A small courier business can use AI in three places: drafting quote replies from your rate card, turning the day's job list into clear driver messages and a checked plan, and condensing proof-of-delivery records into answers for disputes. It writes and sorts. Your rate card sheet, route tool and delivery app still do the sums, the maps and the evidence.
Fleet size decides which of the three matters most. A one-van owner-driver gains most from fast quote replies and a tidy end-of-day record. At five vans the gain moves to dispatch messages and exceptions. Guides written for enterprise fleets, with predictive routing and autonomous delivery, say little about either, and none of that is needed to start.
A sheet prices the job, the assistant writes the reply
A quote has two halves that need different tools. The price is arithmetic with rules, so it belongs in a spreadsheet that gives the same answer every time. The reply is writing, which is where the assistant earns its place. If you mix the halves by pasting your rate card into a chat and asking "how much?", you get a confident number that is sometimes wrong and impossible to audit.
A one-van business might use a rate card like this (illustrative):
| Item | Rate |
|---|---|
| Base fee (collection plus the first 5 miles) | $12.00 |
| Each mile beyond 5 | $1.60 |
| Weight band, 10 to 25 kg | +$5.00 |
| Stairs or no lift at either end | +$6.00 |
| Same-day booking | +25% of the total so far |
| Waiting after the first 10 minutes | $0.60 a minute |
In a calculator sheet, put the base fee in B2, the miles in B3, the per-mile rate in B4, the weight-band fee in B5, extras such as stairs in B6, and the same-day uplift in B7 (zero if the job is not same day). One formula does the work:
=(B2+MAX(0,B3-5)*B4+B5+B6)*(1+B7)Now an enquiry arrives (illustrative): two boxes, 14 kg in total, from a workshop to a clinic reception 18 miles away, wanted today. Thirteen miles are beyond the included five, so 13 x $1.60 is $20.80. Add the $12.00 base and the $5.00 weight band and you have $37.80. The same-day uplift of 25% brings it to $47.25. That figure comes from the sheet. Only then does it go into the prompt:
You are replying to a delivery enquiry for a one-van courier business.
Use ONLY the facts below. Do not change the price or add promises.
Price: $47.25 including up to 10 minutes' waiting at each end.
Job: two boxes, 14 kg in total, collect from a workshop, deliver to a
clinic reception, 18 miles, today.
Collection window: 1:00 to 1:30 pm. Delivery target 3:30 pm. We do not
guarantee delivery times.
Extra waiting: $0.60 a minute after the first 10.
Payment: invoice on delivery.
Proof: photo of the boxes and the name of the person who accepts them.
Write a friendly reply of under 120 words, signed [your name].
End with one question: please confirm the contact name and phone
number at the workshop.A typical result (illustrative):
Hi [first name], thanks for your enquiry. We can collect your two boxes (14 kg) between 1:00 and 1:30 pm today and aim to deliver to the clinic reception by 3:30 pm. The price is $47.25, which includes up to 10 minutes' waiting at each end; after that it is $0.60 a minute. You will get a photo and the name of the person who accepts the boxes, and we invoice on delivery. Could you confirm the contact name and phone number at the workshop? [your name]
That draft is usable. The failures to look for are the ones that creep in when the prompt is looser: "guaranteed by 3:30", "fully insured" or "handled with care" appearing from nowhere, a price quoted twice with different rounding, or a promise of a call before arrival that you did not offer. Search the draft for "guarantee", "insured", "will arrive" and "always". The prompt's last instruction also matters, because a reply that ends with a question gets answered faster than one that ends with "let us know".
Two controls keep this safe as volume grows. Give every quote a number and a "valid until" line, because fuel and your own costs move. And for anything above a value you choose, have a second person approve the reply, as setting up an approval step for AI-written quotes describes. The full rate-card workflow, with more job types such as multi-drop runs and return journeys, is in AI courier quotes from your rate card.
One van to five vans: what changes at each size
The same three jobs look different as the fleet grows. Job counts below are illustrative, so replace them with your own.
| One van | Two or three vans | Four or five vans | |
|---|---|---|---|
| Who dispatches | You, between drops | You or a partner, early morning | A part-time dispatcher, or the owner for half the day |
| Jobs a day | 20 to 30 | 40 to 80 | 100 to 150 |
| What AI does | Quote replies; checks the day plan; customer ETA texts; an end-of-day record | The same, plus one standard job message per driver and a shared exceptions list | The same, plus the dispatcher's exception messages and a daily summary of late, failed and disputed jobs |
| Tools | A chat assistant on a free or $20 plan; a free driver app, or Google Maps for a handful of stops | A route planner with a free tier; one shared message group | A dispatch platform with proof of delivery and customer tracking |
| Stays with a person | Price sign-off, liability | Who drives what, price sign-off | Allocation, exceptions, pricing, claims |
On tools, a few checked details. Spoke's homepage describes its Route Planner driver app as free. Routific's pricing page says it charges for the orders you deliver rather than per driver or vehicle, and that its free plan has every feature except SMS notifications. The page no longer states the free plan's monthly order cap (it was about 100 orders in September 2026), so confirm it before you plan around it. At the dispatch-platform end, Spoke Dispatch lists Starter at $125 a month including 1,000 stops, then $0.04 for each extra stop, with unlimited drivers.
Take the five-van end as a sum. Five vans at 28 stops a day over 22 working days is 3,080 stops. That is 2,080 beyond the 1,000 included, so 2,080 x $0.04 is $83.20, and the month costs $208.20, about seven cents a stop. If a part-time dispatcher earns $20 an hour, the platform pays for itself once it saves roughly two and a half hours of dispatching a week ($208.20 over four and a third weeks is about $48 a week). Confirm how the vendor counts a stop, because pickups and deliveries may count separately.
Google Maps is the free tool most people start with, and it is fine for a handful of drops. It does accept several destinations, but its "Depart at" and "Arrive by" options only work for a route with one destination, so it cannot plan around delivery windows. Past that point you want a route tool, and can a small courier firm use AI to plan delivery routes goes through the options.
Checking the day plan before the van leaves
A chat assistant is a poor route optimiser, because it has no real map and will happily invent a drive time. It is a decent checker for a short list when you supply the drive times from your map app. Take this illustrative morning, allowing 10 minutes at each stop:
| Job | Window | Drive from previous stop |
|---|---|---|
| 1. Collect, workshop | 09:00 to 09:30 | Start here |
| 2. Deliver, clinic | By 10:30 | 35 minutes |
| 3. Collect, print shop | 10:00 to 11:00 | 20 minutes |
| 4. Deliver, office | By 12:00 | 25 minutes |
| 5. Collect, bakery | 11:00 to 11:30 | 40 minutes |
The instruction is: "Work through the jobs in this order, showing arrival and departure time at each stop. Flag any window I miss. If a drive time is missing, ask me instead of guessing." The answer (illustrative):
1. Workshop: arrive 09:00, leave 09:10.
2. Clinic: arrive 09:45, leave 09:55. Deadline 10:30 met.
3. Print shop: arrive 10:15, leave 10:25. Window met.
4. Office: arrive 10:50, leave 11:00. Deadline 12:00 met.
5. Bakery: arrive 11:40. The window closes at 11:30, so this misses by 10 minutes. Options: ask the bakery for a window from 11:45, or give me the drive time from the print shop to the bakery so I can test collecting it before the office.
The assistant has found the clash, and you have decided nothing yet. Calling the bakery is your call, and so is giving up the office deadline. Check one line by hand before trusting the rest (10:25 plus 25 minutes is 10:50, which agrees). Ask for the times line by line, because a plan given as a paragraph hides arithmetic slips. This approach suits five or six jobs. Beyond about ten, use a route tool.
Driver messages that leave no room for guessing
Most dispatch errors start as a phone call or a message typed in a hurry: missing gate code, no call-ahead request, wrong contact. The fix is one standard job message that every driver reads the same way, with the assistant filling it from your rough note. Here is a rough note (illustrative) and the instruction:
Note: [name], delivery to 14 Elm Road, 2 boxes cake stuff, call before,
back gate, 10-12, cash 35
Instruction: Turn the note into the standard job message below. Do not
add any instruction that is not in the note. Write NOT STATED for any
field the note does not cover.
JOB [number] | [type] | [window]
Contact and address:
Parcels:
Call ahead:
Access:
Payment:
Proof required:
If no answer:What comes back (illustrative):
JOB 4 | Delivery | 10:00 to 12:00
Contact and address: [name], 14 Elm Road
Parcels: 2 boxes, cake supplies. Handle upright.
Call ahead: Yes
Access: Back gate
Payment: Collect $35 cash, give receipt
Proof required: NOT STATED
If no answer: NOT STATED
Two things to fix. "Handle upright" is not in your note, so it must go, because a driver will follow it literally and a customer will quote it back. And a "Proof required: NOT STATED" line is the system working: it shows that your default (photo plus recipient name) needs adding, which is better than a silent blank. Both fixes take ten seconds, and they are why you read before you send.
Customer messages follow the same rule. An ETA text such as "Hi [first name], your parcel from [sender] is about 25 minutes away. Reply here if you need to change anything." runs to roughly 110 characters. Keep it plain: a text with an emoji or a curly apostrophe is encoded differently, a segment then holds 70 characters instead of 160, and longer messages are split and billed as several (Twilio documents this; ask your provider how it counts). The deeper treatment of tracking messages is in cutting parcel-tracking calls with AI delivery updates.
Keep customer details out of the chat where you can. Full addresses and phone numbers belong in the dispatch tool; when you draft in a chat assistant, use a first name, a street and a job number, and check that the model-training switch in your privacy settings is off if you use a consumer plan.
Proof of delivery: what the record must hold
Proof of delivery is evidence, and the evidence has to be created at the doorstep by an app, not written up later by anyone, human or AI. A record that holds up in a dispute has these fields:
| Field | What it settles | Where it should come from |
|---|---|---|
| Delivery time | "I was told 2 pm" arguments | The app's timestamp |
| Location | Whether it reached the right address | The app's geotag |
| Photo of the parcels in place | Condition and placement | Taken inside the app |
| Name or signature | Who accepted it | Captured in the app |
| Driver note | Why an exception happened | Typed or dictated by the driver |
| Instruction followed | "Leave in the porch" was authorised | The original job message |
Check what your tool actually captures, because plans differ. Spoke Dispatch's pricing page shows signature capture, geotags and timestamps on its Starter plan, a cap on photos (five on Starter, ten on Premium, per its comparison table), and data history kept for 30 days on Starter against a year on Premium. That last figure is the one people miss: if a claim can arrive in week six, export the records before they expire.
The assistant's part is downstream of the record. It can draft the reply to a claim, summarise the day's exceptions and explain a timeline in plain words. Take a claim (illustrative): "My parcel never arrived, but your tracking says delivered at 2:14 pm on Tuesday." The record shows a delivery at 2:14 pm, a geotag about 20 metres from the address, a photo of two boxes on a doorstep beside a blue door, and a driver note: "No answer, left in porch as instructed in order notes". The first draft the assistant writes says: "Our records confirm your parcel was delivered to your door." That overclaims. A geotag 20 metres away could be next door, and the customer has not yet said what colour their door is. The reply you send reads:
Hi [first name], sorry your parcel has not turned up. Our record shows two boxes left in the porch at 2:14 pm, with a photo of a blue door. Could you tell us the colour of your front door and whether a neighbour could have taken the boxes in? We will check the address against our record today.
That reply is cooperative and does not concede or accuse. The assistant drafted it from the record, but you decided what it could say. The same discipline applies to photos: ask a chat assistant what a delivery photo shows and you may get a fluent description with an invented detail. OpenAI's own help page warns that image descriptions and captions can be wrong, that object counts are approximate, and that ChatGPT does not process file names or metadata, so it cannot confirm when or where a photo was taken. Use it to describe, never to certify.
Finding records with a missing photo or name does not need AI at all; a filter on an exported sheet does it better. Where the assistant helps is the Friday summary of the week's exceptions, written from that sheet. For disputes involving damage, handling damage claims and disputes with AI goes further, and electronic proof of delivery with photos and AI covers the evidence side.
Where a person takes over
- Failed delivery. The assistant drafts the "we called today" text; you decide whether to re-attempt, hold or return, and whether to charge.
- Damaged or leaking parcels. Photos first, always. You decide the next step and contact your insurer; the assistant can tidy the incident note afterwards.
- Restricted or dangerous goods, and medical samples. Never delegated. Check your own acceptance rules and your insurance terms.
- High-value consignments. You set the identity-check and signature policy, and the driver follows it. No draft overrides it.
- Price disputes and cash handling. You decide goodwill; the assistant can word the message.
- Driver safety and accidents. A person calls. Nothing is typed into a chat first.
Fleet-scale features you can ignore at this size
Much of what ranks for this topic describes features aimed at far larger fleets: demand forecasting from years of order history, live re-optimisation across dozens of vehicles, autonomous or drone delivery. A one-to-five-van business does not generate the data volume the first needs, does not have the vehicles for the second, and cannot buy the third. A useful test for any feature is whether it changes something you do before 9 am tomorrow. A standard job message does. A forecasting dashboard does not.
The things that do change your week are plainer: replying to an enquiry while the customer is still looking at your site, giving every driver the same message structure, and holding a record you can defend. At one van the cost of all three can stay near $20 a month, until your volume says otherwise.
A two-week trial on one van
Pick one van and one fortnight. In days one to five, use the assistant for quote replies and customer ETA texts only, and log two numbers: minutes from enquiry to reply, and how many "where is my parcel" calls you took. In days six to ten, add standard job messages and the end-of-day exceptions summary, and log instructions missed (gate codes, call-ahead requests) and disputes you settled from the record.
Then decide with a simple rule. Keep a step if it saves at least 15 minutes a day or prevented a visible error. Drop it if you spent longer fixing drafts than you would have spent writing them. A quick sum shows the stakes: eight enquiries a day at six minutes each by hand, against two minutes with a draft and a check, saves 32 minutes a day. Over 22 working days that is 704 minutes, nearly 12 hours, or about $290 of your time at $25 an hour, against an assistant plan of about $20 a month. Your own numbers will differ, which is why you log them.
Questions from owner-drivers and small fleets
Can ChatGPT plan my delivery routes?
It can group stops into sensible clusters, but it cannot measure real drive times or hold time windows reliably. Test it once: give it ten stops, enter its order into your map app, and compare the total drive time with the app's own best order. If the assistant's order is longer, use a route tool for the sequence and keep the assistant for the messages around it.
Is a photo enough as proof of delivery?
A photo helps most when it is taken inside the delivery app, so the time and location are recorded with it. A picture from a phone's camera roll is easier to dispute. OpenAI also says ChatGPT does not process file names or metadata, so never ask a chat assistant to confirm when or where a photo was taken. Pair the photo with a recipient name or signature.
Do I need a dispatch app with only two vans?
Usually not. A route planner's free tier and one shared message group can carry two or three vans. The tipping point comes when someone spends an hour a day answering where-is-my-driver calls or retyping jobs. At that point a platform such as Spoke Dispatch, listed from $125 a month, can pay for itself.
How long should I keep delivery records?
Longer than your tool keeps them, because claims can arrive late. Spoke's Starter plan, for example, keeps data history for 30 days. Export proof-of-delivery records to a folder each month, and ask your insurer or adviser how long to hold them for damage and loss claims.
Further reads
- Using AI Services for Customer Support: Courier Parcel-Query Replies — Replies to where-is-my-parcel messages, drafted safely.
- AI in Logistics for Small Delivery Firms: Routes, ETAs and Messages — A three-van worked example covering route planning, delivery windows, delay messages and proof-of-delivery replies, with named tools and monthly costs.
- How to Market a Courier Business: B2B Outreach Written With AI — A target list, a three-email sequence and a one-page rate card for a small courier firm, drafted with AI and fixed by hand.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Spoke Dispatch pricing page, Routific pricing page, Google Maps help (directions with several destinations), OpenAI's ChatGPT image inputs FAQ, and Twilio's help article on message length and its glossary entry on SMS encoding, all read in October 2026.