A 1–2 week, fixed-price engagement ($2,500–$5,000 at Cerebrum) that maps how work actually moves through your business, scores every automation candidate on hours, feasibility and risk, and ends with a written roadmap: usually 6–10 candidates, 2–3 recommended for a first build, each with hours saved, cost and payback.
An AI operations audit is the work you do before writing a line of code. At Cerebrum it is a 1–2 week, fixed-price ($2,500–$5,000) engagement: a kickoff, two to three working sessions with the people who actually do the work, a workflow map with real volumes and times, a scored list of automation candidates, and a written roadmap you own whether or not you continue with us. A typical audit surfaces six to ten candidates and recommends two or three for a first build. Here is what is inside, stage by stage.
| Stage | What happens | Time | What you get |
|---|---|---|---|
| 1. Kickoff | Owner / ops lead: what you sell, how money and work move, systems in use, what’s been tried | 60–90 min | Scope + list of workflows to follow |
| 2. Working sessions | 2–5 team members show (not tell) their real work: inbox, systems, yesterday’s tasks | 45–60 min each, 2–3 sessions | Observed steps, handoffs, volumes, time per instance, error points |
| 3. Workflow map | We document each candidate workflow end to end, including the spreadsheet that shouldn’t exist | Our time | Map with weekly volume, minutes per instance, systems touched |
| 4. Scoring | Every candidate scored on cost today, feasibility, risk of a wrong answer, and payback | Our time | Ranked list, usually 6–10 candidates |
| 5. Roadmap + readout | Build / buy / wait call per candidate; 2–3 recommended for Phase 1 with hours saved, cost, payback | 60 min readout | Written roadmap you keep and can hand to anyone |
| Total | 1–2 weeks · $2,500–$5,000 · 2–3 hours of the owner’s time | cerebrumone.ai/services |
The failure statistics are not about bad models. RAND interviewed 65 practitioners in 2024 and found that misunderstandings about what problem the project is meant to solve — named by 84% of them — cause more AI failures than any other factor. MIT’s 2025 review of more than 300 initiatives reported that 95% of organisations see no measurable P&L return from generative-AI pilots. S&P Global found 42% of companies abandoned most of their AI initiatives in 2025, up from 17% a year earlier, and that the average organisation scrapped 46% of proofs of concept before production. Every one of those numbers is a project that started with a tool and worked backwards to a problem. An audit reverses the order.
After the kickoff we sit with the people who do the work and ask them to show us yesterday: open the inbox, open the system, walk through a real order, claim, file or candidate. Process documents describe how work is supposed to flow; the gap between that and reality is where the opportunity lives. The gap is large. Asana’s 2023 survey of 9,615 knowledge workers found 58% of the workday goes to “work about work” — coordination rather than skilled work — and workers estimated 4.9 hours a week could be saved with better processes. Zapier’s 2021 survey of 2,000 SMB employees found 94% perform repetitive, time-consuming tasks. Nobody puts those hours in a process document, so we go and count them.
We are looking for specific things: re-keying (information that exists in one system being typed into another); “checking on” (status requests from clients, tenants, carriers or borrowers that pull someone off real work); read-to-extract (opening a document to pull five fields out of it); templated replies with slight variation; queues that grow overnight; and the spreadsheet that shouldn’t exist, which is nearly always patching a gap between two systems. McKinsey Global Institute’s long-standing estimate is that about 60% of occupations have at least 30% of activities that are technically automatable — but fewer than 5% can be automated entirely, which is exactly why the human review step is designed in from the start rather than bolted on.
Cost today: hours per week times loaded hourly cost times 48. Feasibility: does the input already exist in a system, is there an export or API, what share of cases need judgment. Risk: what happens when the output is wrong — an internal correction, an apology, or a regulator. Payback: months to recover the estimated build cost. Candidates that score well on all four go to the top; the ones that score badly on risk get a mandatory human sign-off in the design or are parked. This is where “an AI that knows the whole business” usually dies — high excitement, no volume, no measurable payback — and where the boring workflow that eats 12 hours a week rises to the top. BCG’s research on 1,000 executives found that AI leaders pursue about half as many opportunities as their peers and concentrate on the ones that matter; the scoring is how we do that on purpose.
You leave with a written document: the workflow map, the ranked candidates, and for each recommendation what the system would do, which system it lives in, who reviews its output, expected hours saved, estimated build cost and payback period — plus a build, buy or wait call. “Buy” means the software you already run has the feature switched off. “Wait” means the process is broken, the data lives in someone’s head, or the volume doesn’t justify it yet; in a meaningful share of audits the first recommendation is a free process fix and the AI build comes second. You can hand the roadmap to your own team, to another vendor, or to us. Most clients go on to a Phase 2 build; some don’t need to yet, and we say so.
Two to three hours of the owner’s or ops lead’s time across the engagement. 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 and no homework. If the team is remote, sessions run over screen-share and are, if anything, easier, because people show their actual screens. McKinsey’s 2025 survey found AI high performers are nearly three times as likely to have fundamentally redesigned individual workflows; that redesign starts with someone seeing the workflow as it really is, and that is what the sessions are for.
We charge $2,500–$5,000 rather than folding it into a build for a simple reason: you are paying for the answer, not for the build, so we have no incentive to inflate what we find. Gartner’s prediction that at least 30% of GenAI projects would be abandoned after proof of concept listed unclear business value first among the causes; the audit is designed to make the value explicit — in hours and dollars — before anything is built. If we come back and say don’t build anything yet, you have still bought the map, and the map is what stops the next vendor selling you the wrong thing.
One to two weeks from kickoff to readout, including two to three working sessions of 45–60 minutes with your team.
$2,500–$5,000 depending on the size and complexity of the business. It is a fixed price and stands alone; you are not committing to a build.
No decks and no homework. A list of your systems, a few real examples of the work, and access to the people who do it.
It happens, and we say so. You still leave with the workflow map and a ranked list — usually with a process fix or a feature in your existing software that pays off at zero build cost.
Yes. It is not a slide deck about AI. It is an observed, measured map of your operations with a scored, priced list of what to do first.
These take a while to write. One email when the next goes up, nothing else.
Book a free 30-minute operations call. No pitch, just an honest conversation about where AI can help.