The operations audit: how we find where AI fits

Most AI vendors open with a demo. We open with your operations. The tool is the easy part — knowing what to point it at is the whole game, and you can’t know that from the outside. So the first thing we do with every client is an operations audit. This is what it actually involves.

What the audit is (and isn’t)

It’s one to two weeks, two or three working sessions, and it ends with a prioritised roadmap: every AI opportunity we found in your business, what each is worth in hours and dollars, roughly what it would cost to build, and an honest buy-build-or-wait call on each. It’s a fixed deliverable that you own, whether or not you continue with us.

It is not a strategy deck, and it is not a sales process in disguise. Sometimes the answer is “don’t build anything yet, fix this handoff first.” We say that when it’s true.

Week one: follow the work

We start with a 60–90 minute kickoff with the owner or operations lead. What does the business sell, how does money and work move through it, what’s the software stack, what’s already been tried, and what’s the thing you keep meaning to fix. That gives us the map at altitude.

Then we get to ground level. We sit with the people who do the work — usually two to five people, 45–60 minutes each — and we ask them to show, not tell. Open the inbox. Open the system. Walk us through yesterday. While they do, we’re writing down the steps, the handoffs, the systems touched, how many times a week this happens, how long each one takes, where errors and rework show up, and where things sit waiting for someone.

This is the part that can’t be skipped. Every business believes its process works the way the process document says. It never quite does, and the gap is where the opportunity lives.

What we’re looking for

Certain patterns light up almost immediately:

• Re-keying. Information that exists in one system being typed into another.
• “Checking on.” Status requests — from clients, tenants, carriers, borrowers — that pull someone off real work to look something up.
• Read-to-extract. Someone opening a document to pull five fields out of it.
• Templated replies with slight variation. The same email written thirty ways a week.
• Queues that grow overnight. Anything that piles up while nobody’s watching.
• The spreadsheet that shouldn’t exist. Every business has one. It’s usually patching a gap between two systems.

We’re equally alert to the anti-patterns: processes nobody on the team agrees on, data that lives only in someone’s head, and decisions with regulatory or legal weight where a human step must remain. Those don’t get automated. They get flagged, and if it’s a judgment call, we design the human review in from the start.

Week two: scoring and the roadmap

For every candidate we found, we score four things. What it costs today (hours per week times loaded cost). How feasible it is (does the data exist, does the system have an export or API, how much of it is judgment). What the risk of a wrong answer is (an internal correction, an apology, or a regulator). And how fast it would pay back.

A typical audit surfaces six to ten candidates. Two or three get recommended for a first build. The rest are parked with reasons — “worth it after volume doubles,” “needs six months of clean data first,” “your software already does this, it’s just switched off.” Each recommendation spells out what the system would do, where it lives, who reviews the output, expected hours saved, estimated build cost, and payback period.

A composite example

A property management company with around 900 doors. Maintenance requests arrive by phone, email, and portal, and a coordinator triages them by hand — roughly 25 hours a week. Owner reports are assembled manually each month, about 40 hours. Leasing follow-up is inconsistent because it depends on who’s busy. The audit ranked triage first: high volume, mostly rules, output reviewable, lives entirely inside their existing property-management software. Owner reporting came second, leasing follow-up third. The first build was scoped at three weeks and paid back inside a quarter. Nothing exotic — just the right order.

What you need to bring

Two to three hours of your own time across the engagement. Access to three to five team members for about an hour each. A list of the systems you run, and where practical, read-only access or an export. A handful of real examples — last week’s emails, a recent order, a live file. No prep decks, no homework. The whole point is that we come to you.

What it costs, and why we charge for it

Audits run $2,500–$5,000 depending on the size and complexity of the business. Charging for it keeps everyone honest: you’re paying for the answer, not for a build, so we have no incentive to inflate what we find. And you keep the roadmap either way — hand it to your own team, another vendor, or us. Most clients go on to a Phase 2 build; some don’t need to yet, and we tell them so.

If you’re not sure which problem to solve first, that’s normal — it’s the single most common reason AI projects fail, and it’s exactly what the audit is for.