A Nebis field guide

The AI Operating Playbook

How to put AI to work without putting your compliance at risk. Written from inside a regulated Singapore practice that was rebuilt on it.

Free · No email required
The playbook

Written for people running a business, not for specialists. Read it once end to end, then use the scans and checklists as working tools.

Free · No email required

The AI Operating Playbook

For owner-managers, founders and finance leads at Singapore SMEs. 27 pages.

  • Where AI actually pays in a back office — and where it doesn't
  • Worked examples: receivables chasing, the month-end pack
  • PDPA, record-keeping and the InvoiceNow timetable
  • What a machine should never be the last step before
  • A 30-day pilot roadmap with a decision at the end
Download the playbook PDF · 27pp
What's inside

Not a technology catalogue. It names capabilities, not products — because the tool you pick this year won't be the tool you run in three.

01

Choose the right work

An opportunity scan and a scoring method that stops a long list of interesting ideas becoming a scattered programme.

02

Design it so it can be trusted

The control points, the exception paths, and the difference between a workflow and a demonstration.

03

Stay inside the rules

What Singapore's data protection and record-keeping obligations mean in practice, for work nobody wrote them for.

04

Know where the line is

The decisions that cannot be delegated to a machine, regardless of how capable it becomes.

05

Prove it in 30 days

A week-by-week roadmap where each week ends in evidence and a decision — including permission to stop.

06

Start tomorrow

Checklists, a success contract, and eight starter task specifications you can put to work immediately.

Most guidance about AI is written by people who advise on AI. This is written from the other side of the problem — under a statutory framework that punishes carelessness, in a jurisdiction that expects records to be produced on demand years after the fact.

That forces questions a pilot project never has to answer. What happens when the output is wrong and someone has already relied on it? Who is accountable when the machine did the work? Which decisions can never be delegated, no matter how good the technology gets?