How to Audit Your Business for AI Opportunities: The 4-Phase Method We Use

We run software products of our own alongside client work, and several times a year we put each one through a full audit: inventory every feature the product claims to have, test whether each one actually works, fix what's broken in priority order, then retest to prove the fixes hold. It is unglamorous work, and it finds problems every single time — including in products we built ourselves and thought we knew.

At some point we realized the same four phases are exactly how a business should decide where AI fits. Not by asking "what can AI do?" — that question produces brainstorms and decks. By asking "what does this business actually do all day, and which parts of it are broken or expensive?" That question produces a ranked list you can act on. Here is the method in full. You can run it yourself with a spreadsheet and a couple of honest afternoons.

Phase 1: Inventory

List every recurring task in the business. Not departments, not job titles — tasks. "Answer inbound phone calls." "Prepare the Tuesday production schedule." "Chase unpaid invoices." "Re-key orders from email into the system." Go person by person and ask each one: what do you do every day, every week, every month? Write down everything, including the things that seem too small to matter.

Two rules make this phase work. First, capture frequency and time: how often does this happen, and how long does it take each time? Multiply those and you get hours per month, which is the number everything else hangs on. Second, capture the tasks people do that aren't in anyone's job description. Every business runs on invisible glue work — the office manager who reconciles two systems by hand, the estimator who retypes the same specs into three documents. That glue work is usually where the biggest automation wins hide, precisely because nobody ever examined it.

Phase 2: Test

Now score every task on the list against four questions. Is it frequent? Something done fifty times a month justifies automation that something done twice a year never will. Is it low-variation? Does the task follow roughly the same shape every time, or does every instance require fresh judgment? Is the input digital or easily made digital? A task that starts from an email or a form automates far more easily than one that starts from a phone call and a sticky note. And is the output checkable? Can you tell, quickly and objectively, whether the task was done right?

Score each question 1 to 3, add them up, and sort. This takes an afternoon and it is the single highest-value afternoon in the whole process, because it replaces "I feel like we should automate something" with a defensible ranking. The top of the list is almost never what the owner guessed. Owners guess the tasks that annoy them personally. The scoring finds the tasks that quietly consume the most hours with the least judgment.

"High frequency, low variation, digital input, checkable output. Every reliable AI deployment we've ever shipped scores well on all four. Every failed one was weak on at least two."

Phase 3: Fix

Take the top three tasks — not the top ten — and automate them properly. Properly means: the system connects to your real tools, a human owns review of anything customer-facing until trust is earned, every action is logged, and there's a defined escalation path for the cases the system can't handle. It also means measuring a baseline before you start. If invoice follow-up takes six hours a week today, write that down, because in ninety days someone will ask whether the system was worth it and "it feels better" is not an answer.

Resist the temptation to start all ten at once. Three systems that ship and get adopted beat ten that stall in parallel, and the lessons from the first three make the next three cheaper and better. This is the difference between businesses that compound their automation and businesses that have a graveyard of half-finished tools.

Phase 4: Verify

Thirty days after each system goes live, retest it the way you'd test a feature: is it still running? Is anyone silently working around it? Did the hours actually come back, per the baseline you wrote down? Is the error rate acceptable, and are errors getting caught by the escalation path or leaking to customers?

This phase is the one everyone skips, and skipping it is how companies end up "using AI" on paper while the staff quietly reverted to the old way. In software, an untested fix is assumed broken. Treat automations the same way. The verification pass takes a couple of hours per system, and when a system passes it, you have something rare: an automation you can trust enough to stop thinking about. Then you go back to the ranked list and pick the next three.

Run it yourself, or have us run it

Everything above is doable in-house with a spreadsheet and some discipline, and if you run it yourself, you'll get value from it. The place most businesses want help is the scoring and the builds — knowing what current AI can actually do reliably (Phase 2) and building systems with proper guardrails (Phase 3) are where experience changes the outcome.

Our free assessment is essentially Phases 1 and 2 done for you from the outside: we research your business, apply the scoring, and hand you a brief with the three to five highest-leverage opportunities and honest impact estimates. If the list is worth acting on, we can build. If it isn't, you'll have a clean audit and a spreadsheet, and that's worth having either way.