The AI audit
Two weeks inside your operation. Out the other side, a ranked plan: what to automate, in what order, and what each piece is worth. Before anyone builds anything. The AI audit is where every Nezam engagement starts.
Two weeks, fixed fee. Ten to thirty hours of a senior engineer. Yours to keep, whoever builds it.
The two weeks, day by day
Week one, watch the work.
- Day 1, kickoff. An hour with your team. Who does what, which tools are load bearing, where the phone rings. Then we get out of the way.
- Day 2, watch. A day beside whoever runs the busiest desk. We time things. We do not ask how long things take, because nobody knows.
- Day 3, watch. The busiest workflow followed end to end: where it starts, every hand it passes through, every place it gets typed in again.
- Day 4, watch. The tools. What has a login, what has an export, and what exists only in one person's head.
- Day 5, the long list. Everything that came up, in the order it came up. Nothing ranked yet.
Week two, rank what is worth building.
- Day 6, size. Each item sized from your own figures: the call log, the calendar, the payroll. Never an industry average.
- Day 7, size. What each one would take to build, and what could go wrong once it is running.
- Day 8, check. An hour with you on the numbers. A figure that is wrong gets caught here, not in the plan.
- Day 9, the cut. Anything that does not pay for itself comes off the list. Anything that needs judgment stays with people.
- Day 10, the session. The ranked plan, and you argue with the order. The document is yours when you leave.
Your team is in the room on three of those days. The rest is us, on our own.
What you walk away with
Four parts, one document, yours whoever builds from it.
- The operations map. Your busiest workflows as they run, not as the org chart says they run. The org chart says a call comes in, the coordinator books it, the job gets done, the invoice goes out. As it runs: a call at 4:50 PM on a Friday goes to voicemail and is heard on Monday, gets typed into the CRM by the coordinator after two callbacks, typed again into the scheduler with the unit number wrong, and invoiced from a spreadsheet by the owner at month end.
- The opportunity list. Every automation worth building, ranked by what it gives back against what it takes to build. The thing that came up first in the meeting is rarely the thing worth building first: the chatbot the owner wanted demos well and returns little, intake on the main line is usually the build to start with, and pricing and discounts stay with people.
- The numbers behind each one. What each piece is worth in hours and dollars, worked from your figures and never from an industry benchmark. Calls on the main line and the hour they arrive come from your call log. Hours spent typing one record into three systems are counted beside the desk. Quotes that went quiet come from your CRM. What an hour of the busiest person costs comes from your payroll. Every figure in the plan has one of these under it, so you can check the arithmetic before you agree with the order.
- The build order. What to do first, what to wait on, and what stays with people. The plan says so in writing, so nobody builds it later by accident.
Five questions, in order
Everyone wants to start at the fifth question. Starting there is how AI projects waste their first month.
- Where is the time going? Not what your team does. That answer is always a role description. We want something closer to a calendar: how many hours last week went to renewals, to triage, to the escalation that has been bouncing around the office for a month. You get it by sitting beside the team and counting.
- Where is the money going? Whose hours are worth the most, and what those hours are spent on. The owner assembling a report a coordinator could assemble. Nobody invoices for it, so it never shows up as a cost, and it is usually the biggest number in the audit.
- What is fragile? Work that runs only because one person remembers how. Build on it before it is written down and you ship something that looks right for a month, then breaks on the first case that only ever lived in someone's head. The fragile list says what to write down first, and what not to build first.
- What is repeatable? Consistent inputs, stable rules, a bounded set of ways to fail. The maintenance ticket flow usually is. The owner who emails at 11pm with something that needs judgment is not, and should not be a build target. The repeatable list is where the build will live.
- What is the next dollar worth automating? The one everybody wants to start with, and the only one that cannot be answered until the other four are. Ranked by return on the constrained person's time, not by how well it demos. For almost every operation we have looked at, the answer is boring: a database, a clear schema, structured notifications, a few carefully written prompts.
Why we lead with an audit
You should never buy a build blind. We should never sell one.
Most AI projects fail before they start, because they were picked in a meeting instead of measured on the floor. If the honest answer is that there is nothing worth building, the audit says so, and you keep the plan either way.
Here is what measuring looks like. The answering service industry sells the after-hours window on volume: up to 65 percent of property management calls arrive outside business hours, a figure that traces back to a vendor page with no source under it. We hold a complete log for one Toronto property management line, every call answered and time-stamped for a quarter, so we counted.
12.7 percent of its calls arrived outside a nine to six week. Small. Those calls held 41 percent of the quarter's tagged emergencies. One after-hours call in four carried the emergency tag; in business hours, one in twenty. The case for covering the window turned out to be about risk rather than volume, and the build order changed because of it.
Source: property management line at a Toronto property management and brokerage operation, 1 June to 10 September 2026, 102 days, 2,142 calls, 158 of them carrying the operation's own emergency routing tag. After hours is outside Monday to Friday, nine to six. Client anonymized. The full count, with the method.
Questions, answered straight
How long does an AI audit take? Two weeks, start to delivered plan. We set the start date on the discovery call so it lands when your team can give it a few hours of attention.
What does an AI audit cost? A fixed fee, set on the discovery call once we know the size of the operation. What sets it is the work under it: ten to thirty hours of a senior engineer's time, and you should be able to divide any quote by the hours before you agree to it. Every other price we charge is on the pricing page. If we do not believe the audit will pay for itself, we say so on the call instead of selling it to you.
Do we have to build with you afterwards? No. The plan is written so your team, another firm, or Nezam can build from it. It is yours to keep either way.
What do you need from us? A few hours across the two weeks: an hour at the start, an hour on the numbers, and the working session at the end. In between, we sit beside the people doing the work while they do it, and we look at the tools you already run.
Is the audit a report or a slide deck? Neither. It is a working plan: a ranked list with the reasoning shown, delivered in a session where you can push back on it.
Ask the rest on the call. Thirty minutes, free, and if the audit is not worth your money we say so.