Audit the roles, not the tools

There is a booming market right now in AI audits. Readiness assessments, maturity scorecards, stack reviews. Most of them are competent, many of them fixed-fee, nearly all of them built around the same object: the tool.

They inventory your software. They score your data. They map your workflows against a framework. And they hand you a roadmap of tools to add.

Here is what almost none of them examine: what your people actually do all day.

AI does not replace tools. It changes tasks. And tasks live inside roles. If you do not know which tasks make up which roles in your business, every AI purchase is a guess wearing a scorecard.

What a role actually is (and why job titles lie)

Ask an owner what their office manager does and you will get the job description. Sit with that office manager for two days and you will find the real role: forty distinct recurring tasks, maybe a dozen of which appear anywhere in writing. Some are pure repetition: reformatting the same report, chasing the same signatures. Some are pure judgment: deciding which unhappy customer gets a call from the owner. Most are a braid of both.

Three questions for every task

The Role Audit scores every task in every role on three axes:

• How much judgment does it require? Not "could AI do this in a demo." How much context, taste, and consequence-awareness does the task carry in this business?
• How often does it repeat? AI compounds on repetition. A task done once a quarter rarely justifies any tooling.
• What happens when it goes wrong? A misformatted internal report costs a shrug. A wrong number on a client invoice costs a relationship. Error consequence, not technical capability, should decide where AI operates unsupervised.

Score every task on those three axes and something clarifying happens: the AI opportunities in a business stop being a matter of opinion. High repetition, low judgment, low consequence: automate it. High judgment, high consequence: keep the human, maybe give them an assistant. The map draws itself.

The finding nobody selling software will give you

Run this exercise honestly and the first result, in most businesses, is subtraction. Tools bought for roles as imagined, not roles as performed. Two platforms doing the job of one. Seats billed for people who left a workflow months ago. The typical founder-led business is carrying 20–40% overlap or underuse in its software spend, which means the honest first move in most AI strategies is to cut the tool bill before adding anything.

A vendor cannot tell you that. Their assessment is scoped to find a reason to buy their product, and it will, every time. That is not corruption; it is just incentives. Which is exactly why the diagnosis has to come from someone with no software to sell and no referral fee waiting on the recommendation.

Roles are also where the fear lives

There is one more reason to audit roles instead of tools: your people. Every AI conversation in a small business carries an unspoken question: is this about replacing me? A tool audit cannot answer that, because it never looked at the person. A Role Audit answers it explicitly, task by task: here is what AI takes off your plate, here is what stays yours, here is why the judgment you carry is the part that was never automatable. That specificity is the difference between a team that adopts AI and a team that quietly resists it.