AI consulting in Newcastle, NSW: where AI actually pays.

It is looking at how a business really works, finding the jobs that are repetitive and written down, and deciding which of them are worth automating and in what order.

The work itself is plain enough. Somebody sits with the people doing the job, writes down what actually happens, and looks for the parts that happen the same way every time.

What it covers

• How the business actually works.

  What people really do day to day, rather than the written process, including the spreadsheets holding it together.

• Which of the jobs repeat.

  The ones with the same steps every time are the candidates. The ones needing a judgement call are not.

• What each one would take.

  For every candidate: what has to exist first, what it touches, and what happens when it gets something wrong.

• The order to do them in.

  Which one comes first, and which ones are not worth doing at all. This is the part that earns the fee.

What comes out is a short document naming what to do, in order, with the ones to leave alone named as well. Building any of it is a separate job, and it is the easier half.

The tools are cheap and plentiful. What is scarce is knowing which job in your particular business is worth handing to one, and that answer does not come out of a product demonstration.

There is a lot of software with AI in the name at the moment. Most of it is decent software. The question worth answering first is which job in your business it is for.

What follows is a familiar pattern. A subscription that goes unused a month later, because it never fitted how the work happens.

The deciding is what stops that, and it costs less than the tools do.

It will not make a judgement call you would want to stand behind, it will not fix a process nobody has written down, and it will not know anything about your business you have not told it.

Being clear about the ceiling is the part of this conversation people skip, and it is what makes everything else usable.

The limits worth knowing

• It has no judgement of its own.

  It will produce a confident answer to a question that needed somebody to think. That is the failure mode, and it is why the limits go down on paper first.

• It cannot fix an undefined process.

  Where three people do the same job three different ways, there is nothing consistent to hand over. The writing down is the actual work.

• It knows only what it was given.

  Your prices, your area, your rules. Nothing beyond that, and it should say so rather than guess at the rest.

• It does not remove the responsibility.

  Where it sends something wrong to a customer, that is your business speaking, not a supplier's.

The federal government's AI ethics principles put accountability and transparency at the centre for the same reason. They are short enough to read in one sitting.

The job has to be written down, the facts it needs have to exist somewhere, and somebody has to own reading what it does. Missing any of the three, building is early.

Almost every automation that quietly fails was built before one of these was true. Each of them belongs to the business rather than to whoever builds it.

What has to exist first

• The job itself, written down.

  The steps in order, as they actually happen. Where it cannot be written down, it cannot be handed over.

• The facts it answers from.

  Prices, hours, area, rules. Written once, and kept somewhere that can be changed the same day.

• The limits on what it may say.

  What it must never say or decide, written before it goes live rather than after a complaint arrives.

• Somebody who reads what it did.

  Every message, for the first while. That is what keeps it improving rather than drifting.

Somebody sits with the work. The repeated jobs go on a list and carry an honest cost, one comes off it first, its facts and its limits go down on paper, and the smallest version of it goes live and is read.

Every step produces something on paper, which is why the decision at the end takes minutes rather than meetings.

1. Somebody sits with the work

With the people who do it, rather than with the owner's description of it. The gap between those two is where most of the findings turn up.

2. List the repeated jobs

Everything that happens the same way every time. The small ones go on the list too, because how often a job comes up counts as much as how big it is.

3. Put an honest cost on each one

How often, how long, and what happens when somebody misses it. That is what turns a list into an order of work.

4. Pick one

The one with the best ratio of frequency to difficulty. One job is set up and left running before the next one starts.

5. Write its facts and its limits

What it answers from, and what it must never say. About an hour, and it decides whether customers like the result.

6. Build the smallest version, and read it

Then somebody reads everything it does for a fortnight. Then, and only then, the next job starts.

The looking is short. Writing down how a job actually happens takes longer than anybody expects, and it is the part deciding whether the building works at all.

The order it happens in

• First somebody looks at the work.

  With the people doing it. It is a short piece of work, and it is what produces the list.

• Then one job goes down on paper properly.

  Steps, facts and limits. This is the part that stretches, and it belongs to the business.

• Then the smallest version goes live, and somebody reads it.

  Then corrected. Then the next job, which is always faster than the first one was.

A business wanting something no tool currently does is asking for software to be built. That is a different job with a different timeline, and it is worth naming as one rather than discovering it.

Ask what they look at before recommending anything, whether they sell the tools they recommend, who owns what comes out of it, and what they would tell you not to do.

These are worth asking of any consultant, including this one, so the answers given here are underneath each question.

• What will you look at first?

  How the business actually works. An answer starting with a product has skipped the job entirely.

• Do you sell the tools you recommend?

  It is worth knowing either way. Nothing recommended here is owned here, and sometimes nothing is needed.

• Who owns what comes out of it?

  The business does, in accounts in its own name, including anything holding customer details.

• What would you tell me not to do?

  A consultant with nothing at all on that list has not looked properly at the business.

Plenty of businesses go through this and find one small thing worth doing. That is a good outcome and it is worth saying, because the alternative is a project nobody needed.