The CoReCo Map
Monday morning. You run a forty-person firm, and the board wants “an AI strategy” by Friday.
By lunch you have ten requests. Sales wants proposal drafting. Finance wants reconciliation automated. Ops wants scheduling. Support wants reply macros. The founder's nephew wants “an internal ChatGPT.” Three vendors have sent decks; each says their use case is the transformative one.
You have the budget — and the organisational patience — for exactly one pilot. Pick wrong, and 'AI' gets a bad name here for two years.
Before any framework: what would you honestly need to know about each of those ten tasks to rank them? Not about the AI — about the tasks. Write down two or three properties.
Friday is coming. Which pilot do you fund first?
Pick one — committing first is what makes the answer stick.
the lesson continues after you choose
The instinct most leaders bring to Friday is to start from the technology: what can AI do, and where could we use it? It feels rigorous — it's how every vendor deck is organised.
But it inverts the decision. The question that actually ranks your ten requests has nothing to do with models and everything to do with anatomy: which tasks, exactly, does this firm perform — and what is each one made of? That inversion has a method.
The method is CoReCo — Think North's framework for AI-adoption prioritisation. It works at the task level, not the department or 'use case' level: list every task each person does — not roles, tasks: “chase overdue invoices,” “assemble the Monday ops report,” “answer where-is-my-order emails.” Then score each task on three dimensions:
- Co — Complexity: how clear are the steps to execute? Could you write them down such that a competent temp gets it right? (Clear steps = low complexity.)
- Re — Recurrence: how often does the task occur — hourly, daily, weekly, quarterly?
- Co — Cost: what does it cost in total — time × people × materials × money — across a year?
Plot every task as a bubble — recurrence across, step-clarity up, bubble sized by cost — and the map hands you four strategies. Clear steps + high recurrence: automate now. This is where AI ROI actually lives; big bubbles here are your Friday answer. Unclear steps + high recurrence: clarify first. This quadrant is process debt — the task recurs constantly but exists only in someone's head; standardise it, then automate it (often the biggest long-term win on the map). Clear steps + rare: script it. A checklist, an SOP, a macro — using AI here is renting a rocket to cross the street. Unclear + rare: leave alone. Judgment calls and one-offs; this is where your humans are irreplaceable, and where forced automation goes to die.
Why task-level granularity is the whole trick: the research on failed pilots (see the ROI lesson on this shelf) keeps finding the same pattern — tools bolted onto departments ('AI for sales!') rather than fitted to tasks, so nothing integrates and nothing gets measured. And the jagged frontier from the Limits shelf makes the same demand from the other side: AI capability varies task by task, so only a task-level map can even ask the right question. CoReCo is where the capability map and the value map meet.
So Friday's board question was never “which AI?” — it was “which task?”, and that question is answerable with a spreadsheet, three scores and an honest hour per team, no vendor required. The ten competing requests aren't politics anymore; they're bubbles on one map, and the map does the arguing. That's the shift: AI strategy stops being a technology bet and becomes an exercise in knowing, precisely, what your firm does all day.
Your rule: when anyone proposes an AI initiative, ask three questions in order — “Which task? How often does it happen? Can you write down its steps?” No named task: it's a slogan. Rare: it's a script, not a platform. Steps nobody can write: it's a process-clarification project wearing an AI costume — valuable, but budget it honestly.