Pilot selection — picking the right team to start with

TL;DR. Most rollouts pick the wrong first pilot. Score every candidate function on four criteria — pain, measurability, owner readiness, sponsor alignment. A function below 3 (out of 5) on any criterion is not a Wave 1 candidate. Start with one function. Three use cases inside it. Never five.

The four criteria#

Score every candidate function on:

• Pain. Is there real, named, repeated pain in this team's recurring work? (1–5)
• Measurability. Will the team be able to measure hours saved or quality improved at week six? (1–5)
• Owner readiness. Is there someone willing to own the pilot for six weeks at four hours a week? (1–5)
• Sponsor alignment. Will the function head defend the pilot if priorities shift? (1–5)

A function below 3 on any criterion is not a Wave 1 candidate. Move on.

The shortlist filter#

Quickly eliminate:

• Functions with active reorganisation or leadership change.
• Functions mid-system migration (ERP, CRM swap).
• Functions whose primary documents are in formats Cowork handles poorly (CAD, video).
• Functions in a regulated zone where a deeper security review has not started.

These are not "bad teams" — they are bad timing. Come back in two quarters.

The scoring grid#

The function with the highest total is the candidate. If two are tied, pick the one with higher measurability — measurable wins compound; unmeasurable wins do not.

The "first 3 use cases" rule#

Within the chosen function, pick three candidate use cases — never one, never five.

• Three gives you optionality if one falls flat.
• Five disperses focus.
• One leaves you with no fallback.

Pick from the pilot scope quadrant of the pain/gain map.

Anti-patterns#

• "Start with the team that complains the least." They will not give honest feedback.
• "Start with the team that complains the most." They are usually angry about something else; the Cowork pilot inherits the anger.
• "Start everywhere at once." Adoption dilutes; quality drops; nothing ships at week six.
• "Start with the team that already uses ChatGPT." Often a good fit, but verify they will adopt the workspace pattern — which is more disciplined than ChatGPT habits.

The pilot kickoff one-pager#

Once selected, produce a one-page kickoff doc:

• Function
• Sponsor and pilot owner names
• 3 use cases
• 6-week timeline
• 2 success metrics
• Workspace folder and CLAUDE.md owner
• Rollback procedure

This is the artifact that goes back to the steering group.

Tinkso's take#

We turn down engagements where the sponsor wants us to start with five functions in parallel. The math is simple: a successful Wave 1 makes Wave 2 easier; a struggling Wave 1 of five functions costs the sponsor political capital and stalls the entire programme. One pilot, then expansion. Always.

Try this#

Score your three best candidate functions on the four criteria right now, from memory. The lowest-scoring criterion across functions is your real bottleneck — usually owner readiness or measurability. Fix that before you score again.