Adam Yee Workflow diagnosis · Singapore

Singapore · AI implementation

The tools arrived.
The systems didn't.

Capability is cheap now. Almost nobody has installed it.

You can change how your company runs in weeks. A company ten times your size cannot.

01

None of this looks like an AI problem.

Count the ones you recognise.

  • The same fact, typed into three systems.
  • Two systems disagree. No tie‑breaker.
  • A thirty-second question takes two days.
  • One person is the system. They took leave.
  • Nobody owns the first reply.
  • The mistake surfaces at month end.

All six are cheap to fix now.

02

The tools are easy. The order is not.

First

The facts have to be true in one place

One version of a customer, a job, a price.

Then

The work has to have a shape

A state, an owner, a definition of finished.

Only then

Intelligence, pointed at one thing

Cheap now. Worthless without the first two.

Almost everyone starts at the third one.

03

Same material. Same volume.

One body

Every change waits for the whole thing to come round.

In parts

Already where you pointed.

One of these is you.

04

Three processes, taken apart

Archetypes, not clients. Open one.

05

The questions people actually ask

Data, lock-in, what happens if you leave, and the rest
Where does our data go?
Wherever you decide, and I will put it in writing before anything is built. Plenty of these systems run entirely inside infrastructure you already own and pay for. Where something does leave — a document-reading service, for instance — I will name exactly which service, exactly what is sent, and what it retains, before you agree to it. If a class of document cannot leave your building, that is a design constraint, not a blocker.
Our staff will think this is about replacing them.
Sometimes they are right to be suspicious, so it is worth being straight about it. What I build removes copying, chasing and status-checking — the parts of the job nobody defends. If your actual plan is a headcount cut, tell me, because it changes the design and I would rather know than find out. In my experience the fastest way to sink one of these projects is to surprise the people who do the work.
We tried automating this and it did not stick. Why is this different?
Usually it did not stick for one of three reasons: it could not handle exceptions so people quietly went back to the old way; it needed the input to change and the input did not change; or one person understood it and they left. I would want to hear which of those it was before proposing anything — the previous attempt is the most useful information you have, and I would not treat it as a blank slate.
Do we get locked in?
You own the code and the documentation, it runs on your accounts, and another developer can pick it up. Any third-party service in the design is named up front with what replacing it would cost. If I disappeared, the thing keeps running.
What if it breaks at 2am?
Anything I build fails visibly rather than silently — the point being that a broken automation nobody notices is far worse than one that stops and says so. Every process keeps a manual path, so a failure slows you down instead of stopping you. Support after handover is a separate conversation and a separate price, and it is optional.
Do you actually build it, or subcontract?
I build it. You are dealing with the person doing the work, which is the main practical difference between me and an agency, in both directions — there is no team to absorb it if I am busy.
Can you work with our existing systems?
Usually. If it has an API, yes. If it does not, there are still options and they are uglier, so I will tell you honestly whether working around it is worth it or whether the real answer is to leave that system alone and build beside it.
How long does one process take?
I will not give you a number before understanding the process, because one invented on a first call is a number I would then be managing rather than the work. You get a written timeline with the fixed quote, before anything starts.

06

Bring me one stubborn process

Describe it in your own words, however messy. I will tell you what I would build first, what I would leave alone, and whether it is worth building at all.

A reply in writing, with an actual opinion. Not a calendar link.

I will say no if

  • It runs a few times a week. Under about twenty cases, a careful person is cheaper.
  • Nobody can describe how it works today. Then the first job is mapping, and mapping cannot be skipped.
  • You want AI because the board asked. I have nothing that survives that brief.

I would rather lose the enquiry than the first month.