Build-over inventory
Five layers of ground under every AI tool; this page asks one question per layer.
You answer the five questions about your own company. Run it in one sitting with the people who know, and write the answers down: what you finish with is the list of what you have not prepared yet. If your people are already building tools, the failure-mode guide covers what happens as they grow.
The five questions
Where your data actually lives
- The question
- Which system is production, which is a copy, and which logins can reach each one?
- Why it matters
- The tools never touch production, and that rule starts with knowing which system is which.
- What an answer looks like
- One line per system: what it is, where it runs, who reaches it, read-only or not.
- What the gap costs later
- A report shows a number nobody recognizes, and nobody can say which copy produced it.
- The next question
- Which copy do the tools read, and who verified it is read-only?
Whether your definitions agree
- The question
- What does gross margin mean, which tables produce it, and who owns it?
- Why it matters
- Every report, dashboard, and model query inherits them.
- What an answer looks like
- One page per term: the meaning, the tables and columns, and a named owner.
- What the gap costs later
- Two dashboards show two numbers for the same metric, and nobody can say which is right.
- The next question
- When two reports disagree, whose definition wins?
What the agents are allowed to touch
- The question
- What can a tool touch, through what gate, and where is the record?
- Why it matters
- Every agent runs on the access its builder already had.
- What an answer looks like
- Written rules of engagement: which systems, what path, what gets logged, who owns each connection.
- What the gap costs later
- Forty-five agents query in thirty-seven ways through personal keys, and none of it is logged.
- The next question
- If a tool misbehaved yesterday, could anyone show what it touched?
What the terms actually say
- The question
- Which AI tools are you paying for, and what do the terms do with your data?
- Why it matters
- Where nobody has decided, whoever clicked accept did.
- What an answer looks like
- One page: tool, tier, what the terms say, who decided. Counsel can sign it.
- What the gap costs later
- Nobody can produce the list of tools when counsel asks.
- The next question
- If counsel vetoed one tool tomorrow, which workflow stops?
Where it runs
- The question
- Which cloud, whose accounts, and what exists between a laptop and production?
- Why it matters
- A laptop prototype has to move somewhere real before the company relies on it.
- What an answer looks like
- A written decision: AWS or Azure, shared Python environments, a replicated reporting database, QA separated.
- What the gap costs later
- Credentials sit in laptop config files, with nowhere to move a tool that outgrows them.
- The next question
- Who gets the 2 a.m. call when the environment breaks?
Reading your answers
A gap is a sequencing decision: it tells you what to prepare first, before the work that depends on it.
Each item needs an owner and a written answer; an answer that lives in one person's memory of a meeting is still open.
Where this comes from
These questions come from workshops we have run and from conversations with buyers through 2026. We add a question when the field shows us one we missed. Last reviewed September 2026.
Sequence it with us
Bring your list to a Working Session and we will sequence it with you: what to prepare, in what order. The verdict scopes the Sprint, if there is one worth doing.
Book a Working SessionThe Advisory page describes how a Working Session works.