AI helps a small haulage or courier firm by building the day's routes from the order list, flagging vans that are drifting behind plan, and drafting the customer message before the customer rings. A dispatcher still approves changes. For ten vans, the gains come from fewer wasted miles and fewer "where is my delivery?" calls.
The money is rarely in the route itself. A decent planner saves a few per cent of mileage, but the larger saving is in dispatcher time: ten vans generate a surprising number of calls, and most of them are about the same thing. Below, one modelled day for a 10-van firm shows where AI sits in each step, and where it gets the ETA wrong.
The firm: ten vans, about 180 stops a day
Take an illustrative regional delivery firm: ten vans, each making around 18 drops on a weekday, so roughly 180 stops, with two people in the office, one of whom dispatches. Customers are mostly trade accounts with delivery windows of two hours, plus some households. Most stops take 4 to 6 minutes on site, but a few need a pallet unloaded and run to 20 minutes. Orders arrive by email, a customer portal and a daily spreadsheet from the three biggest accounts.
Before any AI, the dispatcher spends about 90 minutes each morning assembling routes in a spreadsheet and a map, then handles around 25 inbound calls a day, many of them "is my van late?". Average mileage is 105 miles per van per day. These are made-up figures for modelling, but they are plausible, and you can replace them with your own from last month's data.
The day in four stages
| Time | Stage | What AI does | What the dispatcher does |
|---|---|---|---|
| 05:30 | Order intake | Reads emails and spreadsheets into one clean stop list, flags incomplete addresses and odd weights | Fixes flagged rows, confirms special handling |
| 06:15 | Route build | Planner proposes routes by van capacity, windows and shift hours | Reviews outliers, moves stops, locks the plan |
| 07:00-17:00 | Live board | Compares planned and actual arrival at each stop, drafts delay messages | Approves messages, reassigns stops, calls drivers on real problems |
| 17:30 | Review | Summarises late stops, failed drops and driver notes | Decides rule changes for tomorrow |
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowStage one: turning the morning's orders into a clean stop list
The route planner is only as good as the list that goes into it. In the 10-van firm, bad input causes more late deliveries than bad routing: an address with a typo puts a pin in the wrong town, a missing weight lets a van overload, and a note like "back entrance, ask for Sam" sits in an email nobody carries across.
A general assistant can do the tidying before the list goes into the planner. Paste the combined rows (business names and addresses are normal working data here, but check your own contracts and use a business plan, which doesn't train on your content by default) and ask for a check.
Here is today's delivery list. For each row, check:
- the address looks complete and the street matches the town
- weight and item count are present and plausible
- a delivery window is present
- the notes mention access, appointment or special handling
Return only the rows with a problem, one line each, saying what to fix.
Do not change any addresses yourself.
[paste rows]An illustrative output lists nine rows out of 180: three with an address that doesn't match the town, two with no weight, three with no window and one with the note "call before arrival" that needs a phone number. The dispatcher fixes them in six minutes, where before each would have surfaced as a failed drop at 14:00. Tidying the list is the cheapest AI job in the whole day and often the most valuable.
Stage two: route building, and what the planner can't see
Route planners such as Routific handle the sums: vehicle capacity, time windows, driver shifts and the order of stops. As an example of cost, Routific's pricing page shows a free tier for up to 100 orders a month and a flat $150 a month up to 1,000, then per-order bands with volume discounts (15 cents an order from 1,001 to 2,000, falling to 10 cents between 3,001 and 5,000). A firm doing roughly 3,900 stops a month should work out its own total from those bands, or ask Routific to confirm it, rather than assume the $150. Text-message updates are a paid add-on with rates that vary by country, which matters for the delay messages below. Spoke Dispatch (previously Circuit for Teams) and other planners sit in the same space; four route planners compared on cost and AI sets them side by side. Every figure above should be checked against the vendor's page on the day you buy.
What the planner can't see is the local knowledge that makes a route workable. Examples from the 10-van firm: the industrial estate gate that closes to vans between 12:00 and 13:00, the school road with a loading ban at pick-up time, the customer whose unit needs a tail-lift. Each of these should become a rule or a stop note in the planner rather than living in the head of the driver who knows. Ask drivers for their top five "never send a van here at this time" facts, and enter them in one afternoon.
The dispatcher's review takes about 15 minutes. Look for four things: vans with very different stop counts, any van with a finish time later than its shift end, stops with tight windows at the end of a long route, and any two vans crossing the same area. AI can summarise the plan against those four checks if you export it, but the final judgement is a person looking at a map.
Illustrative result: the planner builds ten routes that average 96 miles instead of 105, a saving of about 9 miles a van. At a running cost of, say, $0.65 a mile, that is about $58 a day across the fleet, or around $1,200 over 21 working days. The saving is real but modest, and it depends on how messy the old routes were. If the dispatcher was already good, expect less.
Stage three: the live board and the drift alert
This is where AI pays for itself. Each driver's app reports when a stop is completed. The board compares the planned time with the actual one, and the planner re-estimates every later stop. When the estimate for a stop moves past the end of its window, that is a drift alert.
Without rules, a board produces so many alerts that the dispatcher ignores them. Three thresholds keep the board useful:
- Amber, 15 to 30 minutes behind plan: shown on the board, no message sent yet.
- Red, estimate falls outside the customer's window: a message is drafted for the dispatcher to approve.
- Black, more than 90 minutes behind or a van off the road: the dispatcher calls the driver, decides which stops to hand to another van, and tells affected customers personally.
For ten vans, a good day produces perhaps three or four red alerts. A bad day, with a motorway closure, may produce twenty. That is the point of the black level: when the day is breaking, the AI's job is to list the stops at risk, ordered by window end, and suggest which spare capacity can take them, while the dispatcher makes the calls. See a daily exceptions report from route exports for how to produce that list.
Delay messages: a template and a worked example
A delay message should say three things: what has happened, the new window, and what the customer can do. It should not blame a driver or promise an exact minute. An illustrative message for a red alert:
Hi [first name], your delivery from [company] is running about 40 minutes behind this morning. Your new window is 12:30 to 14:30. The driver will phone from the van 15 minutes before arriving. If nobody will be on site then, reply to this message and we'll rebook for tomorrow at no charge.
The AI drafts it from the stop's data: name, original window, re-estimated arrival, driver phone rule. A dispatcher approves a batch of three or four in under a minute. For the template library and the missed-drop wording, late and missed drop texts with AI goes further.
Once a message type has been right for a few weeks, you can let the amber-to-red 30-minute slip send automatically for trade accounts that have opted in to text updates. Keep longer delays and failed deliveries with a person, because those messages need judgement. A quick sum shows why this is worth doing: if 14 of 180 stops (about 8%) run late and each one used to produce a 4-minute call, that is 56 minutes of phone time. If proactive messages halve those calls, you save about half an hour a day, which pays for a planner subscription several times over.
How a wrong ETA happens, and four checks to catch it
An ETA is a prediction, and the planner makes it from three inputs: distance and road speed, an assumed time on site at each stop, and the driver's break pattern. Each can be wrong in a predictable way.
| Cause of a wrong ETA | How it shows up | Check |
|---|---|---|
| Default time on site too short | Vans are 10 to 20 minutes behind by mid-morning, every day | Compare average actual stop time with the planner's setting for each customer type |
| Wrong pin on the map | Driver arrives in the right street but at the wrong gate or the wrong end of the road | Count stops where the driver's completion location is more than 200 m from the pin |
| Traffic assumptions | Same two-stop stretch is late at the same time each day | List stops late more than three times in two weeks and look for a pattern by hour |
| Missed driver break or loading time | First stop is always late, or lunchtime stops slide | Check the first-stop arrival against shift start plus loading allowance |
| Stale stop notes | Driver sits waiting for a gate code or a closed bay | Review every note flagged "waited" in the driver app each Friday |
Two weeks of data is enough to calibrate. Export planned and actual arrival times for every stop, ask the AI for the average error by customer type and by hour of day, and pick the largest cause to fix first. An illustrative finding: stops at trade counters average 9 minutes on site where the planner assumed 5, which alone explains why every van loses about 40 minutes a day. Changing one setting makes the ETA more honest for the whole fleet. A related check is accuracy of the window itself: if 85% of deliveries land inside the window promised, you can tighten it by 30 minutes. If it is 65%, widen it and tell customers the reason.
The driver who ignored the plan
A realistic failure: the planner sends a van to a trade estate for a 10:00 to 12:00 window. The driver knows the site's yard is chaos at 10:00 and goes to his last two stops first. He returns at 11:40, an alert has gone out to the customer saying "running 90 minutes late", and the customer's manager has already phoned twice. The AI had worked from planned stops, not from the driver's real sequence.
The fix is two small ones. The driver app should let drivers re-sequence stops, so the board always knows the real order, and the dispatcher should review overrides by close of play, turning the good ones into permanent rules. A driver who is thanked for a useful override will keep telling you things the software can't see, and one who is blamed will stop. For the broader picture of what AI can and can't do in a fleet like this one, AI in the trucking industry for a small fleet owner is worth reading next.
Getting drivers to use the app properly
None of the board's alerts work if drivers tap "complete" at the end of the shift rather than at each door. This is the most common reason a pilot looks disappointing in week two. The planner thinks the van is still at stop 6 when it is at stop 11, so the ETAs it gives for the remaining stops are fiction and the messages that follow go out wrong.
Three habits fix most of it. Explain the benefit in the drivers' terms: fewer calls from the office while they drive, and no argument afterwards about when they arrived. Make the tap a single action, so that the app asks for nothing more than "arrived" and "done" unless a problem needs a note. And review completion times with the team on the first Friday, with real examples: "Van 4 shows five stops completed at 15:52. What happened?" The answer is usually a sensible one, such as the signal was poor on a rural stretch, and the fix is a setting, not a telling-off.
A filled-in morning checklist for the dispatcher
A short written routine keeps the day repeatable when the dispatcher is off sick and someone else steps in. An illustrative version for the ten-van firm:
- 05:30: open the stop list, run the tidy-up check and fix the flagged rows (target: under 10 minutes).
- 06:00: import into the planner, apply today's known exceptions (road closures, a van booked for service) and generate routes.
- 06:15: look at the four plan checks: uneven stop counts, finish times after shift end, tight windows at route ends, vans crossing areas.
- 06:40: lock the plan and push it to the driver apps; send the van-by-van sheet to anyone who prefers paper.
- 07:00 onwards: watch amber alerts, approve red messages in batches of three or four, and phone drivers only for black alerts.
- 17:30: export planned versus actual, run the late-stop summary, write one rule change for tomorrow in the shared notes.
The last line matters most. A single rule change a day, such as adding two minutes to trade-counter stops or a note about a gate code, adds up to over 20 improvements in a month. The plan gets steadily more accurate without anyone ever sitting down for a "big optimisation project".
What this costs and what to measure
For a ten-van firm, the realistic monthly cost is a planner subscription (check the vendor's page, as pricing above the entry tier is banded), a business-plan AI assistant for list tidying and messages at roughly $20 to $25 a seat, and possibly SMS charges for customer texts (Routific, for one, sells SMS as an add-on to its paid plan). For a fleet this size, buying everything at once is overkill; the discipline is to buy one thing at a time, and to measure it.
- Miles per van per day before and after route planning. Expect a small gain.
- Inbound "where is it?" calls per day. The 25 in our model should fall noticeably once proactive messages go out.
- Percentage of stops inside the window. The real service measure.
- Failed first-time deliveries per week. They cost a van return trip each, so a fall here is money.
- Dispatcher morning planning time. From 90 minutes towards 30 is a fair target.
Record these for four weeks before you change anything, so the comparison is honest. To work out what each drop costs once the planner is running, cost per drop from route and fuel data shows how to build the figure. If you'd rather map your own dispatch board with someone, that is what my AI implementation consultation covers.
Where to stop and keep a person in charge
Some decisions stay human however good the planner gets: which customer's delivery to sacrifice when a van breaks down, whether to send a driver over their hours to rescue a late stop, and what to say when a delivery fails and the goods are perishable or urgent. AI can list the options and the stops affected. A dispatcher weighs the customer relationship, the driver's hours and the cost. If a vendor tells you the system "handles exceptions automatically", ask which exceptions, and what happens when it gets one wrong at 16:30 on a Friday.
Questions fleet owners ask about AI dispatch
Do I need telematics in every van for AI dispatch?
Not to start. A route planner with a driver app on each driver's own phone gives you stop-by-stop progress, which is enough for delay alerts. Telematics adds speed, idling and engine data, which matters for maintenance and cost tracking but is not required for the first month of better dispatch.
Can AI promise customers an exact arrival time?
It can estimate a window, and that is the safe thing to give. A single arrival time sounds precise but breaks as soon as one stop overruns. Give customers a one- or two-hour window, update it when the plan changes, and measure how often vans arrive inside it before you narrow it.
Who sends the delay message, the AI or a person?
Start with a person approving each batch. The AI drafts the message from the stop data and the dispatcher releases it. Once a message type has gone out correctly for a few weeks, you can let the routine ones, such as a 30-minute slip, send themselves while longer delays and failed deliveries stay with a person.
What happens when the AI route is worse than the driver's local knowledge?
Let the driver override it and log why. Drivers know that a loading bay closes at noon or a gate needs a code. Add each of these facts to the planner as a stop note or a time rule, so the next day's route already knows. Treat overrides as data, not as insubordination.
Further reads
- Can a Small Courier Firm Use AI to Plan Delivery Routes? — Whether a small courier firm can plan routes with AI.
- AI for a Small Delivery Service: Dispatch, Routes and Customer ETAs — Dispatch, routes and customer ETAs for a small delivery service.
- AI Dispatch Automation for Small Delivery Firms: Where to Begin — Where small delivery firms should begin with dispatch automation.
- AI in Logistics for Small Delivery Firms: Routes, ETAs and Messages — Routes, ETAs and customer messages for small delivery firms.
- How Can AI Help Courier Services? Quotes, Dispatch and Proof — How AI helps couriers with quotes, dispatch and proof.
- How Can AI Assist in Delivery Services? 6 Jobs for a Small Courier — Six jobs a small courier can give to AI.
- Pool Route Software: Can AI Plan Routes and Stop Density? — What pool route software really optimises, how to measure stop density yourself, and where Skimmer, GorillaDesk and Jobber differ.
- Fuel Surcharge Calculator: Set a Weekly Rate for Haulage Quotes — A weekly fuel surcharge policy for haulage and courier quotes: base price, fuel share, step table, spreadsheet formulas and customer clause wording.
- AI for Shipping Companies: Courier Quotes, Tracking Replies and Claims — A small-courier workflow for checked quote drafts, accurate parcel updates and complete damage-claim intake.
- Industries Where AI Is in Use: What Small Firms in Each Actually Do — Seven industries, one concrete small-firm AI job each, with the monthly cost, a sample of the work and the check that keeps it safe.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Routific pricing page; Spoke Dispatch pricing page and help centre; vendor help pages for route planners and driver apps; general dispatch practice.