Digital Transformation of Last-Mile Delivery

thank you Tim a my own Andy okay it's a pleasure to be back again I've given lots of lectures around here and but not lately so I'm not I don't know if I'm still used to it or not we'll have to give me a chance to warm up practice a little bit right I'm going to talk about my favorite topic last mile delivery I've been working on last mile delivery I caps logistics released its first software in 1980 the first delivery software so routing software so I've been at a long time I guess that means I now know a lot it certainly means I'm persistent I am about to launch a new company called delivery dynamics and so I thought that I was invited to speak that I would talk about why anybody would want to launch a new company in this area given that there are at least a hundred companies in this area that said last mile delivery software so I'm gonna explain a little bit to you about the wine and then I'll try to give you an idea of what the software is like I've been working with a software team or about four years now software is harder to build than it used to be not much software is built in your garage anymore so it takes some effort to do it but we've got it done we're in final testing with most of it now and so we will launch the company sometime after the first year okay I was told that there were gonna be a lot of students here so i thought okay i better start with a more academic picture right here's my here's my academic picture the delivery planning problem from sort of an academic point of view is you got a whole bunch of customers out there and every day there is some subset of that cuz of those customers that need to get deliveries so you've got deliveries to a subset not all of them but a subset and what you have to do is from those you have to take that subset and then you have to group that subject into groups that will fit on a truck and then you have to sequence the deliveries on the truck right so the problem is a very simple one to at least to talk about the results have to satisfy typically a lot of delivery constraints they have to actually fit on the truck they have to get to the customer at whatever time the customer at least when they're open right when they're available to receive it so there's a variety of constraints so it has to satisfy those it needs to be efficient and it should make the customer happy the hardest thing always is making the customer happy it used to be the case that we only worried about the trying to be efficient but now making the customer happy if you're in the last mile delivery business is probably more important than being efficient all right this problem from an academic standpoint is is difficult if you had 200 customers which is a very very small problem by the way 200 customers and 10 trucks then there are about according to my calculation 55 trillion groupings all right and then each one of those groupings you have the problem of developing the sequence which is a very famous problem called the Traveling Salesman problem okay and so there are for each one of those about 2400 trillion possible sequences so I don't know what big means but there there are a bunch of possibilities and so what that means in addition to there being a very very large number of possibilities we don't have any mathematics which is terribly helpful in in getting us to filter through that set of huge number of possibilities to pick out the good one like so it is impossible to generate delivery routes that you can prove our optimal it's just not possible we talked about optimizing delivery routes and but if you look at it from trickly an academic point of view it's not possible not only that it's not possible now but it will never be possible we will never ever be able to do it and you will never ever in your lifetime be able to go through that number of possibilities to look at it it can't be done right all right so when people talk about a route optimizer basically what it means is that they have tried to compare combined computer processing the ability to enumerate possibilities with ideas concepts from optimization and so if you have better concepts for which solutions to look at that helps and if you're able to look at loksabha that helps so what we do is we use parallel processing to look at lots of them now lots of them means typically millions but millions is a teeny little fraction of the trillions of possible solutions right now what the good news is that all of all of these solutions there are typically a lot of them that are pretty good so while it is impossible to prove you have the best one if you are clever about how you do it and you look at a very large number we generally are able to to get a good and in fact we have a route optimizer which looks and acts like any other optimization engine so you would you would be hard-pressed to tell that it wasn't actually getting you the best solution having said that the only way anybody can evaluate how good an optimist if I say I've got one and she says she has one optimization engine the only way you can tell which one is better is you've got to actually look at it you've got to compare what it does in your environment with what mine does and my hers does right now having said all that the thing is in delivery this is not the hard problem in fact you could argue that this is the easier part of delivery all of the stuff which sort of theoretically is not solvable the biggest problem in delivery is that the environment is continually changing delivery environments do not stay the same they don't even stay the same over the same day and there's lots of reasons that change one of the big things the bigger things that's happened is we have an increasing number of urban deliveries people are moving to cities and people in cities want deliveries to themselves but it's also the case that they won't be deliveries to businesses that they frequent and as you get into big cities particularly in developing countries countries like Panama and Peru the retail is largely made up of small stores there's not a lot of Walmart's and targets and the huge big-box stores there's a lot of little media stores in Peru 80% of all retail is done in the little mom-and-pop stores right so the thing about little mom-and-pop stores is that's what you can have in cities they have a small footprint if they have a small footprint that says they get lots of deliveries because there's no room to put a lot of stuff they don't have a big back room to store stuff right so it means that we have in many cases smaller deliveries even when we're delivering to stores traffic patterns are continually changing right so the time it takes me to get from here to my house will not be the same going back today as it was the last time I was down here parking availability is becoming a huge problem in cities you this is over here the FedEx truck it was an article in The New York Times if you hadn't read it it's a pretty nice article about what's happened in Manhattan so that pictures taken in Manhattan and so there's no place to park and the people like FedEx and UPS and anybody else that goes there they park in the street not legal they park in the street and they get tickets and they paid millions of dollars in fines every year for the tickets in fact in some cases they make a deal with the city that says we'll just pay you a fixed amount we don't have to get tickets right which parking the street government governments want to put restrictions on things to make it better for other people other than trucks so they do things like restrict usage of streets in Peru they've just passed a new set of regulations that restrict the main highways so that trucks can't be on them during certain parts of the day right well the problem the deliberate problem was complicated before now it's way more complicated because you can't be on that road you're in certain parts of the day customers the amazons of the world have dramatically changed customer expectations now then a customers expect to get stuff in many cases the same day and many times within a very small time frame so you say okay I want even between 1 and 3 and I want it today all right so customers continue to change and the fact that Emma's on offers it makes a lot of other people who are not buying from Amazon think they should get better service as well there's not much time for planning now people order or do something to to let the provider know they want to buy something and then they won't it soon so you have a very short time to plan it so it's not like it was when I first got in this business where you know you'd have a week or a month to plan now you've got an hour two hours to plan labor is continues to be more expensive and that's the labor to drive the truck the labor to sell the product but also the labor to plan planning labor is the most expensive kind of labor so if you have some of you that want to go and work for a company that it has technology that's going to be used to plan you don't want to work cheap right you also don't like to work at night so you want to charge them more so that's a problem deliveries are smaller a lot of reasons one of the things that's impacted things a lot is that the the stop time to drive time ratio has gotten bigger what that means is that that trucks are stopped more than they're driving and in fact if you look at the to say when I first started to point out what the stem is the stem just then we talked about the stem the stem is the the part of the route that's from the DC to the first customer and from the last customer back to the DC all right so we call it thing to stem now the stem there's no optimization essentially no optimization to what you do for the stem part right so I know I'm going from here from here to there and so I can get on Google and say hey Google tell me the best way to get from here to there right so there's there's not much you can do about the stem part if you look at the the other part the the the whole driving for most of the city routes driving accounts for something in the 20% or less of the day's work all right now if you port it take 40 percent or less and half of that generally is in the stem so you get 20 percent or less that's in the part where sequencing matters so that says that all of this traveling sales and stuff is working on 20% of the problem so if you're able to decrease the sequence by say 10% you've only impacted the solution by 2% because it's only 20% of what you do right so it said okay all this technology that we worked so hard to do that part is not the important part there's a lot of other sort of innovative things that people had to do one is what are called satellite delivery points again I just got back from Peru so I know what there trying to do in Lima but coca-cola in Lima they have a truck and the truck has a great big cab on it and this truck goes down the street and it will stop and there are five guys that are just riding in this cab they all spill out well the truck stopped and they all grabbed their trucks in trucks and go and deliver their product - I have different sets of stores and then they all get back on the truck right well this makes things appreciably harder if you're trying to figure out how you're going to do your routing and that sort of thing so you might ask okay what's wrong with with current delivery management we've spent a huge amount of time on technology to do this why is it that this stuff doesn't work as well as it should well one of the things is that technology it puts too much emphasis on the pieces this is a problem perhaps a fatal problem for operations research generally in that operations our search tends to pick out some part of a great big problem and put a lot of effort on that part in some cases you can do that part quite well but in order to solve the problem you've got to do everything right so one of the problems is if you go out and you buy a routing software generally you've got this piece of routing software and it doesn't hook up to much of anything else you have to you have to make the hookup happen so if you get a technology that is going to provide you visibility provide you some visibility but it usually didn't do much for your routing technology right so you got all these individual pieces and then they don't hang together much planners probably the biggest problem is that planners assume that typically that the data is good enough all right this is good enough Bator chunk it in this fancy optimization scheme I've got and out will come the answers right and I'll show you how do you bore but show you that the data is is typically not good and just think about there's some data that is pretty critical one piece of data that is really critical is where is the customer if I get the customer wrong if I say the customer is over their computer thinks is over there and he's over there I'm not gonna have a good result nothing good is gonna happen with that right if I say that it's going to take me 20 minutes and it's really gonna take me 60 minutes nothing good is gonna happen with that so no matter how good the technology is if the data is not good isn't work and generally the technology does very little to help you with the data you can look at the answers and sometimes as they're sufficiently poor you can say okay something is screwed up in the data but the the technology doesn't try to make it better for you to make it easier for you it still takes a lot of manual effort when when in in the 1980s when we started caps logistics all of the technology was really about going from doing this on a map on the wall to doing it and a map on a computer and the map on the computer is still the main way that routing is done if you look at the technologies the some of the more the most popular ones they still require a huge amount of human interaction to make it work you can't just take the answer and make it work so this is not only expensive but it results in routes that are not very good delivery plans that are not very good manat very systematic everybody does their own thing so you go in and you know this guy thinks he's really slick and she thinks she's really good and everybody does their own thing and you get answers and so tomorrow if you say okay I want to do better you have anything it's systemic to do better one of the things about industrial engineering that we know is sort of the more systematic the most the same the more standard it is the easier it is for us to make it better and we have done very little in delivery planning to make it better we assume that there is no variability or uncertainty it is when you plug this stuff into the model you say it's going to take twenty two point three minutes to get from there to there we know it's not going to take twenty two point three minutes right that's for sure it's not gonna take that but we assume there is no variability and we assume there is no uncertainty and there's really not much in the technology that helps you if you recognize that it's uncertain there's not much you can do to fix it too many times people consider this to be a static environment there's an inclination to come in put effort on installing the technology trying to get it so that it at least looks right and then assuming that it just is going to stay that way and if you think about it a little bit you know that's wrong you know the customers are going to change you know the traffic patterns are going to change you know that what happens that the customer is going to change we've had a big flurry of mobile technology that's come in mostly on cell phones so now we can do a lot of tracking of things on cell phones and this is really pretty technology because you can see I'm gonna big screen you can see for all your trucks are and they're all sort of running around but it hasn't done much to help the planning process it looks pretty it's sold pretty well people feel better knowing where they're all their trucks are but other than feel good there's not been a lot of other things that it's done so far and the last one is that companies even big companies generally don't have the capability or do-it-yourself delivery software it just takes skill sets that they don't have right so we have problems the question is what are you going to do about it and the the thing is that we need to have a an approach that's different and the fact is we've learned a lot in in industrial engineering over the last 40 or 50 years the one thing that we've learned we should have learned is that continual improvement is the way that you should run almost everything and so you can look at it was sort of a core part of the whole quality movement when you look at things like Six Sigma it is a core part of that so this idea of doing something and then keep cycling around is obviously the way to go so then the question is okay can you use this idea in delivery planning well the closest that I could find to somebody who tried to do this there is a an area which is called adaptive management and it's mostly used in trying to manage natural resources forests fisheries those kind of things things which for which there is a lot of uncertainty and it uses based us basically the same thing it's just a little bit different context so I say ok this is not rocket science so I'm sure I can figure out at least what I want to do so here's what I think you should do what you should do and how you do it or two completely different things right so this is what you should do the first thing is to have a data hub and I guess I didn't say start with but everything I'm talking about is or things that are have been enabled by what we can do in the cloud now all right so nice thing about the cloud we can just send everything up we can do all the computing up here and the ether somewhere and we can send all the answers back down and we don't have to worry about right so all the data and everything can be done no question that we can do it and we can have about as much computing as we want so the in in logistics we've always had a problem in getting all the data in the right place and so a data hub is critical to bring everything that you need into the same place so you want to bring it into the same place you want to organize it and then you want to use it for both planning and for keeping up with how well you're doing organizing is a much much bigger deal than you might think and in many cases people don't really think through what they're doing when they organize the data for example if you want to compare if you think okay one of the thing I'm going to want to do is I want to compare whether that truck driver is doing a better job than that truck driver if I want to know that I'm looking at two different drivers that are doing different things every day this one's got one route that it's got another route all of the stuff that they have is somewhat different so if I want to compare them I have to have some way of figuring out which ones of those things they're doing are the same and so if it's the case that for example I have 5,000 7-elevens if I identify all the 7-eleven stores you think okay I should have that but many cases you don't identify all the 7-eleven stores and then when I get through I'll look and see how long it took for 7-eleven stores that had this about this amount of stuff so people that are doing things that are the same so this again not rocket science but it is also not easy it takes more than average smarts to get it organized in the right way so first is to get all the data second the mobile devices the mobile devices while people talk about them mostly in terms of giving you visibility their real value is as a data capture device we talk about the Internet of Things in the thing that is most important is the cell phone a cell phone is a little computer it is the ability to capture GPS so I can know where the vehicle was every second if I want but typically I look maybe every 10 seconds or so when I go in to make a delivery I can indicate on the cell phone what I delivered I can indicate what the problems are so I can capture a lot of data so it's a data capture device that gives me information about what's going on right now but more importantly data that I can use to analyze sort of how well I've done and then there's the ability to apply data science data science is a fancy word that we've done data science pretty much my whole career it all it means is you know you're trying to be smart about the way you do deal with data and there's a lot of things from being able to clean the data to being able to use machine learning to do things right but it is the case that if you're systematic you can do things to clean up today the first thing is to try to make sure that data is right you don't make sure the data is right everything you do is garbage right and it is not easy not easy to figure out if the data is right so use data science to help you clean up the data and to help you to to get better planning estimates again the cloud is a big plus for this because we can store huge amounts of data we used to only can have a little bit of data about customers now we can have lots of data about customers question is how we're going to use it and then there's the the issue of identifying the problems and inefficiencies so we can do that both automatically on a computer in theory in by various kinds of data visualization alright so we can figure out at least where the problems and efficiencies are now it may be that the problem is it is taking too long at each one of the stores so that's not a computer problem that's a people problem but if it's taken too long at each one of the stores then you say ok there's where the problem is so what I have to do is I have to go and look and see if I have the right processes if I've standardized the processes if people are doing this efficiently so all of those kind of things so everything is not technology but it's still something you have to do then once you've done that what you want to do is you want to adapt the framework and parameters it doesn't make any sense if you're running a thousand trucks a day to have people looking at a thousand routes a day trying to make them better that is nonsensical what should happen is that your people ought to be looking at how you're generating the thousand route today and they I want to be working to make the thousand routes be good if they're automatically generated right so you should work on the process not on the final product if you work on the process then you've got something that is repeatable that you can do something about all right so that is maybe the hardest thing in all of this of all these boxes that may be the hardest one is okay I'm now yeah you see a lot of the newer technology now will tell you here's your plan and here's how much you miss the planet well that's all fine but it's pretty much useless unless somehow from that you can figure out how to plan better tomorrow the fact that I was off by 40 percent today okay makes me feel bad but I need to have some way of saying okay I got to do better tomorrow right so the adapting is what are you going to do to make it better tomorrow and then to be able to automatically optimize and buy again optimize I mean try to get a good one delivery plans alright so and then everything sort of starts over they just keep at it so you start and then whatever changes you make you try to make little changes not big changes little changes try to see what the impact is and you try to continually improve that's the fundamental notion out of that you can hook on business intelligence to the data hub and you can hook on a variety of things that makes you look better to your customer you can know what your expected time of arrival is at the customer you can send a messages you can do whatever does allow them to track the trucks all of this is technology that's that's doable and then on that side in looking at how you identify the problems you've got two big buckets of stuff one is analytics and that's the buckets you like the best because most of the analytics the computer's gonna do for you going Peter is gonna go in there and try to figure out what the problem is visualization you got to figure out what the problem is now some cases visualization is the only thing you got I you can't get smart enough to make the computer do it you're always working toward making that computer take a bigger part of that so you always want that one to be bigger and then bottle indistinct now all right with business intelligence you have to be careful because you got a boatload of data so yeah huge amounts of data so it is easy to draw all the pretty bar charts and circle charts and whatever charts you could have and they tell you nothing they are not helpful at all in fact in many cases they're unhelpful they're the opposite of helpful right so business intelligence ought to be simple you have to remember that in the delivery world you got smart people but mostly you don't have rocket scientists right so if you're gonna give the guy who is managing you know 50 delivery drivers something business intelligence to work with you don't want them out there having to write macros that's not what you want right you want something that's simple it's configured that looks the same sort of across the board all right once I had my picture drone I started trying to figure out how I was going to actually build the technology to do that the data hub is relatively easy with today's technologies you can connect things up they're relatively easy to connect to connect to people so that's that's not so bad we've been working a long time on the the route optimizer and we have one that is very good G coding Google has been a big help with geocoding you input an address and if you live in the United States you get back a latitude and longitude which most of the time is in the ballpark now it is not in many cases where you actually want to go it the the best year coding you can get is what they call rooftop geocoding and a rooftop is just what you think it is it's the about where they think the building is but but if you're going to deliver you're not going to deliver to the rooftop you're gonna deliver to the dock door or you're going to deliver somewhere so you need to have something which is better than that it's also the case that there is a small percentage maybe a percent of the G codes are just garbage you don't notice so much when you have your you're doing it on your cell phone but I spent five years working with the post office and I've looked at thousands and thousands of postal routes and you you will have some that will geo code in the middle of the streets some that were geo code there's the bite of a forgotten where the zero zero is there's a zero zero the latitudinal on diffusion it takes you this one place in the world there's a lot of the data is not not perfect so you there's a way to get started but then you have to do something to improve it if you're in a lot of countries like Panama Panama you you will get about as close as to say you're in Fulton County County not real helpful right you're gonna deliver the Fulton County that's tough the we've gotten over the years we've gotten better and better at both analytics and business intelligence easier to do business intelligence and we have some real capability in in doing in leaks the the hard part there is getting it trying to figure out exactly what analytics you want to do and what business intelligence dispatching is always exciting because you don't know exactly what people want to do some people want to have different strategies for which truck drivers they want to do things with but there's a part to do that there's the part to do monitoring and the exception handling that is looked where things are all the time communicate with the drivers and then their mobile assistance and we have a mobile assistant we actually have three of them one for drivers one for sales people and one for merchandisers so the people who come along behind and put stuff up so we built all of the pieces right and so the technology is not I would say other than the the route optimizer is not really rocket science Raoh optimizer more or less is rocket science but the rest of it is mostly not but it is not at all easy to build so that it actually works one of the hardest parts is are these mobile assistants you've got an app to a mobile phone and mobile phones got an operating system and it doesn't have the same version of the operating system even if you're got the same brand a mobile phone right so they changed the operating system and sometimes they'll do it to you when they change the operating system all right in particularly if people try to install the apps on old phones then your phones most of the newer phones we can make work be pretty sure it's going to work before ever having tested it but the older ones maybe not so much and there are a lot of older phones still out there okay all right let's go back to the problem again just for a minute there are a lot of different flavors of delivery problems and so when you walk in and somebody that says I have a delivery problem and you got technology that doesn't necessarily mean your technology is going to solve their problem so you have to sort out what it is that is important about their problem and the things that matter these are the ones I think matter the most probably the the way they do sales is the thing that makes the most difference water is called traditional sales is you will send out a sales representative they'll go to a store they will take an order and then either tomorrow or two days from now they will make the delivery so you'll have a set time when you're gonna have to make the delivery all right so that's one kind of delivery and they're the you want to have almost always you want the deliveries to be on a thick sequence that is if they're once a week you don't want them to be every Monday or every Tuesday or every Wednesday mostly your customers don't like one week you come this day and another week you come that day because you have to think about the customer custom you got to figure out how much to buy right and they want to buy week earth of stuff and if they don't know if you're going to come back in a week or ten days I don't know how much stuff to buy so if they don't want to run out that to buy ten days so customers ecommerce has created a humongous headache for everybody in the delivery business because if you're doing the traditional sales I say I'm gonna go to those people on Monday and those people on Tuesday and those people on Wednesday and if I'm really tough I won't let them be delivered except on those days now if they're an important enough customer they have a lot of what are called off cycle orders but mostly what I want to do is I want to balance out the work that's gonna take place every day ecommerce generally you don't have any idea when people are going to buy so everybody is going to buy own Monday what's worse your sales staff screws up the buying because they'll run black friday during the delivery business you don't like Black Friday you don't like anything that causes this big spike because if you order it today they're going to expect it to be delivered at some time in the future and for all of delivery this is what you want this is not what you want right because if you're going to satisfy the customer and you got this then you got to haven't have enough capability to be up here and that means when you're down here you're still gonna pay for that capability you've got up here okay the vendor managed is the easiest because that says I can send it whenever I want to right and then delivery times delivery anytime I like that right delivery during delivery windows is the most typical them and and the most common delivery one is you have or when the stores open so nine to five but that's much better delivering to a store than delivering to a individual if you're delivering to an individual they're not open 9 to 5 all right so what has to happen is you somehow have to have a scheduled delivery if you're gonna actually deliver it to the person so scheduled deliveries first it's harder to solve all of this delivery problem if you have to have scheduled because that says I've got a pretty narrow window but it's also the case I got to work out with the customer when what that one there's gonna be right somehow I've got to say okay I'm gonna get you to your house between two and three and you've got to say yep I'm gonna be there between two and three so this one is the hardest when you have scheduled deliveries in much of the u.s. you if if I'm not home UPS leaves the stuff at the door if it doesn't say Apple on the front and it's not booze and they leave it on little right if you're in Panama nothing is left on the door because if you leave it on the door it simply won't be there right so it just doesn't happen so in many of the developing countries and also many of the big cities us big cities you have to leave it outside they just won't do it access is a big deal if there are restrictions on the access then you know restrictions can be like for right now Amazon everybody talked about Oh Amazon's in the delivery business but Amazon has a very hard time delivering to be because I happen to live in a gated community you know if you want to FedEx can deliver to me because we give FedEx the code of the game you'd be asking to deliver to me you the post office can deliver to me Amazon Kate delivered to me my not you're not Amazon drivers right they're crowdsourcing the drivers so it can be a different driver every time so giving Amazon the code it's just like I might have what posted on the front gate right so any kind of restrictions cause issues parking is more and more of a problem people get terribly aggravated about trucks parked on the street in Panama the trucks generally are the the company allow the drivers to decide what the route is going to be now you might think that's a good idea because drivers should know what their route is going to be one of the problems with that is the guy who's loading the truck doesn't know what the routes gonna be so they got a loading truck loads the truck but he doesn't have it loaded in delivery order so if it's not loaded in delivery order it takes a guy a lot longer when he arrives there because he's got to find the order you got a scramble in the truck where you got you know 150 degrees and dark to do it walk in requirements more and more particularly big cities you have to park and do some walking and the more you have to walk then the more time it's gonna take you to do it now all of those things cause your estimates to be different if I have to walk if you if the post office if they deliver a package to me they have to walk about ten feet I live in Vinings the streets their house is there this is easy right but they'll also deliver to somebody that lives 100 yards off the road and that takes a lot longer so it's very hard to figure out how long it should take them to do it the last thing here is one of the one of the things that is a huge deal and we'll talk a little bit more about it is driver familiarity if you try to estimate how long it's going to take to find a customer this is pretty easy to do if the same driver is going there every time and knows what to do the same with servicing a customer they going back to the same place every time they know what to do if you send a different driver then they don't know what to do so first it's expensive in terms of it usually takes longer and it also you don't know what kind of number to put in there when you're doing the planning right with planning you'd like to have lots of customers on each route but not so many that you can't get them delivered and you have to bring the stuff back so if you have to bring something back that general is way more expensive than having left it off altogether all right uh there's design issues I guess I better speed up a little the design issues are something that what we have technology that helps with this it generally the company is wise to have some hotshot consulting somebody help them to use the technology to do this so are you hotshot consulting people to listen up and the the first thing is with regard to delivery day assignments truck drivers are typically on salary and so if their own salary that says that that every day they're going to get paid for eight hours if you don't send them out they still get paid for eight hours right so if you are very efficient one day and you only need 20 trucks but the next day you need 40 trucks and you have 40 trucks and drivers on salary you're going to pay the 40 truck pay for the 40 trucks and drivers on the day you only need 20 all right so what that means is that you can't wait you shouldn't wait and and just let your customers decide when they're going to get their delivery what you want to do is decide when they're going to get their delivery because if they do that you can spread it out so if there's one thing in any almost any delivery environment where you have the capability to do that the capability to spread it out you can save more money by spreading it out than you ever could by being more efficient with the routes all right having said that this is not an easy thing to do first it's not an easy thing to do because particularly for the ones that are what we call traditional sales we've got sales reps because another one do you have to spread it out over the drivers you also have to have the sales reps have about the same amount of work every day so the technology is not so easy to do that with route strategy there's always this debate between dynamic routes and what some people call fixed routes but what I prefer to talk about is master routes dynamic route means it's a route that I have all the data and I now build a route for that data a master route means I am going to approximate what the routes are going to be live based on past data so I've got an approximation now the fact that you've got to master route doesn't mean you actually have to run that route it means you start with that route all right so it's today and now I either am going to start from scratch that is I'm just gonna pretend I know nothing and build a new route set of routes or I'm gonna start with a set of master routes my approximate route so I try to improve them if it were the case that I could truly optimize if what I said right to start was a lie right and I really could optimize it wouldn't matter which one of those things you do but generally people have found it easier to start from scratch to get a good solution than to improve the solution all right so we've had to work hard on the starting with a master outside but that clearly is the better way to do it almost always you're better off if you can take advantage of what you know so you have to have some way of taking advantage of what you've learned if you're doing dynamic routes every day you you have a hard time learning and then you have a hard time taking advantage of what you've learned all right so one of the the big strategy issues you have is are you going to do that the biggest difficulty with master routes is that you have to keep them up because having bad master routes is worse than having no routes at all because that's where you're going to start your optimization so if you're not going to pay attention have somebody take care of the master routes you're better off not to have master route because if you start with our headmaster route the computer is sitting there needs trying to make that thing better right but the worse it is the more effort it takes to make it better if you let the computer decide itself what the routes going to be it can generate a pretty good route so it can get a pretty good starting place now if you work hard on the master routes you can get a better starting place thank you returns and revisits are a huge issue with regard to how you're going to deal with them returns are expensive if you try to deliver something and you don't it's expensive because I have made the effort I've accomplished nothing by it I got to bring it back it has to be processed back at my distribution center put back in stock whatever it's a very expensive operation so you don't want that to happen but the question is how are you going to deal with this so somebody has to have a strategy for how you're going to deal with it in some places they will make try to make multiple go multiple times to make the same delivery just because they don't want to bring it back okay master routes we got pretty good master routes one of the things that's nice thing I give you my take on on optimization optimization the algorithms in optimization are mostly pretty limited mostly you cannot solve the problems that I want to solve you cannot solve the delivery problem with I sort of off-the-shelf routing algorithm having said that the concepts you can use to solve the problem all right so we can go in and do a pretty good job of generating master routes that are reasonably well balanced now here each one of these is a different day so you break them up my day and here I've got geographies that so Monday it's down here at Tuesday's up there Wednesday I did it wrong first words but this is one day down here now if you're going to let this this works fine if you're going to make people adhere to the the day you've assigned them the delivery but if you're going to let that guy up at the top order on this day down here it means you're not going to have if I do this I'm not gonna have a truck up there so this is not necessarily the best strategy somehow you have to decide am I going to enforce my delivery days if I am it's a good strategy most efficient I'm not not a good strategy I'm better off to have trucks going all over the area at the same time just I don't have time to talk about it but we saw a relaxation of this problem and then take the solution and convert it into what you do every day and we can do some things to match up what we're working on a problem right now where there's a distributor and they distribute Pepsi products not peppery beverage but Pepsi other products Gatorade that sort of thing and they distribute coolers and Pepsi and coolers will let them deliver on the same truck but then won't let them order by the same flavor threats so they send her sales reps they send Pepsi sales reps and they both have to go out there on say Monday and they'll both be delivered on Wednesday and so now you have to have your delivery routes have to make sense so you have to assigned each one of those customers a delivery day so that you're going to have efficient routes but then your Pepsi sales reps have to have a pretty nice group of customers to visit and your quarter sales reps have to have a pronounced customers to visit so if you want to look for a hard problem for your masters PhD whatever that comes in that category hard problem what they district wired we got it the customer data you got G codes that is where are they you got to stop times how long are they going to be stopped you got the delivery windows when can you deliver to him you got the delivery restrictions so this guy's truck can't go in there right the drive data you need drive times and drive distances so we call them a drive time matrix and a drive distance matrix and that's between every pair this you might want to go between and it's not just us everybody in the world who does this has to have that data sometimes they approximate it by straight-line great circle miles and if they do it almost always will give you garbage okay there's order data's this data is usually pretty good there's Product data that is pretty good if it's weight I don't know how volume got put over to the side it should be this much bigger volume is much harder if they deliver in things like cases that's okay you can count those if they're delivering a lot of other stuff it is hard to know how much volume this this order he's gonna take up on the truck there's truck data driver data availability that's usually pretty good and then stop beta parking again is is always a problem so here is the kind of baby you have all right with regard to stop time data traditionally what we've done is we've created a equation that says the stop time is going to be equal to something's time plus the amount of stuff you're delivering times some constant that's what we've done and you apply it across the board and for sure it is going to be really bad for some circumstances because often it's the case particularly for delivering to little people that the fixed cost is a much bigger deal than the variable cost all right you have to wait because grandmother's got to take somebody out or whatever so what we can do now that is is much better is we keep up with a distribution of stop times for every customer so every customers got a distribution right so we know what they what they had delivered we know how long it took we know what day it was when what time of day it was so we know all that stuff and so because you got that you can make a much much better estimate and every time you go out there you get another piece of data and so you are continually improving and if it changes if something is different now you know that it's different so this is one of those things that the computer can keep up with right the the drive time data is probably the hardest you can get drive times and distance matrices and you can get them from Google Bing open streets all of those you can get a prompt the there's a variety of issues one is you can have bad G codes and if you have bad G codes what has to happen is you got a G code and then wherever you get it somebody has to match the G code to this underlying street network all right it's got it's going to take it and usually it's going to match that there's an underlying network it's going to take the G code and it's going to try to put it to the closest place all right now if this is a rooftop geo code and the closest place happens to be on the other side of the interstate behind the house equal put it there right so it pretty easy to get it screwed up and it takes pretty good technology to figure out that you've got it screwed up all right so you can have a bad Network assignment bad G codes drop time is variable and these companies typically don't tell you when they're when they're giving you the time for they don't say we this your estimate is 20 minutes to get from here to here and that will be at 9 o'clock in the morning all right so what has to happen is that you have to if you want to do it well you have to revise it based on the GPS trail now what we can do got the GPS is on the phone we're getting a GPS signal every 10 seconds so we knew where the vehicle is so we know how long it took to go from A to B for any a and B you want to have here I've color-coded them so the yellow ones are 42 200 240 kilometres per hour so I can figure out what the time is and then again I just do the same thing it's more complicated but the same thing I was doing with the stop times so I've got samples if you use it master routes this is much better because a master out will tend to take you along the route about the same time every day if it's about the same thing time every day you're going to be much better at estimating mostly what you want to do is you would like to take out the variability and uncertainty you'd like to make it go away so the more you can do that almost always the better off you are the idea that you can take advantage of sort of knowledge about the order and that's going to do better almost always that's wrong route optimisation we have a very good at optimizer we can do dynamic routes if that's what somebody wants to do and in some cases that is the thing to do or we can generate master routes and then improve the master routes right biggest problem you have the route optimizer is these things will work pretty well as long as you don't get too many customers on a route it's making lots and lots of looking at lots and lots of different solutions and if it's looking at a solution that has lots of customers on the route then it takes it a while even if it's using a heuristic which it is to solve the travelling servant problem and if you're it doesn't matter if you're only solving one I mean you know it takes a second but if I want to solve a million right now a second is a long time so I've got to do better than that so there are still some issues with that once we have the master route or we've got the the the optimization engine set up then it should not be the case then anybody has to do anything to it now this is a concept that John and I tried quite a few years ago we weren't smart enough then to make it work but now we can actually make it work the thing is that what you've got to do is if it if you did something wrong today then you've got to figure out what you did wrong today and then you've got to make a correction in the in the way the algorithms work rather than try to correct the route so you just have to say ok that route was messed up it's probably the case that nobody is smart enough to figure out exactly that it was miss before it left anyway the mobile platforms have gotten better one of the problems is they're too easy to build so a lot of people have built them and now somebody's got one and they want to use their platform the problem is it's it's easier and cheaper for me to provide the platform that it is for me to try to hook up with whatever platform they've got to get the data all right but what they do is they can capture the GPS trails they can capture reasons for turns what you'd like to do is not have returns if you want to not have returns you need to figure out what's causing the returns all right they can capture reasons for no sales in a lot of cases as salespeople go out they have a group of customers they're supposed to visit but they may not visit those customers especially if they're running out of time their tendency is to go to the ones that they can sell the Moose to so the guy they can't sell very much to may not get a call and if they don't get a call ultimately you lose that guy they can provide real-time visibility and you can provide all kinds of alerts to drivers sales people's customers that sort of thing all of that's easy we can do that and just to give you an idea these are just things off phone basically what phone does is you download the order in delivery order so there they are now you pick one you say okay I'm gonna pick the first one that's when I pick over there anyway you pick one and now then you say okay there's some choices maybe I want driving directions if you say I want driving directions what it does is just go to whatever app you have on your phone so if you use Google or ways or whatever it just takes you there puts in the latitude and longitude so you have to do that and then you get driving directions right if you know how to get there you go ahead and get there if you get there and it's closed then you skip it or whatever reason if you say you're going to skip it then you have to say why you skipped it then because all that information is kept when you get there you indicate that you've you've arrived many many many companies capture the time taken for the delivery from putting the when the driver says I arrived and when the ever since I deported if you did that you almost always have bad data cause drivers forget they don't do it at the right place sometimes I don't even do it at the store they're supposed to be all right so what this hopefully does is gives you an indication of where they are the GPS trail you know where they are geographically but you don't know which store they're at right so then when you get there you can either put in more information or if you've if there's some reason that there's some problem you can specify what it is we've tried doing things like letting them put notes in that turns out to be a bad idea for two reasons one is that they usually put a lot of things in particular when they're unhappy and so it takes somebody a lot of work to figure out what they've said so it works usually not better if you have defined pretty carefully what is you want to do this is this is why they didn't know the uke on the phone you can you indicate what was delivered and what wasn't delivered he can you can calculate what the what the cost is it totals it up it keeps up with it if it didn't if there's anything it didn't take in the order you can specify a blank wasn't taken so somebody can figure out what the problem was particularly if it's something like an order picking issue or something like that you know what that one is oh this is the same thing for sales so the salespeople typically they'll take one and so if they go to some place and they'll get a sales order then that indicate plan all right visibility we can anytime you can look and you can see where everybody is you can see how much they've got left on there their route so this was about 56% done I can look and see where they are I can look and see how much time they've spent at every place so forth so I can see where they are it's not ain't useful but pretty right that data is much more useful when it's accumulated and then you do some analytics on it later occasionally they run into a problem that somebody can help them with what we found is that not many people want to look at this all the time so we define what it is that they believe to be a problem events not too long ago in place for their more than 10 minutes late and so the manager gets sent an alert so it goes onto another app every time you go out when they make the delivery if they indicate it on the phone we try to match it up with where they actually made the stop so you've got a GPS indication that you made a stop and hopefully you've got at least one point on the phone in that time frame when they hit the button if they did that now I can match them up and say ok they were at this location I know it's this location and so now I have one more sample of the latter human latitude glad if you longitude that there is error associated with latitude and longitude so the more samples you get the better your estimate of what it is the circle you see here is how good we believe it is so every time we get another sample we squeeze down the circle all right and then when he gets good enough we say aye that's where it was instead of where we thought it was we thought it was over here but that's where it really is we knew something very similar with the stop times of drive types analytics like I said the problem year now is mostly you get too much rather than too little the same analytics are used to make business decisions and provide business insights but you also use this to try to figure out how you're going to do better the next time and we've got a lot smarter about what we do so the key features sort of the this new generation of stuff you can optimize the sales territory is the delivery days you can balance workloads you can look at alternate delivery strategies you can automatically optimize the routes you can use either master routes or dynamic routes you can monitor the status of the drivers you can provide business intelligence about all the stuff you can send alerts to anybody that you want to you can clean the data so there's a lot of tools associated which is making the data work and so everything is sort of in one spot and it all connects together all right maybe that's the biggest deal is it all connects together so all it has to do is you have to have airs api's that connect used to connect to the company so you've got to get data from the company every day about what is going to what be delivered that day and then at the end of the day they have to get the data back about what actually was delivered that day so they can reconcile and then everything in the middle is handle sort of away from the company so that's the only connection it has to have there's other ways to do it but that's the cleanest way now you can ask a question ago yes we use we use the big iron to to do the enumeration and we use it for some of the data GPS data you end up with a tremendous amount of data because you're getting a lot of human longitude every 10 seconds from every truck right and so you get a huge amount of data the the thing is you don't have to have you can but you don't have to have so much power because you have time to do the processing so I can process that overnight it doesn't you know the only thing that that I've got a process quickly is to generate the routes I have to process that quickly and so there we you do need big iron and you need a lot of smarts in the algorithm other than that even for the the GPS on the phones that it takes very little data to do it is only to stay within the maps that maps will chew up a lot of data but just just sending them the GPS coordinate is insignificant it's not doesn't doesn't take much volume of data and then you're not doing anything with it except just looking at where they are at the time you're capturing but you're looking at where they are so certainly if there are as you get more and more data stored then it's more and more of a problem there's a big problem with business intelligence because the customer can reach and get you know a year's worth of data well you have a bunch of data so we restrict how much and get just because it gets all choked down once you get lots of data just give it to you yeah we we haven't tried real hard I mean it obviously is is difficult to quantify one of the biggest problems is you know most delivery systems are currently not managed so if you ask the question of how much is it worth to me to have a managed system versus an unmanaged system it's a it's a not an easy thing to do we do quantify things like what is the direct delivery cost so we keep direct delivery costs for every customer every customer we do that now but we don't keep up that that's just what it costs for that truck to go out and deliver to that customer in addition to that there's if any trucks were at all that's not counted in their and returns it's a very hard to put any kind of number on returns if you go out and look at any return operation you can see that it is atrociously expensive and there's a lot of loss in the returns a lot of the returns don't get returned as part of the problem but no it is it is very very difficult to quantify it's also in some cases hard to get people to admit that their current system doesn't work very well as you know there I believe there are very very few delivery systems where the technology and concepts are working very well and so I think it's going to take a different generation of technology but it's also going to take people who know what to do with the technology and like I said most companies I don't think do yes would would do with what yeah that is that is the one thing that monitoring can do because the guys back at the distribution center wherever the control tower is and control tower can be just one room or it can be thinking the size of this but they know that the bridge is al are the spending big acts are done or whatever so they can notify all of their the the drivers that that is the case and they can try to either give up advice or tell them to look on ways in Google or Google to try to get advice there there's still a problem with ladies and Google in that they are geared for cars and so you have some difficulties if cars and trucks are not allowed on the same same street so you have bigger difficulties so there's a bridge in Panama that the trucks can't go across during the daytime so they have to get over there real early in the morning or and get back later at night so there's some some issues with that but I think again there are a few dynamics like that sort of things that instantaneously come up that you have to deal with but if you wanted to look at how you wanted to run a better delivery operation it wouldn't be I think usually to handle those things better the the thing in delivery is you are just sending out a bunch of stuff you make in a bunch of deliveries so if you can save you know 50 cents per delivery you can save a huge amount of money and the delivery is is that one point where most companies maybe the only point where they actually interface with the customer and if they screw it up the customer is going to be very unhappy about it unhappy enough to lose the customer a lot of times so I think that the I would advise anybody you know that the trend around the delivery system to try to make all of the the sort of bread-and-butter ones work make make all of those where you know try to eliminate as much variability and uncertainty as you can even if you you're going to pay a little extra for it you're going to have to not be quite as efficient to try not to introduce yourself variability and uncertainty which is going to cause a problem with the customer and it causes a problem or your planners so mad enough when it happens and you can't do anything about it but you shouldn't do it to yourself usually the thing the problem you have with cheat codes is even the look your delivery location doesn't necessarily stay the same right I mean it at one time we were happy if we could just get within a buck or two but now you know if you're if you're trying to make 50 or 60 deliveries a day you just can't waste time looking for it so if if it's got to be at the back door you got to go to the back door and if they change it and now it's going to be at the front door you got to go to the front door right so you have to keep looking at your samples and see if they jive with where you you think the thing was and you know as soon as you you you have to say okay I'm getting these new samples now are they consistently at a different location if they are you got you've either got a caller people and say have you changed what we're doing you got talked to your driver chef do something so the the biggest difficulty is trying to figure out if there's change yep you don't have to have a chief at people talk about geofences and and they're pretty I mean geofence all that means is you're going here's the point now I'm gonna go off fence around the point and then I'm gonna figure out they go across the fence but the simplest thing to do is just say how far I am from the point you know no here's creet right I don't have to have a fence if I'm if I do the geofence somebody has to fence it somebody and you have to think that a moderate-sized we work with a moderate-sized company in Panama and they have 13,000 customers that they visit 13,000 geofences take some time to geofence 13,000 right so anything anything where on the on the planning side where you're introducing something that people have to do you try to stay away from that you don't want to do that automation we would like to have the whole process automated and I mean basically when you people talk about digital transformation you know that you could say automation is what we're trying to do we're trying to automate stuff and the transformation for a lot of what we do is translating something that somebody's head on to something the computer can understand post office I work with post office for the last five years that post office doesn't know where any of the the locations of mailboxes are the computer doesn't know your your house they don't know where your mailbox is right I don't know where your door yes so the the question of how you can get that from the the delivery guys head into the computer is not a trivial matter but that's one of the things we digitize yes yes drivers it is absolutely the case the drivers do not want to do this drivers have spent a long time doing whatever they wanted to we have one client and I was looking we put it first it depends on where you are if you're in the US this is negotiated if you're in Panama there's no negotiation about this this is they say you're going to drive it or you know you're not going to work for me tomorrow that's all the wait works but we we put the the the mobile phone app out and then I'm trying to look at the result and I noticed that all the trucks go to the same area that they try to beat traffic panama city traffic panama city panama traffic is very bad and they they get the trucks out early to beat the traffic and so i looked and all the trucks are going to the same location first so they go this location they stay here an hour and now there's traffic everywhere and now they go out somewhere else and it's the best breakfast spot in panama right so drivers have been used to that sort of thing when when when the that's one thing that all of the monitoring does is it does stop a lot of that to start with we were having about from an hour to two hours of what we call unidentified stop time that means we know the jeep from the gps it stopped we just don't know what it stopped for right because we don't have anything when you Mitch you can match it to and now the average unidentified stop for the same company same drivers is thirty minutes and thirty minutes is what they get in for lunch they drop any of them want to it for thirty minutes so they don't like it and I'm not that's not my business it's the company's gonna have to figure out how to deal with that we one of the things we try to do is tell the monitors to not get too overbearing sometimes they'll you know the driver will have been at a stop for 20 minutes and they want to call them to find out why they're stopped for 20 minutes and but there may be very good reasons one of the drivers stopped there for 20 minutes particular if it's a customer you know they have to wait to do whatever but no they're not crazy about it salespeople like it even less the sales before everybody you tried to provide something that is helpful to them one of the things we found that the drivers didn't like at all was that when they got back in the evening they had to reconcile they had to bring the stuff back they in Panama they they pay cash so almost everything is a cash deal after bringing the money back and after bring back what they didn't deliver and then they have to reconcile and the problem was that they came back at odd times and sometimes there wouldn't be anybody there for them to reconcile with so they just had to sit there and wait until they reconciled now the boss man tells the people doing the reconciliation you can see where they are so when they get to the door there better be somebody there to reconcile a hide to wait for the salespeople and set of people say what features on the sales guy you know he sales people are very cranky about being observed but they they want one of the things that we provide to them is on the mobile app they have when they go to our customer they can see if the customer rejected something from the last order and a lot of times the customers reason for rejection they have to say why they rejected it feda if it turned it out and a lot of times I say they didn't order it so now the salesperson is sitting there and they say uh last time they rejected this and they said didn't order it but you know I know they ordered it so I got to work it out with the salespeople so they liked that part they also liked the part that those people are often assigned far more customers than they can go visit they don't go visit them because they can't you know they give you 60 stores to go take orders in today you take orders in 60 stores and so by having the technology you can see what they did you can see it is nobody had any idea how long it took a salesperson though somehow it has to be some advantage for them are they complaining a lot and then it's up to management to the or we what all we try to do is to make it as attractive as we can make [Applause]

Channel: GTSCL

Views: 5,827

Rating: 4.8846154 out of 5

Keywords: supply chain, logistics, Georgia Tech, last-mile delivery, digital transformation, business intelligence, artificial intelligence, predictive analytics

Id: gRrb4gQZ3UI

Length: 85min 52sec (5152 seconds)

Published: Tue Dec 03 2019