Ops audit and automation roadmap

What this actually solves

There is a particular point a growing company reaches where the owner knows something is wrong but cannot name it. Headcount has gone up faster than output. Everyone has invented their own way of doing the same job. Three people maintain three spreadsheets that mostly agree. Nothing is on fire, and yet Friday afternoon still gets eaten by work that should have taken an hour. The question is not whether automation would help — it obviously would. The question is where to start, and that is the question nobody inside the business can answer, because everyone can see their own corner and no one can see the whole thing.

There is usually a second reason too, and it is often the real one. The owner has to persuade somebody — a partner, a board, a finance director, sometimes just themselves — that spending money on this is sensible. A conversation cannot do that. A document with task-level hours, a cost against each phase and an explicit list of what will be left alone can. This engagement produces that document, and it is written to be read by someone who was not in any of the interviews.

What it is not is a build. Nothing gets connected, no tools get bought, no account gets created. It is looking and telling. That distinction matters commercially as well as practically — at the end of it you have a plan you own outright, and you are free to hand it to your own developer, to us, or to nobody at all.

How the audit runs

1. One conversation with each person who touches the work. Thirty to forty-five minutes, and mostly one question — walk me through your whole day, start to finish, and tell me where it grinds. Not a form, not a survey. The frustration is the signal; people are reliable narrators of what annoys them and unreliable narrators of how long it takes. One to three days, depending on headcount.
2. Watching three or four of them actually do it, on screen share. This is the step that gets skipped and it is the one that earns the fee. We are watching for three things specifically — every moment something is copied from one window and pasted into another, every moment the person pauses to think (a pause is a hidden decision, and a decision becomes a filter or a branch later), and every time someone says "and sometimes it happens that…". That last sentence is where the exceptions live, and exceptions are what break automations after handover. Half a day to a day.
3. Costing every recurring task. How often it happens, how long each time, who does it, and roughly what that person's hour costs fully loaded. That produces monthly hours and monthly money per task. Where a number was estimated rather than observed, it is labelled as an estimate and the basis is written next to it, because a roadmap built on quietly invented figures is worse than no roadmap. About a day.
4. Scoring each task for effort and for benefit, then ranking. Two numbers on a simple scale, each with a one-line reason. High benefit and low effort rises to the top; high effort and modest benefit gets named as such and pushed down rather than deleted, so you can see the reasoning rather than take it on faith. Half a day.
5. Writing the roadmap. Three phases — the first 30 days, the following 60 to 90 days, and later — with the work, an estimated cost and an estimated monthly saving against each. Alongside it, the deliberately unglamorous part: the tasks that should stay manual, and why. Low volume, changes shape too often, or a judgement call where being wrong costs more than the hours saved. One to two days.
6. Presenting it and answering the questions it raises. The document is walked through live, because the objections that surface in that room are usually the ones that decide whether phase one ever happens. You also get the list of things we could not answer and what would need to be opened up to answer them. Half a day.

The toolkit here is deliberately boring — a document, a spreadsheet and a deck. There is no platform to buy and no software to install for this piece of work. The instrument is the questions and the watching.

What moves the price

The range is $500–$5,000, and three things set where you land in it.

How many people and how many departments. A five-person company with one process is a fundamentally different exercise from a fifty-person company where sales, operations and finance each have their own version of the truth and the interesting failures happen in the gaps between them. Interview time scales with headcount; reconciliation time scales faster.

Whether we are allowed to watch, or only to be told. If the work can be observed on screen share, the estimates are grounded and the audit is quicker. If access is restricted and everything has to be reconstructed from description, it takes longer and the confidence markers in the document get more honest and less flattering.

How far past the document it goes. A written roadmap on its own sits at one end. A roadmap plus a board-ready presentation, plus follow-up calls while the owner takes it through internal approval, sits at the other.

What usually goes wrong

The audit becomes a list of fifty things. It looks thorough and it is the most common way this deliverable fails. An owner opens a fifty-item findings list, feels the weight of it, closes it and does nothing — and a year later the same conversation happens again. A useful audit says three things clearly. Do this one now, because it returns the most for the least. Then these two. And do not touch these, for these reasons. Everything else belongs in an appendix. The job is not to inform you; it is to make the next decision easy.

The savings column gets read as a redundancy plan. Hours returned and salaries removed are not the same thing, and a roadmap that blurs them causes two separate problems. The first is that the promised saving does not appear on the P&L and the whole project is judged a failure. The second is worse and happens during the audit itself — if the people being interviewed suspect the exercise is about replacing them, they will describe the process inaccurately, and they are entirely within their rights to. The roadmap is then built on a fiction. Every finding in this document is framed as hours returned to a named person and what those hours go to instead.

A phase gets costed from what somebody said rather than from what was opened. The line "our data is all in Excel" appears in almost every audit, and it is worth nothing until the file has been looked at. Open it and you routinely find one customer entered under three different names, dates in three formats, and a single column holding both numbers and free text. That is two days of cleanup that was priced at zero, sitting in phase one where it will blow the first estimate you gave and poison confidence in the rest. Any phase touching existing data is either costed after the actual file has been opened, or written up explicitly as excluded pending that.