Workflow maps are indispensable. They show how work moves across systems and people, where approvals occur, and where delays or controls sit. They are excellent instruments for process improvement and automation.

They are not,however, a complete picture of work.

A workflow diagram describes the plumbing: request received, information checked, decision made, output issued, record stored. An activity map reveals what people do inside those boxes: read, compare, investigate, interpret, draft, consult,negotiate, explain, and decide. The latter is where much of the human effort—and much of the near-term augmentation opportunity actually lives.

Take contract review. A workflow view may show: request, draft, review, negotiate, approve,archive. It makes the routing, approval, and document-management opportunities visible. But “review” is not a single activity. It can involve identifying clauses, comparing them with policy, researching precedent, assessing commercial exposure, discussing trade-offs, proposing revisions, and escalating material risk. Some of those activities can be assisted by AI today. Some should remain entirely human-led. A workflow diagram alone cannot make that distinction.

This is whyworkflow-led future-of-work programmes naturally lean toward automation.Workflows make structured, repeatable, hand-off-heavy work highly visible. Theinformal, judgment-heavy, and cognitive work that surrounds it often becomes alarge box labelled “review” or “assess”. If leaders only use that view, they see the automatable edge of work and miss the augmentable core.

An activity-led model corrects the bias. It asks what work is actually being performed, how much time it consumes, what capabilities it requires, and how technology may alter it. Each activity can then be assessed against two independent questions:

• How feasible isit to automate this activity or a defined component of it?
• How valuable isit to augment the person performing it?

That produces aportfolio rather than a binary answer. Some activities are high automation andlow augmentation: standard routing, structured validation, routine datatransfer. Others are low automation but high augmentation: research synthesis,first drafting, pattern spotting, or preparation for a difficult decision. Someare high on both, where AI can prepare work now and parts may later bestandardised. Others remain predominantly human because trust, accountability,physical context, or complex social judgment are intrinsic to the work.

This mattersbeyond technology selection. Activities are the bridge to skills, careers,workload, and employee communication. Employees may not recognise themselves ina process map. They recognise “I spend too much time searching forinformation”, “I repeatedly draft similar material”, or “I need to reconcileconflicting evidence before advising a client”. An activity-led view givesleaders language that is both analytically useful and human.

The practicalconclusion is not to discard workflows. Use workflows for what they do best:designing and operating automation. Use activity maps as the primary substratefor understanding how AI changes work across roles, including both automationand augmentation. Link the two. Do not confuse them.

**Next article:**How do you move from this principle to a credible map of what people actuallydo?