Select the First AI-Native Workflow

Why workflow selection matters

The workflow is the unit of AI transformation.

Most failed AI initiatives are framed too broadly:

Those are themes, not workflows.

A workflow has a trigger, inputs, decisions, outputs, owner, quality bar, and measurable outcome.

Better examples:

Turn a new inbound lead into a qualified CRM record and follow-up draft.
Convert a customer call transcript into a support summary, risk flag, and account update.
Review an RFP and produce a compliance matrix, unanswered question list, and proposal outline.
Convert a board reporting pack into variance commentary and executive talking points.

Score candidate workflows from 1 to 5 across eight criteria:

Criterion	What to look for
Frequency	Happens often enough to matter.
Pain	Current process is slow, inconsistent, or expensive.
Clear outcome	Everyone agrees what good output looks like.
Context availability	Inputs, examples, and policies can be gathered.
Risk boundary	Mistakes are recoverable or reviewable.
Owner	One person can make process decisions.
Measurement	Cycle time, quality, cost, adoption, or revenue can be tracked.
Reuse	Pattern can transfer to other teams or workflows.

The best first workflow is rarely the flashiest. It is the workflow where the organisation can learn fastest with controlled risk.

For the workflow you select, write a one-page brief:

Workflow name:
Use a verb phrase, not a department name.

Business outcome:
What changes if this works?

Trigger:
What starts the workflow?

Inputs:
What documents, records, messages, data, or API calls are needed?

Current steps:
What happens today?

AI role:
Where should AI draft, extract, classify, route, recommend, or act?

Human role:
Where should a person approve, edit, reject, or escalate?

Quality bar:
What does a good output include? What mistakes are unacceptable?

Metrics:
What will you measure before and after?

Avoid these as your first AI-native build:

Too vague: "Improve knowledge sharing."
Hard to measure, easy to drift.

Too risky: "Let agents approve refunds over a certain threshold."
Start with recommendations and human approval first.

Too political: "Automate management reporting across every function."
Cross-functional alignment will slow learning.

Too low volume: "A quarterly board process."
Important, but slow feedback.

Too context-poor: "Answer any customer question."
Start with a specific question category and approved sources.