
Systems replaced
Target audit trail
Platform Build
Operations Platform
The internal tool a team has been running out of spreadsheets, rebuilt as one system with roles, an audit trail, and a number the business is willing to act on.
Timeline
14 wks
Industry
Multi-Team Operations
Service used
System Design, Platform Build, Data Work
Challenge
Four spreadsheets and a shared inbox are load-bearing, and nobody agrees which one is correct
(nai™ — the problem)
Four spreadsheets and a shared inbox quietly became load-bearing, so the number the business plans against depends on which copy someone opened and who last remembered to update it.
Platform Walkthrough
0:10 sec film
(nai™ — solution)
From four spreadsheets to one system with an audit trail
Spreadsheets do not fail loudly. A file gets copied, a column is added to one copy, and a year later the number the business plans against depends on which version someone opened and who last remembered to update it.
The replacement has to be duller than the sheet, and harder to get wrong.
What we would ship is a Next.js application over a Postgres schema that models the work rather than the sheet: records with an owner, the states a record may be in, and transitions that write who changed what and when. Roles decide what each person sees, and nothing is edited without a history.
The parts that make people stop keeping their own copy are the unglamorous ones:
one record per thing, with an owner named on it
states and transitions the system enforces
an audit trail that cannot be edited afterwards
exports, so nobody has to keep a private copy
Migration is the risk, not the build. The existing sheets are read as they are, including the rows that break their own rules, and every exception is either modelled or written down as a decision before anything is loaded.
Adoption follows the same logic. The system runs alongside the sheets rather than instead of them, the team keeps working while the two agree, and a sheet is retired only once the platform is the copy people reach for first without being asked.
Reporting then stops being a weekly assembly job. Because the states are real and the history is complete, the numbers are queries against Postgres rather than a formula somebody maintains in the corner of a tab.
The measure is not the feature list. It is whether, three months after launch, anybody is still keeping a private version of the truth in a file on their own machine.

(nai™ — Technology Stacks)
(nai™ — our commitment)
The measure of an internal tool is whether people stop keeping their own copy. That is the thing we build for, and the thing we check afterwards.
Nukes AI
Delivery commitment
(nai™ — 05)
MORE BUILDS

(nai™ — 15)
Work with us
Book a
free call
We build AI that changes how your work runs —
not how the demo looks.
We’ll map how your
workflows run, show
where AI pays off,
and leave you
with a clear plan.
No prep needed — we’ll steer the conversation
and keep it on what matters.




4 practices
7+ yrs
minimum, every engineer





