INSIGHT · AI AND TRANSFORMATION

Where to start with AI transformation: begin with a decision inventory

Most AI projects fail not because of technology but because of the wrong starting point. The right start is not choosing a model but choosing the decision you want to improve.

Key takeaways

  • AI transformation starts with a decision inventory, not tool selection.
  • Keep the first pilot narrow: one decision, one team, one measurable win.
  • Write governance alongside the pilot; deferred governance blocks scaling.
  • One piece of evidence in 90 days is the strongest argument for the next budget.

Why not start with the model?

Model and tool decisions age fast. Whatever you choose today will have a better alternative in six months. An organisation's decision structure, by contrast, changes slowly. So start from the layer that lasts — the decision.

Projects that start with a tool also run into an inevitable question: “where will we use this?” Asked at the end rather than the beginning, it usually has no answer.

How to build a decision inventory

Run short sessions with teams and list decisions with three properties: they repeat often, they take time, and errors are costly. Proposal preparation, request triage, technical document search, candidate screening, stock decisions.

Then score each on two axes: potential benefit and implementation difficulty. Difficulty must include data access, privacy sensitivity and the process owner's willingness.

How to choose the first pilot

The first pilot should not be the one promising the largest benefit but the one most likely to produce evidence. The criteria: data is accessible, the process owner is willing, the result is measurable, and mistakes can be reversed.

Keeping scope narrow is discipline, not weakness. A measurable win in one team on one decision is more persuasive internally than ten presentations.

Do not defer governance

Governance is usually deferred — “let it work first, we will write the rules later.” That deferral stops the project at the scaling stage: legal, audit and information security all arrive at once and the pilot is shelved.

The right approach is a three-to-five page framework written alongside the pilot: usage policy, risk classification, human oversight rules and logging. That document becomes ready ground for every subsequent use case.

Tags
WRITTEN BY

Avenra AI and Transformation team

This insight was prepared by the Avenra team running field projects in ai and transformation. Editorial responsibility sits with founder and chief strategist Taylan Yıldız; the material is based on applicable findings from client work.

03FAQ

Frequently asked questions

Our data is scattered — should we do a data project first?
No. Comprehensive data projects take years and AI often never starts. Prepare the minimum data needed for your chosen use case and grow data maturity alongside use cases.
We are a small company — is governance still necessary?
Yes, but lightly. A three-page usage policy, a list of which data may enter which tool, and a human approval rule is a sufficient start for most companies.
Employees already use their own tools — what should we do?
That is shadow usage and banning it does not work. The right approach is to offer a safe official alternative and clarify the rules. A ban does not remove usage; it only makes it invisible.
→Let's begin

Let's turn this insight
into practice.

If what you read is heading towards a decision, let's define an applicable first step.