On this page
“Should we be using AI?” is now a board-level question in businesses of every size, and the honest answer is almost always: for some things, and not yet for others. An AI readiness assessment exists to draw that line before money is spent on the wrong side of it.
Here is what one actually examines.
1. Which tasks are even candidates
The assessment starts with an inventory of the work, not the technology: what does the team spend time on, and which of it involves reading, categorising, looking things up, or drafting? Tasks that are pure rule-following belong to automation, not AI. Tasks where being confidently wrong would be expensive — pricing, legal commitments, anything customer-visible without review — go on the “not without a human in the path” list from the start.
What survives is usually a short list: internal questions answered repeatedly, documents that need reading and classifying, first drafts that a person currently writes from scratch.
2. Whether your information is ready to be an answer source
An assistant is only as good as the sources it is allowed to use. The assessment looks hard at those sources: Do the documents exist? Are they current, or does the real process live in someone’s head and diverge from the written one? Is there one authoritative version, or three conflicting ones?
This step produces the least glamorous and most valuable finding of the whole exercise. Very often the first project is not an assistant at all — it is getting the handbook, the price list, or the process docs into a state where anything — human or AI — could rely on them.
3. What the assistant must never do
Scope is a design input, not an afterthought. A useful assessment defines, in writing: which data sources are approved, what the assistant refuses to answer, what permissions it runs with, and which actions require a person’s sign-off. An assistant that answers from an approved document set, cites what it used, and declines everything else is dramatically more trustworthy than one that will attempt anything — and trust is what decides whether the team actually uses it.
Scope is a design input, not an afterthought.
4. How you’ll know it’s working
Before rollout, output gets checked against real examples — actual past questions, actual documents — not demo prompts. After rollout, usage and quality get monitored: what is being asked, where the assistant declines, where people override it. Without that loop, quality drifts silently and confidence drifts with it.
What comes out the other end
A short written document: the ranked list of suitable processes, the ones ruled out and why, what needs fixing first, and a right-sized recommendation for the first build — which is sometimes “start with automation instead,” and occasionally “not yet.”
That last answer costs the least and saves the most. The businesses that get burned by AI are rarely the cautious ones; they are the ones that skipped the assessment and pointed a confident system at a problem nobody had defined.