How Can AI Assist in Delivery Services? 6 Jobs for a Small Courier

AI can help a small courier with six jobs: planning the day's routes, answering rate enquiries, sending arrival-time messages, sorting proof-of-delivery exceptions, matching deliveries to invoices, and reviewing the week's numbers. Route software does the first; a chat assistant handles the others. Begin with the job that eats the most dispatcher minutes.

None of the six needs a developer. Each turns something your dispatcher already does by hand into a repeatable routine, with a person still approving anything that touches money or a promise to a customer. This list is for firms with a handful of vans or riders, not for national networks, so costs and examples stay small.

Follow me on Instagram@sagnikteaches

Job 1: Plan the day's routes in ten minutes

Route planning is the clearest win because the computer is better at it than any person: choosing the order of 25 stops from thousands of possible orders, respecting time windows and vehicle capacity. This is not a chat-assistant job. Use route-planning software, and use ChatGPT, Claude or Copilot only to tidy the input and explain the output.

Connect on LinkedInSagnik Bhattacharya

The common choices are Routific, Spoke Dispatch (formerly called Circuit for Teams), Route4Me and Onfleet. At the pages read for this tutorial: Routific's price starts free for up to 100 orders a month, then a flat $150 a month covers up to 1,000 orders with per-order rates above that; Spoke Dispatch's Starter plan is $125 a month including 1,000 stops, then $0.04 a stop, and its Premium plan is $200 for 2,000 stops; Onfleet's Launch plan starts at $619 a month for up to 2,500 completed deliveries. Route4Me's page lists plans at $199, $299 and $349 a month without saying whether that is per user, per driver or per vehicle, so ask before you compare it with the others. Prices change, so confirm before you buy. For a closer look at tools and what to ask of them, can a small courier firm use AI to plan delivery routes goes through the options in more detail.

Subscribe on YouTube@sagnikteaches

The main worked example: a courier with three vans does about 1,500 stops a month, around 23 stops per van per working day over 22 days. The dispatcher currently sorts orders on a map for 70 minutes each morning. With a planner, importing the spreadsheet of that day's jobs and optimising takes about 10 minutes. That is 60 minutes a day, 22 hours a month. On those volumes, an illustrative bill is about $145 on Spoke's Starter ($125 plus 500 extra stops at $0.04) or about $225 on Routific ($150 for the first 1,000 orders plus 500 at the 15-cent band). If the dispatcher's time is worth $20 an hour, the saved 22 hours are worth about $440 against that bill, before counting any change in miles driven. Measure the miles for two weeks before and after instead of trusting a vendor's promised percentage.

Where the chat assistant helps

The messy part is the input: orders arrive as emails, phone notes and a shared sheet, with addresses spelt three ways. Paste a day's raw list into the assistant and ask it to return a clean table with one address per row, the delivery window, the parcel count and a "needs checking" flag for anything ambiguous.

Clean this list for import into a route planner. Return a table with columns: customer, address line 1, town, window start, window end, parcels, notes. If an address is missing a number or the window is unclear, put "CHECK" in notes and do not guess.

[paste raw list]

Illustrative output:

| customer | address line 1 | town | window | parcels | notes |
| Dental surgery | 12 [street] | [town] | 09:00-12:00 | 2 | |
| [customer name] | Flat 3, [building name] | [town] | 14:00-17:00 | 1 | CHECK: no street number |
| Pharmacy | 8 [street] | [town] | before 11:00 | 4 | Fridge item |

Ten stops flagged "CHECK" in a day is a signal to fix the order form, not to buy better software.

The constraints a planner needs to know

A plan is only as good as the rules you give it, and most bad plans come from rules nobody entered. Before the first run, write down the constraints your dispatcher carries in her head and put each one into the tool, or into the spreadsheet the tool imports:

• Time windows. Which customers need a morning slot and which accept any time, entered per stop.
• Vehicle limits. The small van that cannot take a pallet, the van with the tail-lift, the rider who carries only envelopes.
• Service time. A pharmacy drop takes two minutes; a delivery up three flights of stairs takes ten. Planners that assume a flat three minutes per stop will run the afternoon late.
• Start and end points. The depot, or the driver's home if a van goes home overnight.
• Breaks. Where the driver's lunch falls, and how long it lasts.

Ask a chat assistant to turn your dispatcher's explanation into that checklist. Record her talking through a normal morning for five minutes, paste the transcript, and ask: "List every rule or constraint she mentions, one per line, and mark each as hard (must happen) or soft (preferred)." Hard rules go into the tool as constraints. Soft rules are the ones you accept breaking when the day gets tight.

A simple test of the first plan: ask the driver. If a driver says the order is wrong in the first week, find out why, because a missing rule is usually the answer and the tool is right for the rules it knows.

Job 2: Quote delivery enquiries from your rate card

A courier loses jobs by replying slowly. A business owner who asks for a quote at 9am has often asked three firms by 9:15. If your rates sit on a card (zone prices, weight bands, out-of-hours surcharges), an assistant can draft the reply in under a minute for a person to approve and send.

The approach is covered in detail in AI courier quotes: replying to delivery enquiries from your rate card. The short version: put the rate card in a Project (ChatGPT Projects or Claude Projects) so it is always in the assistant's context, along with three rules (never quote outside the card, never promise a time without a dispatcher check, always state what is excluded), and paste each enquiry.

A filled-in example. The enquiry: "Two boxes, about 12 kg each, from our unit to a client 18 miles away, today if possible." The rate card, illustrative: base $14 up to 10 miles, $0.90 for each further mile, $6 for each parcel over 10 kg, same-day surcharge of $12. The draft the assistant returns:

Hello [first name],

Thank you for your enquiry. For two boxes of about 12 kg from [pickup] to [drop], 18 miles, the price is $50.40:
- Base rate up to 10 miles: $14.00
- 8 extra miles at $0.90: $7.20
- Two parcels over 10 kg at $6: $12.00
- Same-day service: $12.00 (subject to availability; we will confirm a collection time shortly)

Regards,
[your name]

Check the sum: 14 + 7.20 + 12 + 12 = $45.20, not $50.40. The assistant made an arithmetic slip, which is common when it adds several lines. That is why the human check stays: confirm the total against the card, or have a spreadsheet do the sum and let the assistant only write the wording. Better still, ask it to show the working in a table so the check takes ten seconds.

Job 3: Arrival-time messages that cut "where is my parcel" calls

The cheapest call to answer is the one that is never made. Most customer calls to a courier ask one question: when will it arrive? Routing tools that include tracking send an automatic text or email when a driver is a set number of stops away. Onfleet's pricing page lists automated recipient SMS notifications and a branded live tracking page from its Launch plan, with dynamic ETA notifications added on the higher Scale plan. Spoke Dispatch lists live tracking and customised SMS notifications on all three plans, with a branded tracking page and a custom sender name from Premium upwards. Routific's free tier leaves SMS notifications out. Check that the notification feature you want is on the plan you can afford, and that texts are billed separately in some cases (Onfleet says SMS and voice usage is billed apart from the plan).

The AI's role is the wording. Default templates are stiff and uninformative. Ask the assistant to draft three versions of each message type, then fix the best into the tool:

Write three versions of an SMS a courier sends when a driver is 5 stops away. Maximum 160 characters. Include the first name, a two-hour window and a way to reply if they will be out. Friendly, no emoji, no exclamation marks. Placeholders: [first name], [window], [reply number].

Illustrative result and edit: "Hi [first name], your parcel is next on our route and should reach you between [window]. If you will be out, reply to [reply number]." Counting the characters with real names and a window like "2:10-3:10pm" often pushes past 160, which splits the text into two billing units, so shorten until a typical real message fits.

The before-and-after is the evidence. A courier counts "where is it" calls for two weeks (say 38 calls across 10 working days), turns on the messages with the improved wording, and counts again. If the number drops to 15, that is 23 fewer calls over the fortnight, about 11 a week. At roughly five minutes each, counting the time it takes to get back into the routing she was doing, the dispatcher recovers close to an hour a week. Use your own counts. For a longer treatment of tracking messages, cutting parcel-tracking calls with AI delivery updates has the templates.

Three standing messages worth writing once

Beyond the "five stops away" text, three more messages repay an hour of careful drafting. A "booked" message sent when the order arrives, stating the delivery window and how to change it. A "missed you" message after a no-answer, with the next available slot and one reply option. A "delivered" message with the photo link, so a customer who is not at home can see where the parcel was left. Ask the assistant for versions at three lengths and read each aloud; a message that sounds odd spoken sounds odd read on a phone. Keep one master document of all the final wordings so a new dispatcher does not rewrite them.

Job 4: Sorting proof-of-delivery exceptions

Proof of delivery (a photo, a signature or a note) is simple when a handover goes well. The work is in the exceptions: no-one in, parcel left with a neighbour, damaged box, refused delivery, wrong address, signature missing. Each exception needs a decision, and drivers' notes are free text typed on a phone: "no1 in left card", "nbr took it #4", "box wet corner".

A chat assistant can sort a day's notes into categories and draft the customer message, which saves the dispatcher from reading 30 lines of shorthand. Ask for a fixed set of categories so the output is consistent and can be filtered:

Classify each driver note into exactly one category: DELIVERED-OK, NO ANSWER, LEFT WITH NEIGHBOUR, DAMAGED, REFUSED, WRONG ADDRESS, OTHER. For each non-OK row, add a one-line suggested next step. Do not decide whether a claim is valid.

Notes:
[paste]

Illustrative output:

| stop | note | category | next step |
| 14 | no1 in left card | NO ANSWER | Rebook, text customer |
| 19 | nbr took it #4 | LEFT WITH NEIGHBOUR | Confirm neighbour name with driver, tell customer |
| 22 | box wet corner | DAMAGED | Collect photo from driver, hold for dispatcher decision |

The line "do not decide whether a claim is valid" matters. A damage dispute can carry cost and contract terms, so a person reads the photo, applies your terms and replies. Handling damage claims and disputes with AI goes through the steps, and the draft reply the assistant writes should always be read before it goes out.

A realistic slip in this job: a driver typed "left safe place" for a parcel at a gated address, the assistant filed it under DELIVERED-OK, and the customer then said it never arrived. The category should have been a flagged one. The fix was a rule in the prompt that any note containing "safe place", "porch", "gate" or "behind" goes to a category called LEFT UNATTENDED, which needs photo evidence. Add words like these as you learn how your own drivers write.

Drafting the customer reply to an exception

Once a dispatcher has made the decision, the assistant can write the message. For the wet-corner box, the dispatcher decides to offer a redelivery or a refund of the delivery charge under her own terms, and gives the assistant that decision. The prompt carries the decision, not a request for one:

Write a short, plain email to a customer. Facts: parcel delivered today at 14:20, outer box corner was wet when handed over, our photo is attached. Decision already made by our dispatcher: we will collect the parcel tomorrow for inspection and refund the delivery charge of $9 whether or not the contents are damaged. Do not accept liability for the contents; do not promise a replacement. Apologise once. Under 90 words.

The assistant's version will be warm and usually too long; trim it, check that every sentence matches the decision and that it promises nothing extra. A slip to watch for is an apology that reads as an admission, such as "we are sorry we damaged your parcel", when the facts say only that the box was wet.

Job 5: Match completed jobs to invoices

Many small couriers lose money after the delivery, not before it: an account customer's weekly jobs sit in a driver app or a spreadsheet, and the invoice is built by hand on Friday. Jobs get missed, waiting time is not billed and surcharges are forgotten.

Export the week's completed jobs and the customer's rate agreement, then ask the assistant to build the draft invoice lines and to say what it could not price. The sum below is illustrative: a business customer with a standard $9 per local drop, $4 extra for a return leg and $15 an hour waiting time beyond ten minutes.

Using the agreement below, price each job and total the week. Show the arithmetic per line. List any job you cannot price, with the missing detail.

Agreement: local drop $9; return leg +$4; waiting beyond 10 minutes $15 per hour, pro rata.
Jobs: [paste]

A sensible check is to total the week on a spreadsheet with a formula as well, and compare to the assistant's total. A difference means a line was misread. Once the habit holds, the assistant's value is catching the unbilled extras: waiting time and return legs, which typically sit in the driver's note rather than a field. Suppose three return legs at $4 and 50 minutes of waiting at the pro-rata rate ($12.50) were found in one week; that is $24.50 that would otherwise have gone unbilled, or roughly $1,270 over a year if it recurs weekly. Treat that as an illustration of how the sum is done, not a forecast.

This job touches customer data and prices, so use a business account with training off, and paste only the columns you need. Reading delivery notes and updating stock automatically describes the same extraction idea for incoming goods.

Job 6: Review the week and find the repeat problems

A small courier usually knows the day's problems and rarely sees the weekly pattern: the same industrial estate always late, the same customer always out, one driver's route always overrunning. A weekly export of jobs (time promised, time delivered, outcome, area, driver) is enough for the assistant to find them.

Here is a week of deliveries: stop id, area, promised window end, delivered time, outcome, driver. Tell me: (1) on-time rate overall and by area; (2) the three areas with the most late or failed drops; (3) any customer with more than two failed attempts; (4) one change I could test next week. Show the numbers behind each answer.

The principle is the same as for any spreadsheet question; what to upload when AI analyses a spreadsheet, and what to check lists how to verify. Let the assistant propose, then check two or three of its numbers by filtering the sheet yourself.

Illustrative finding: of 310 drops, 273 were on time (88%). Two areas account for 21 of the 37 late drops. One customer has three failed attempts. The assistant suggests moving that customer to an evening window and testing a 20-minute earlier start on the two late areas. The dispatcher tries the earlier start for a week and compares. Nothing here is clever. The value is that the review happens every Monday, in 15 minutes, rather than never.

A Monday review you can finish in fifteen minutes

1. Export last week's jobs from your routing tool to a sheet and delete personal columns (names, phone numbers).
2. Paste into the assistant with the standing prompt above.
3. Check two figures by filtering the sheet yourself: the on-time rate and the worst area's count.
4. Pick one change to try, write it in a "tests" tab with the date, and compare next Monday.
5. If a driver's name keeps appearing at the top of the late list, look first at that driver's route and stop count before you look at the person.

The order to introduce them in

You do not need all six. A sensible sequence for a three-van firm:

1. Weeks 1-2: route planning (Job 1). Biggest daily time saving, tool free or cheap to trial.
2. Weeks 3-4: arrival-time messages (Job 3). Needs the tool from step one and a better template.
3. Month 2: exception sorting (Job 4), once you have enough drivers' notes to see patterns.
4. Month 2-3: quotes (Job 2) and invoice matching (Job 5), which carry money and need the most checking.
5. Ongoing: the Monday review (Job 6).

For each, write down the minutes it takes today before you change anything. Without that number you cannot say whether it worked. After two weeks, keep it, change it or drop it.

A day with all six running, for a three-van firm

Illustration only. At 7:30 the dispatcher pastes the overnight orders into the assistant, which returns the clean table with two rows flagged CHECK; she phones both customers and corrects them. At 7:45 she imports the sheet into the route planner and has three routes by 7:55; she checks them with the drivers over a coffee. The tool sends "booked" texts as drivers set off. Mid-morning, a business customer emails for a same-day quote; the assistant drafts the reply from the rate card and she corrects the total before sending. In the afternoon the tool's arrival texts go out automatically, and the "where is it" calls that used to fill her hour between two and three come in at a fraction of that. At 5:30 she pastes the drivers' notes into the exception sorter and finds one damaged parcel and two no-answers; she decides the damaged parcel's outcome and has the assistant write the message. On Friday she runs the invoice matching, and on Monday the weekly review. None of that needs a technical person, and each step has a human decision in it.

What to keep away from AI in a delivery business

Some jobs are poor fits, whatever the tool's promise. A chat assistant should not decide liability for damage, promise delivery times on its own, or see customers' full addresses and phone numbers when a routing tool already holds them. It should not be the only route check on a day with heavy or fragile loads; a driver who knows a bridge has a weight limit outranks any suggestion.

Rules that hold up in practice:

• A person approves any message that states a price, a time or a refusal.
• Names, addresses and phone numbers stay inside the routing and tracking tool; paste only what a task needs into a chat assistant, and prefer a business plan where your content is not used for training by default.
• Drivers can override the plan. Keep a log of overrides, because they teach you where the data is wrong.
• Every automated text includes a way to reach a human.

A quick cost check before you buy anything

Add up four numbers: the monthly cost of the tools you intend to use, the minutes each job takes now, the number of times a month you do it, and the hourly value of whoever does it. Illustration for the three-van firm: routing $145; chat assistant $20; time saved on routing 22 hours; messages and exceptions 8 hours; quotes 4 hours. That is 34 hours a month against $165 in tools. At $20 an hour the saved time is $680. If a month in you cannot find the saved hours, the tools are not the problem; the process around them is. Look for steps where the dispatcher still re-types data from one place to another, and fix those first.

Also count the costs people forget: the hour a week of checking, the extra SMS charges that come with more messages, and any plan upgrade when your stop count crosses a limit. A courier at 1,000 stops a month on a plan that includes 1,000 stops will feel the change when a busy December adds 400.

Questions couriers ask before they add AI to dispatch

Which job should a small courier automate first?

Start with the job that costs you the most time every day. For most small firms that is morning route planning, followed by customer update messages. Pick one, measure the minutes it takes now, run the new way for two weeks, and compare. Do not start three at once.

Do I need a route planner as well as ChatGPT?

Usually yes. Route planners such as Routific, Spoke Dispatch, Route4Me and Onfleet solve the stop-order problem with maps and time windows. A chat assistant is better at writing messages, sorting driver notes and summarising the week. They do different jobs and work well together.

Can AI decide whether a delivery counts as failed?

No. It can sort driver notes into categories and draft the customer message, but a person decides whether a parcel was delivered, damaged or disputed, because that decision has money and contract consequences. Treat the AI's category as a suggestion for a dispatcher to confirm.

What does this cost a courier with three vans?

At about 1,500 stops a month, the list prices checked for this tutorial put a route planner at roughly $145 to $225 a month on Spoke Dispatch Starter or Routific; Onfleet starts at $619. A chat assistant costs about $20 a month per person on an individual plan. Check each vendor's current price before you commit, and compare it with the dispatcher time saved.

Further reads

• Pizza Delivery Software: Which Options Handle Dispatch and Texts? — See how delivery texts work for food rather than parcels.
• AI Route Planning for Small Delivery Fleets and Mobile Teams — Fleet-size thresholds for choosing a route planner.
• When Should an AI Chatbot Hand Over to a Human? — When a website chatbot should pass a customer to a person.
• AI Software for Trucking Companies: What to Check for a Small Fleet — What to check in software for a small trucking fleet.
• AI Job Scheduling and Dispatch for Trades and Field Service Teams — Scheduling logic for visits rather than drops.
• Rental Car Companies Using AI: Seven Uses a Small Fleet Can Copy — Seven uses of AI for a small vehicle fleet.
• How Can AI Help Courier Services? Quotes, Dispatch and Proof — What AI does for a courier firm with one to five vans: quote replies, driver messages, day-plan checks and delivery records, and what stays with a person.
• AI Dispatch Automation for Small Delivery Firms: Where to Begin — A small delivery firm's first dispatch pilot: capture jobs, suggest assignments, handle exceptions and measure results.
• How Does AI Help With Transport? Dispatch and Delay Alerts for 10 Vans — A 10-van firm's day modelled step by step: AI-built routes, a dispatch board that flags drift, and delay messages that go out before customers ring.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Routific, Spoke Dispatch, Onfleet and Route4Me pricing pages (read 5 November 2026); Twilio's guide to SMS character limits and segments. Prices are list prices in dollars and change; check the vendor's page. Examples are illustrative.