Before you automate anything, map the process on paper. Pick one painful, frequent workflow and write every step as it really happens — who does it, how long it takes, where it stalls. Mark the rework loops, handoffs, and “it depends” decisions. Then delete, simplify, and standardise with no technology at all. Only what’s left — the repetitive, rule-based, checkable steps — is a real candidate for AI. Most of the win is in the map.
A practical companion to Process-First AI Adoption and The AI Productivity Paradox.
Why map a process before automating it?
Because automating a broken process just makes it break faster. A map shows you where work really stalls, repeats, and changes hands — so you can delete and simplify first. Most of the payoff comes from fixing the process; only what’s left after that is a real candidate for AI.
Here’s the uncomfortable part: a tool inherits whatever process you point it at. Aim AI at a messy workflow and you don’t get a clean one — you get the same mess, running faster and harder to see. That’s a big reason MIT’s Project NANDA found 95% of enterprise generative-AI pilots delivered no measurable return — the cause was mismanaged adoption, not weak models (Fortune). The firms that win do the unglamorous work first: McKinsey found AI high performers were about 3× more likely to fundamentally redesign their processes than everyone else (McKinsey).
Mapping is the front half of how we work — the Map step in Map → Redesign → Match → Embed. The good news is you don’t need us, and you don’t need software, to start. You need a pen, a quiet hour, and a willingness to write down what really happens.
If you can’t draw the process, you can’t automate it.
How do you map a business process? (6 steps)
Pick one painful, frequent process. Write every step as it really happens, not the official version. Mark who does each one, how long it takes, and the tools used. Find the stalls, rework loops, and handoffs. Delete, simplify, and standardise — then, and only then, flag what’s left for AI.
The whole method fits on one page. Work through these six steps in order — the first five involve no technology at all, and that’s the point.
Pick one painful, frequent process
Choose a single workflow that hurts and happens often — handling an inbound lead, processing an invoice, onboarding a client. Don't try to map everything. One process, one clear trigger, one clear outcome.
Capture every step as it really happens
Write each step in the order it actually occurs — the real version, not the SOP. Sit with the people who do it and record the shortcuts and workarounds. The honest map is messier than the official one.
Mark owner, time, and tools
For each step, note who does it, roughly how long it takes, and which tools they touch. A flat list becomes a picture of cost — the time-eaters and the app-to-app jumps appear straight away.
Find the stalls, rework loops, and handoffs
Mark where work waits, where it loops back for fixes, and every point it changes hands. Flag the “it depends” decision points too — those judgement branches shape what you can and can't standardise.
Delete, simplify, standardise
Fix the map before any tool. Cut steps that exist only by habit, simplify the overcomplicated, standardise decisions that don't truly need judgement. Most of the win lives here — with zero technology spend.
Mark what's left for AI
On the cleaned-up map, flag the steps that are repetitive, rule-based, and checkable. Those are your AI candidates. Judgement, empathy, and true edge cases stay human. Now — and only now — you match a tool.
What does a finished process map look like?
A finished map is a numbered list of every real step from trigger to outcome, each tagged with who does it, how long it takes, the tools used, and whether it stalls. The bottlenecks and rework loops jump out, the “it depends” branches are flagged, and a handful of steps are marked as honest AI candidates.
Take a concrete one: handling an inbound lead. The trigger is a form submission. The official process is three steps. The real one looks more like this — a salesperson sees the email whenever they next check their inbox, copies the details into the CRM by hand, googles the company, drafts a reply from memory, waits on a manager to approve pricing, then chases the lead three days later because the first reply sat in drafts. Seven steps, two stalls, one rework loop, and a handoff that depends entirely on who’s online.
Written down, the problems are obvious: the lead goes cold in the gap between “form submitted” and “someone notices.” The manual CRM entry is pure copy-paste. The pricing approval is a real it depends — sometimes standard, sometimes a judgement call. A good map makes all of that visible before anyone argues about which tool to buy. Use the template below to capture yours the same way.
No download needed. Copy these eight columns into a doc or whiteboard, one row per step, and work down your process.
Step name — What happens, in plain words (e.g. “copy lead into CRM”).
Trigger — What kicks this step off — a form, an email, the previous step finishing.
Who does it — The actual person or role — not who's supposed to.
How long — Rough time it takes, plus any waiting before it starts.
Tools used — Every app, doc, or system the step touches.
Decision points — Any “it depends” branch or judgement call here.
Where it stalls — Waits, rework loops, or handoffs that lose time.
Automate? (Y/N) — Filled in last — only after you've fixed the process.
How do you know which steps AI should handle?
A step is a strong AI candidate when it’s repetitive, rule-based, and checkable — same input, predictable output, easy to verify. Keep humans on anything needing judgement, empathy, or messy edge cases. The test is simple: if you can’t write the rule down, AI can’t reliably follow it.
On the inbound-lead map, the copy-paste into the CRM and the first-draft reply are textbook candidates — repetitive, rule-based, and easy to check before they go out. The pricing decision is not: it carries judgement and risk, so a human stays in the loop (AI can tee up the options, not sign off on them). This is also where you decide you need no tool at all — sometimes the fix is a shared inbox rule, not a model. We tell half our clients exactly that.
- Repetitive, same-shape work (copy-paste, data entry, formatting).
- Rule-based decisions you can write down as if-this-then-that.
- First drafts a human still reviews before they go out.
- Outputs that are easy and fast to check for correctness.
- Judgement calls — pricing, exceptions, trade-offs.
- Anything needing empathy, trust, or a real relationship.
- True edge cases and one-offs with no clear rule.
- Decisions where a wrong answer is costly and hard to undo.
What do you do after you’ve mapped it?
After mapping, you redesign the process around the fixes you found, match the right tool to each remaining AI candidate, then embed it into the daily workflow and measure the result. The map is the input to every step that follows — redesign first, automate second, measure always.
That’s the back half of our method: Redesign → Match → Embed. You’ve already done the hardest, highest-leverage work by mapping honestly. If you want a second set of eyes on the map, that’s exactly what a scoped AI adoption audit does — and where a custom build picks up once the candidates are clear. You can even start by talking a workflow through with Maggie. But the map comes first, and most of it you can do today, for free, with a pen.
If you can’t draw it, you can’t automate it. Map first, then decide what AI is even for.
FAQ
What is business process mapping?+
Business process mapping is writing down every step of a real workflow in order — from the trigger that starts it to the outcome that ends it — and marking who does each step, how long it takes, and where it stalls. It turns a fuzzy, in-people's-heads process into something visible you can fix, standardise, or automate.
Do I need special software to map a process?+
No. Paper, a whiteboard, sticky notes, or a plain document work fine for almost every small business. The value comes from the thinking, not the tool — a clear list of steps in order beats a fancy diagram nobody finishes. Reach for dedicated mapping software only once the process is genuinely large or branching.
How detailed should a process map be?+
Detailed enough that someone new could follow it and do the job, but no more. Capture every real step, decision point, and handoff — including the messy rework and the “it depends” branches — but don't drown it in sub-steps. If a step never changes the outcome or who acts next, you can usually fold it in.
How long does it take to map a process?+
A single, well-bounded process usually takes a couple of hours to map honestly — longer if several people have to compare how they each actually do it. Most of that time is spent surfacing the unofficial version and the exceptions. Resist mapping ten processes at once; one clear map beats ten half-finished ones.
Can AI map my process for me?+
AI can help draft and tidy a map once you feed it the steps, but it can't see how work really happens in your business — the workarounds, the bottleneck, the unwritten rules. The accurate first version has to come from the people who do the job. Use AI to organise the map, not to invent it.