AI Route Planning for Thai Last-Mile Delivery
How AI route optimization works for Thai last-mile fleets, what data it needs, how to pilot it safely and where it falls short.
AI route optimization in Thailand means using solvers and learning models to decide which driver visits which drops, in what order, and to re-plan when traffic, orders or vehicle availability change. For last-mile delivery, the gain comes from fewer wasted kilometers, fewer missed time windows and dispatchers who spend less time on spreadsheets. Dispatchers keep the final say.
This article is for operations and logistics leads at Thai delivery companies, retailers with their own fleets, and 3PLs. It covers what route planning software does, the data it needs, why Thai cities make it harder, and how to pilot it without disrupting live deliveries.
What AI route optimization actually does
A route planner solves a puzzle: given a set of drops, vehicles, time windows and constraints, find a plan that serves every drop at the least distance or time. This is a long-studied problem in operations research, usually called vehicle routing.
AI adds three things on top of the solver:
- Better inputs. Models predict how long a stop will take and how fast a road really is at a given hour, rather than using a fixed guess.
- Live re-planning. When an order is added, a customer is not home or traffic changes, the plan updates and the driver app shows the new sequence.
- Learning from outcomes. Actual delivery times feed back so tomorrow's plan is more realistic than today's.
Our logistics practice describes this as route and load optimization connected to the driver app, with dispatcher override. See the full logistics and supply chain page for the wider scope.
Why Thai last-mile is a hard version of the problem
Textbook examples assume clean grids and reliable addresses. Thai delivery is messier, and your planner has to cope with that.
| Challenge | What it does to a plan |
|---|---|
| Traffic that shifts by hour and weather | A route that is fast at 10:00 can be slow at 17:00 |
| Narrow soi and one-way streets | Map distance and driveable distance differ |
| Addresses written in Thai, English or both | Geocoding errors send drivers to the wrong spot |
| Condominiums and gated estates | Parking, security checks and lift waits add minutes per drop |
| Motorcycle and van mixes | Different vehicles reach different places at different speeds |
| Rainy-season flooding | Roads close without notice |
| Cash on delivery and failed attempts | Stops take longer or need a second visit |
None of these makes optimization impossible. They make it important to use real local data and to test on your own routes before trusting the output.
What data you need
You do not need a perfect data lake. You need a few sources that are reliable enough.
- Orders and drops. Address, coordinates, parcel size or weight, requested time window and any special handling.
- Fleet details. Vehicle types, capacity, driver shift times, start and end locations.
- Historic delivery records. Timestamps for arrival, completion and failed attempts, ideally from the driver app.
- Travel time data. Observed speeds by road and hour from your own GPS traces or a map provider.
- Rules. Restricted zones, loading windows, driver break requirements and customer commitments.
The most common blocker is the address. If a third of your coordinates are guesses, a perfect solver will still send drivers to the wrong place. Cleaning up address data, and letting drivers correct a pin once and keep it, often delivers more than a better algorithm.
How a good pilot is set up
Do not switch the whole fleet at once. A pilot that runs beside your current process gives you a fair comparison.
Pick a zone and a time period. Choose a district with enough daily volume to matter, such as one depot area. Run for several weeks so weather and weekday patterns show up.
Plan in shadow first. Let the system produce plans each morning but have dispatchers keep their own. Compare distance, stops per route and predicted finish time. This protects service quality and shows where the system disagrees with your experts.
Then run live on a few routes. Give drivers the new sequence on some routes and keep others as control. Use the same days and similar drop density.
Hold a weekly review. Ask drivers and dispatchers where the plan did not make sense. Their answers become new rules, such as "this building needs ten minutes for security."
Measure the right things
Choose a handful of measures before you start and record the baseline.
- Distance driven per drop
- Drops completed per route per shift
- On-time rate against the promised window
- First-attempt delivery rate
- Dispatcher time spent planning each day
- Share of plans changed by hand, and why
The last measure is underrated. If dispatchers override half the plans, you have learned something about missing rules or poor data, and that is as useful as a savings figure.
The dispatcher stays in control
Optimization is not about removing dispatchers. It is about giving them a good first draft and clear reasons.
A usable tool lets a dispatcher:
- See why a stop was placed where it was, for example a time window or capacity limit.
- Drag a stop to a different route and watch the plan recalculate.
- Lock certain stops, such as a VIP customer or a fragile delivery.
- Record why they overrode a plan, so the team can improve the rules.
Explainable changes matter for trust. A driver who is told to skip a stop and come back later will accept that if the app shows the reason. They will ignore a black box.
What can go wrong
Plans that look good on paper but fail on the road. If service times are underestimated, every route runs late and drivers lose faith. Calibrate service times from real data, by building type if possible.
Optimizing one number. Minimizing distance can create routes that break driver shift rules or leave a long drive back to the depot. Include shift length, breaks and fairness between drivers as explicit constraints.
Re-planning too often. If the plan changes every ten minutes, drivers cannot follow it. Set rules for when a change is worth pushing, such as a major delay or a high-priority order.
Poor connectivity. Drivers may lose signal in basements and rural stretches. Test that the app keeps working offline and syncs later.
Privacy and tracking. Driver location and customer addresses are personal data. Under Thailand's PDPA you need a clear purpose, limited retention and controlled access. Tell drivers what is tracked and why, check current guidance from the PDPC, and confirm your approach with counsel.
Unrealistic expectations. Optimization removes avoidable waste, but it cannot create road space or fix a late warehouse dispatch. If vans leave the depot two hours behind schedule, the route plan will not rescue the day.
When not to start with route optimization
Route optimization is a weak first project if:
- You make only a few dozen drops per day. A good dispatcher with a map will do about as well.
- Most deliveries are single point-to-point trips with no sequencing choices.
- Your order data arrives too late to plan, for example orders confirmed after vans have left.
- You have no digital record of deliveries, so you have no baseline and nothing to learn from.
In these cases, begin with capturing delivery timestamps in a driver app. That data will make any later optimization work much better.
How it connects to the rest of the chain
Routes depend on what happens before and after the road. Documents held at the border delay stock that was supposed to be on tomorrow's van. For that side of the problem, read our piece on customs document AI.
If you are a retailer with your own deliveries, our retail and e-commerce page covers how AI fits alongside inventory and customer service. For other options we offer, see our AI solutions.
How to start
- Choose a zone with steady daily volume and a dispatcher willing to give feedback.
- Baseline it for two to four weeks. Capture distance, stops, on-time rate and planning hours.
- Check your addresses. Geocode a sample and measure how many are wrong.
- Run the planner in shadow mode, then on a few live routes.
- Review weekly with drivers and dispatchers, and decide on scale-up only after the numbers and the people agree.
If you want help picking the first zone or checking whether your data is ready, our consultants can review your operation with you. Book a consultation.
Frequently asked questions
What is AI route optimization?
It is software that decides which vehicle serves which drops and in what order, then re-plans as orders, traffic or vehicles change. It combines routing solvers with models that predict travel and stop times.
How is it different from the routing we use now?
Fixed rules and map distances ignore live traffic, real stop times and past outcomes. AI planners use observed data and update during the day, while dispatchers can still override any plan.
What data do we need to start?
Orders with addresses and time windows, fleet details, historic delivery timestamps, travel time data and your operating rules. Address quality is the most common problem, so check a sample first.
Will it work with our TMS or driver app?
Usually through APIs or file exports. We confirm integration options in a readiness check before any build.
Does tracking drivers raise PDPA issues?
Driver location and customer addresses are personal data. You need a clear purpose, limited retention and controlled access. Check current guidance from Thailand's PDPC and confirm with counsel.
Ready to put this to work?
Turn this playbook into a working system
From a first agent to a national platform, the same team plans it, builds it and runs it with you.
