What is the process for mapping our workflow so AI can turn it into a strategy?

Why mapping everything first is the wrong start

The instinct is to document every process in the firm and hand the result to AI for a plan. The published material points the other way. The service design essay in the newer book says many processes of the last era, workshops, committees, stage gates, documentation rituals and periodic reviews, were built to compensate for the absence of intelligence, and that in an AI-native firm many of them become redundant. The other-firms answer on this site reports where most firms are: AI tools added to an unchanged operating model, producing real but small benefits. And the EOS essay says processes increasingly live inside software rather than documents, so the question shifts from who is accountable to whether the system is designed to produce the outcome.

As an inference, a complete map of the current firm is a detailed picture of the thing you are about to change. It is expensive to produce and out of date the moment the first workflow is redesigned.

Step one: choose the workflow from the P&L

The AI strategy answer on this site says the order comes from the profit and loss statement, not from a tool list. The newer book organizes the firm the same way, through the AI-Native Firm Map: revenue, less cost to serve, is gross margin; less sales and overhead, is EBITDA; with nineteen roles arranged under those lines. The automation answer puts it as one decision, one owner, one capability, and then the next.

As an inference, ask where the firm loses the most money or hits a capacity wall first. Margin lost during delivery, proposals that take a senior person days, a pipeline nobody trusts, month-end reporting that arrives too late to act on. Pick one, and name the person who will own it.

Step two: map that workflow at task level, from real work

The replication chapter of the 2020 book describes how to get expertise out of people and into the firm. Start with postmortems of a representative sample of recent engagements, break each one down, look at how the work was performed at the task level, and inventory the knowledge and skills each task requires. The process diagnosis answer on this site adds the classification to apply when something is broken: a design failure, a capacity failure or an accountability failure, because only the first is fixed by redesigning the process.

As an inference, for each task capture what starts it, who does it, how long it takes, what decision is made, what information it needs and where that information lives, what it hands to whom, and where rework or waiting happens. Map the work as it was actually done on real projects, not as the process document says it should be done.

Step three: sort every step by what it needs

The AI agent answer on this site makes the division of labor concrete: give AI the continuous monitoring, synthesis, forecasting and memory, and keep judgment, tradeoffs, escalation and accountability with a named person at a named checkpoint. The EOS essay describes the same split for running the firm: systems capture commitments, monitor progress continuously, detect drift and escalate on thresholds, while people set direction, exercise judgment, resolve tradeoffs and design the system.

As an inference, label each task on your map with one of those needs. Steps that exist only because someone has to remember, chase, copy information between systems or compile a report are the ones AI should take. Steps where someone decides, says no to a client, or signs their name to the result stay human, and the map should show who and where.

Step four: redesign, do not automate

The automation answer on this site is blunt: automation reduces cost; redesign is what creates leverage. The IT essay says the core workflows of selling, scoping, pricing, delivering and expanding work should be re-engineered around intelligence rather than human memory or manual coordination. The AI agent answer adds that the gain comes from redesigning how work happens rather than accelerating the current version of it.

As an inference, once steps are sorted, draw the new workflow rather than marking up the old one. Many steps disappear entirely, because they existed only to move information or check that something happened. Handoffs move to where the system can see them. The human checkpoints become fewer and more deliberate.

Step five: set the standard and prove the measure moved

The AI agent answer says to prove feasibility and cost to serve before launch, count AI and tooling inside contribution margin, and treat any workflow that still needs heroics as unfinished. The delivery automation answer says to decide in advance where recovered hours go, into more engagements, into margin by moving off hourly billing, or into higher-value work, because taking hours out while still charging for time cuts your own revenue.

As an inference, write the definition of done for the redesigned workflow and the one number on the P&L it should move, then measure it before choosing the next workflow.

Where AI helps with the mapping itself

The service design essay says documentation overhead falls when systems generate and update it automatically. As an inference, AI is useful at step two: it can draft a first version of the current-state map from meeting transcripts, project records, proposals and existing procedures, and keep the redesigned map current as the work changes. People still have to check it against what really happens, because a map built only from documents repeats what the documents claim.

What becomes the strategy

As an inference from the AI strategy answer, the strategy is not a document produced from the map. It is the ordered list of workflows, chosen from the P&L, each rebuilt, measured and owned before the next one starts. After three or four, the list of what to do next usually writes itself.

What we do not prescribe

Collective 54 publishes no workflow mapping template, notation or tool, and no target number of workflows to redesign. The published positions are starting from the P&L and the AI-Native Firm Map, one capability at a time, task-level breakdown from postmortems, the split between AI monitoring and memory and human judgment and accountability, redesign over automation, proving cost to serve, and deciding where recovered hours go.

When this answer flips

If the firm has no consistent way of delivering its work yet, as an inference, the first job is standardizing one offer, which the playbooks and SOPs answer covers, before any AI redesign.

If clients or regulation limit where data can go, mark those constraints on the map before sorting steps.

And if the founder is the only person who knows how a workflow runs, step two is also a succession exercise, and the 2020 book suggests renting an instructional designer to get that knowledge out.

The short answer

Do not map the whole firm and ask AI for a plan. Start from the P&L: pick the one workflow where money leaks or capacity runs out, and name its owner. Map it at task level from recent engagements, the way the 2020 book builds certification, capturing who does each task, what it decides, what information it needs and where rework happens. Sort every step: memory, monitoring and synthesis go to AI; judgment, tradeoffs and accountability stay with a named person. Then redesign the workflow rather than automating the old one, set a definition of done and one number it should move, and prove it moved. AI can draft and maintain the map. The strategy is the ordered list of workflows you rebuild, one at a time.