Process Mapping Before You Automate Anything

A Zensurance survey of 1,000 Canadian small business owners published this week found 27% saying they are not sure how AI even applies to what they do. Read that as an operations finding rather than a technology one. In most of those businesses nobody has ever written down how the work actually happens, so there is nothing concrete for a tool to be matched against.

Map reality, not intentions

A process map records how the work gets done. Not how the manual says it should, not how it was designed in 2019, and not the version anyone describes in a meeting.

That distinction is the whole exercise. The gap between the official process and the real one is where the workarounds live, and the workarounds are usually the most informative thing on the page. Somebody re-keys a number because two systems do not talk. Somebody keeps a private spreadsheet because the report is wrong. Somebody phones a supplier because the portal is slower than a call. None of that appears in a process designed at a desk.

The one-hour method

1. Pick one process with a clear start and end. A customer enquiry becoming a booked job. A supplier invoice becoming a payment. Not "sales" and not "admin," which are categories rather than processes.

2. Sit with the person who does it. The person, not their manager. Managers describe the intended process with great confidence and are frequently wrong about the middle of it.

3. Walk through one real recent example. Pull up an actual job from last week and follow it. Hypotheticals produce the clean version; a real case produces the one with the missing purchase order.

4. Ask two questions after every step. What happens when this goes wrong, and what are you waiting for before the next step. Those two questions surface the exception paths and the delays, which is where nearly all the recoverable time sits.

5. Record five things per step. What happens, who does it, roughly how long the doing takes, what it waits on, and how often it goes wrong. Five columns on a sheet of paper.

That example totals about an hour of work spread across five days. Almost every owner shown a table like this points first at the 40-minute quote, because it is the biggest doing number. The four days of waiting are the actual problem, and no drafting tool touches them.

Three things the map usually reveals

Waiting dominates doing. In most small business processes the elapsed time is several times the worked time. Customers experience elapsed time, so that is the number that loses you jobs to whoever quoted on Tuesday.

There is a rework loop nobody mentions. A step that sends work backwards when information is missing, running often enough to matter and invisible in any summary. Ask how often the previous step has to be redone.

One person is load-bearing. Some step depends on knowledge that exists in one head. That is a continuity risk before it is an automation opportunity, and writing it down is worth the hour on its own.

Now score the steps

With the map in front of you, a step is a reasonable automation candidate when it repeats in the same shape, happens entirely on a screen, touches one or two systems, produces output someone knowledgeable can check quickly, and costs little to correct.

A step is a poor candidate when it needs someone physically present, requires judgment on ambiguous inputs, carries accountability for the outcome, or involves getting agreement between people who want different things. Parts of those can be assisted. The step itself stays with a person.

Score the waiting rows too, because they often have cheaper fixes than the doing rows. Two days of phone tag is solved by a booking link, not a model. Two days waiting for the owner to review is solved by an approval threshold, which is a decision rather than a purchase.

The expensive mistake

Automating a process that should have been redesigned makes the wrong shape permanent and harder to change. A tool that speeds up a step existing only because two systems never talked has locked in the disconnection, and now it has a subscription attached.

Two questions before automating any step. Should this step exist at all, and does anything downstream read its output. A surprising number of reports, logs and summaries are produced diligently every week and read by nobody, which we wrote about in AI scales a broken process. Deleting a step beats automating it every time.

What to do with the map

Turn the steps you keep into a written procedure, which is the thing an agent can actually follow and the subject of standard operating procedures an AI can follow. Put the biggest waiting row on the list of things to fix without software. Then take one automation candidate, pick a tool, and measure it against the doing time you recorded, which is why the timings matter.

Our ROI calculator turns minutes saved per instance into an annual figure, and capturing the hours AI saves covers what to do with the time once it appears, which is the step most businesses skip.

Map three processes and the question of where AI fits stops being abstract. The 27% who are unsure are not short of AI knowledge. They are short of a description of their own work specific enough to match anything against.

Frequently Asked Questions