← All posts

Some of the best automation has no AI in it

· Bob Ottley

AI is useful for reading, interpreting and drafting. It is much less useful when a business needs certainty. Knowing the difference is worth more than any model.

£10,000a summer saved in our worked example, with no AI
2independent readings that must agree on every invoice figure
1sentence explaining every decision the planner makes

A planner with no AI in it

A business shipping frozen food by courier had to decide every day which boxes needed an early delivery slot or extra ice packs. Guess wrong one way and you pay for early delivery you did not need. Guess wrong the other way and a box thaws on a doorstep, and the customer is gone.

Everything that decision depends on is known in advance: the Met Office forecast where the customer lives, how long each box stays frozen with one, two or three ice packs, and when it leaves cold storage. So we wrote the rules down. The planner picks the cheapest option that keeps each box frozen and explains why in one sentence. The team can change the rules themselves.

What that is worth, in round numbers

Say a business sends 400 boxes a week in summer, early delivery costs £3 a box extra and an ice pack costs 40p. Playing safe and sending every box early costs £1,200 a week.

Now say the rules show that, on a typical summer week, a quarter of boxes need the early slot and just under a third need an extra pack instead. That is 100 boxes at £3 and 120 packs at 40p: £348 a week. The saving is about £850 a week, or around £10,000 over a twelve-week summer, with fewer thawed boxes, not more.

An illustration with round numbers, not the customer's own figures.

And because the planner also reads the courier's tracking afterwards, the business can see whether early delivery actually arrived early, by courier service and by area. That is how you find out you are paying for a premium service that is not being delivered.

Read, check, approve

  • Read: let AI deal with messy information such as emails, documents and written requests.
  • Check: use rules and the real system wherever the answer must be exact.
  • Approve: keep a person in charge of anything that affects a customer, a supplier, an account or a record.

Where AI does earn its place

Supplier invoices arrive by email as PDFs in every layout imaginable. Reading them is exactly what AI is good at. Even there, we do not trust one reading. Two separate AI models read every figure, and they must agree before anything is accepted. If they disagree, the invoice is flagged for a person, not guessed. A silent error becomes a visible one, which is the whole point.

Then the rules take over again: the account code is matched to the business's own chart of accounts, the VAT rate to its real rates, the supplier to an existing record so nobody is set up twice, and any change to a supplier's bank details carries a warning before anyone pays. Nothing is posted until a person approves it.

Why this matters to you

A supplier who puts AI in everything is selling AI. A supplier who uses it where it is reliable, and rules where it is not, is selling a system that works on a Monday morning when nobody is watching.

Have a job like this? Tell us in a few lines what gets copied, checked or chased by hand. We will come back with questions, not slides.

Book a call

Next: Before you automate it, ask whether it should exist at all