Can a Small Courier Firm Use AI to Plan Delivery Routes?

Yes. Route-optimisation software plans multi-drop routes in seconds, handling time windows, van capacity and driver shifts, and costs from free at very low volumes to roughly $125-$220 a month for a few vans. It pays off once drivers do 20 or more drops a day or routes change daily. A chat assistant like ChatGPT is the wrong tool for ordering the drops.

Most of what's sold as "AI route planning" is an optimisation algorithm working on road-network and traffic data, with some machine learning on top to predict arrival times and how long stops take. That matters for how you choose: judge tools on the constraints they handle, the driver app and the pricing model, not on how much AI the brochure mentions.

Follow me on Instagram@sagnikteaches

What route-planning software is actually solving

Ordering deliveries is a maths problem with a famous name, the vehicle routing problem. With just 10 drops there are over 3.6 million possible orders; with 25 drops there are more than 15 million billion billion. Nobody checks them all. The software uses methods that find a very good answer quickly, while respecting the rules you give it:

Connect on LinkedInSagnik Bhattacharya

• Time windows: the pharmacy before 9:00, the office between 10:00 and 16:00.
• Service time: how long each stop takes. A signature at a reception desk is not the same as a doorstep drop.
• Capacity: parcels, cages or weight per van.
• Driver shifts: start and finish points, breaks, maximum hours.
• Road reality: actual driving times by time of day, not straight lines on a map.

Split across several vans, the software also decides which van takes which drops, which is often where the biggest savings are.

Subscribe on YouTube@codingliquids

Why a chatbot shouldn't order your drops

It's tempting to paste your address list into ChatGPT or Claude and ask for the best order. Here is what tends to come back, shown as an illustrative example:

Prompt: Put these 30 delivery addresses in the most efficient
order, starting and ending at our depot.

Illustrative output:
Here is an optimised route that groups stops by area:
1. Depot  2. 14 Mill Lane  3. 2 Mill Court  4. Unit 5, Station
Road ... 30. Depot
Estimated total distance: approximately 84 km.

It reads well and is mostly guesswork. The assistant has no road network, so the "84 km" is invented. It grouped stops by street names that look alike, which is not the same as being near each other. It ignored the two 9:00 deadlines in the list, and it can't re-plan when a same-day job arrives at 11:00. For a handful of drops, Google Maps is a better free option, but it only takes up to nine stops including the destination and won't reorder them for you; you drag them into order yourself.

Chat assistants are still useful around the edges: tidying a messy address list, writing driver notes, or drafting customer messages. Just not the sequencing.

When route software pays for a small courier

Route tools and what they cost

The pricing model matters as much as the price. Per-stop pricing suits a few vans doing lots of drops only if the stop count stays within the plan; per-driver pricing suits dense multi-drop work. The worked example below shows how different they come out for the same firm. If you're weighing the wider decision between per-seat and usage pricing, per-seat vs usage-based pricing goes through it.

A trial scorecard for two tools

Free trials are short, so decide what you're testing before the first day. Run the same real day's jobs through both tools and score what matters to your drivers and customers, not the feature list. An illustrative scorecard, with the tools left unnamed because the results depend on your data and the plan you trial:

The cheaper tool lost here because the same-day re-optimisation the firm needed sat on a dearer plan, which would have wiped out the saving. Put the must-haves at the top of the scorecard so a lower headline price doesn't decide it.

A four-van courier's first month with route software

The firm in this example is hypothetical. It runs four vans doing about 150 drops a day between them, a mix of timed business deliveries and parcels, over 22 working days a month.

Before. The owner planned routes from 6:30 each morning with a spreadsheet and a maps app, about 70 minutes a day. Each van averaged about 190 km a day and 37 drops, and three or four timed deliveries a week arrived late.

Choosing a tool. At about 3,300 drops a month, Spoke's Starter plan would cost $125 plus 2,300 extra stops at $0.04, about $217 a month. OptimoRoute Lite for four drivers is $140.40 a month on annual billing. Routific would need a volume quote above 1,000 orders. The firm trialled two tools for a week each, then picked on the driver app and time-window handling, not only price.

Setup, week 1. Export the day's jobs as a spreadsheet with address, time window, parcel count and notes. Set service times: 2 minutes for a residential drop, 6 for a business delivery needing a signature. Set capacity and each driver's start, finish and break.

Weeks 2 and 3. The software's routes ran alongside the owner's own for comparison. Drivers flagged two industrial estates where the pinned location was the wrong entrance, and one customer who is never in before 10:00. All three went into the settings as notes and time windows.

Week 4 results. Planning took 15 to 20 minutes. Average distance fell to about 163 km a van per day. Late timed deliveries dropped to one in the week. Each van had room for four or five more drops a day without overtime.

The money. 27 km saved per van, times four vans, times 22 days, is 2,376 km a month. At an assumed $0.18 a km in fuel, that's about $430 a month, plus around 18 hours of the owner's mornings, against $140-$220 in software. The spare capacity is worth more than either: 16 extra drops a day across the fleet without adding a van.

Rural rounds, long stops and mixed vehicles

Two edge cases change the settings more than the tool. On rural rounds, drops can be 5 or 10 km apart, so the saving from reordering is smaller but the cost of a wrong pin is bigger: a farm address pinned to the middle of a field can add 15 minutes. And if your fleet mixes small vans and a larger van, capacity and vehicle type must be set per vehicle, or the software will happily send the small van to the pallet drop.

The data that decides whether the routes are any good

Route software is only as good as what goes in. The usual culprits, in order of how often they cause trouble:

1. Address pins in the wrong place. Industrial units, new-build estates and rural addresses are often pinned to the wrong entrance or the middle of a site. Save corrected locations for regular customers.
2. Service times that are too optimistic. If a business drop really takes eight minutes, not two, every route after lunch runs late.
3. Time windows missing or too tight. Enter the real window the customer needs, not the hopeful one.
4. Capacity ignored. A route that works on distance but needs 180 parcels in a van that holds 140 isn't a route.

This is where a chat assistant does help. Before import, messy job lists can be tidied with a prompt like this:

Clean this delivery list for import. One row per job with
columns: company or name, street address, town, time window
(HH:MM-HH:MM or blank), parcels, notes. Split combined
addresses, fix obvious typos in street names, and put any
instruction like "before 9" into the time window. Do NOT guess
missing house numbers: flag the row as CHECK instead.
[paste list]

Illustrative output for three messy rows:

Chemist, 12 High St, [town] | 07:30-09:00 | 3 | "before 9, back door"
Unit 4B Parkway Ind Est, [town] | 10:00-16:00 | 12 | signature
CHECK: "Mrs Hale, Orchard Rd (the blue door)" - no house number

The third row is the one that matters. An assistant that invents a house number to make the list look complete sends a driver to the wrong door, so the "do not guess" line is essential.

Same-day jobs that arrive mid-route

For couriers taking same-day work, the ability to add a job at 11:00 and re-optimise the remaining stops is the feature that justifies the software. Check on trial that you can lock stops already done, insert the new job into the best-placed van, and push the updated order to the driver's phone without a call. Customers then need a new arrival time; automatic ETA messages, covered in cutting parcel-tracking calls with AI delivery updates, stop that turning into a round of phone calls.

An illustrative case from the four-van firm: at 11:05 a regular client books an urgent collection from a print works, to be delivered to an office across town by 15:00. Van 2 has 18 stops left and passes within 2 km of the print works at about 12:30; van 4 is closer right now but has two timed drops before 13:00. Re-optimising puts the collection into van 2 at 12:35 and the delivery at 14:10, pushing van 2's last six stops back by about 20 minutes, all still inside their windows. Done by hand, the owner would probably have sent van 4 because it was nearest, and missed one of its timed drops.

The customers whose stops moved then get a short update, sent automatically or drafted from a template:

Your delivery is now due between 15:10 and 15:40 today (previously 14:50-15:20). Reply to this message if that's a problem and we'll call you.

Where optimised routes go wrong on the road

• The van is loaded in the wrong order. The route is perfect and the driver spends five minutes a stop digging for parcels. Print a load list in reverse drop order.
• Drivers quietly ignore the route. Sometimes they're right. Ask why, and put their knowledge into the settings.
• One-off problems become permanent. A road closure that lasted a week can stay as a driver's workaround for months. Review overrides monthly.
• Too many drops squeezed in. Spare capacity is a buffer, not an instruction to fill every minute. Leave slack for traffic and failed deliveries.

The load list is quick to produce from the route export. For a van with 38 drops, it looks like this, loaded from the back of the van forward:

Load first (front of van):  stops 38-30  | 21 parcels
Middle shelves:             stops 29-15  | 44 parcels
Load last (by the doors):   stops 14-1   | 39 parcels
Separate, top shelf:        signature items for stops 3, 7, 22

And when a driver's knowledge beats the route, write it into the settings rather than letting it live in one person's head. An illustrative example: a driver kept doing the industrial estate before the town centre, against the plan. Asked why, he said the estate's gatehouse closes for lunch from 12:00 to 13:00 and parcels can't be left. That became a 09:00-12:00 time window on all eleven estate addresses, the software started planning it that way, and a relief driver covering his holiday got it right on the first day.

Measuring whether the routes are better

Track these weekly for a month before the switch and three months after:

• Kilometres per drop, per van.
• Drops per driver hour.
• On-time rate for timed deliveries.
• Failed deliveries and second attempts.
• Minutes spent planning each morning.

An illustrative before-and-after for the four-van firm, averaged over four weeks each:

Failed attempts barely moved, which is normal: they depend on customers being in and on delivery notes, not on the order of stops. That's a separate problem, and ETA messages tend to help it more than routing does.

If kilometres per drop don't fall within a few weeks, look at the data before blaming the tool: pins, service times and time windows are almost always the reason. For firms with mobile service teams rather than parcels, the same approach is covered in route planning for small delivery fleets and mobile teams, and route planning for landscaping crews shows how it changes when stops take hours rather than minutes.

Courier owners' questions about route planning

Do drivers have to use the route software's app?

To get live tracking, proof of delivery and re-optimised routes, yes. Most tools hand navigation to the driver's usual maps app for the turn-by-turn part, so drivers keep what they know. Introduce the app with one or two willing drivers first, and let them tell the others where it helps.

What if my experienced drivers already know the best routes?

On their regular patch, they may beat the software, because they know the access roads, loading bays and which customers are slow to answer. Run the software for two weeks and compare. Where drivers consistently do better, feed their knowledge in as service times, time windows or notes, rather than overriding the whole route.

Can the software handle collections as well as drops?

Many tools support pickups alongside deliveries, but linked pickup-then-deliver jobs, where the same parcel must be collected before it is dropped, are often limited to higher plans. Check this specifically on a trial if you do same-day collections, because it changes the route far more than simple drops.

Further reads

• AI Job Scheduling and Dispatch for Trades and Field Service Teams — Scheduling and dispatch ideas that apply to courier work.
• How to Read AI Software Pricing: Seats, Credits, and Usage Fees — Compare per-stop, per-order and per-driver pricing properly.
• Outgrowing Spreadsheets: When to Replace Manual Excel With AI — When the planning spreadsheet has had its day.
• How to Pilot AI in Shadow Mode Before Customers See It — Test the software's routes alongside your own first.
• AI Software Contracts: Auto-Renewals, Price Rises, Notice Periods — What to check before signing an annual contract.
• How Florists Can Prepare for Valentine's and Mother's Day With AI — An eight-week countdown for florists: forecast, pre-order menu, stem maths, cut-off messages, card-message rules and delivery routes.
• How Driving Schools Use AI to Keep Instructor Diaries Full — Fill cancelled slots in an hour, replace pupils before they pass, cut dead travel between lessons and spot quiet weeks early, with the data each step needs.
• 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 and Spoke announcement that Circuit is now Spoke; Routific pricing page; OptimoRoute pricing page; Google Maps Help article on adding multiple destinations. Checked September 2026.