Framework · free to use

Which workflows should AI touch first?

The most common failure mode in AI rollouts is doing too many things at once. This framework gives you a way to score and rank the 10-30 candidate workflows that emerge from any honest audit — so you can confidently pick the 3 that ship first.

The scoring model

Five dimensions, 1-5 each

Score each candidate workflow on five dimensions. Total score is out of 25. Use the top 3 for your first wave. Everything else goes on a ranked backlog.

Dimension	What 5/5 looks like	What 1/5 looks like
Time leverage	Saves 8+ hours/week per user; multiple users	Saves under 30 minutes/week; one user
AI fit	Pattern is text-heavy, judgment-light, repeatable	Pattern requires deep domain judgment or live human interaction
Data availability	We have 10+ examples of past good output to train against	We have no examples or only the user knows what good looks like
Owner availability	A named human will own this and use it weekly	Nobody specific cares; rollout would be top-down
Low risk profile	Output is internal-facing or reviewable before it ships	Output goes directly to customers or regulators with no checkpoint

Worked example

Scoring a real workflow

Workflow: Drafting B2B sales proposals (12-person SaaS team)

Time leverage: 5/5 — saves ~8 hrs/proposal × 10 proposals/month = 80 hrs/month across the team.
AI fit: 5/5 — text-heavy, follows a repeatable structure.
Data availability: 5/5 — 30+ past winning proposals on file.
Owner availability: 5/5 — VP Sales will own it; she writes 4 proposals a week.
Low risk: 4/5 — output goes to prospects, but always reviewed by AE + senior leader.
Total: 24/25. Top priority. This is what we shipped in week 4 of the rollout.

Workflow: AI-generated outbound sales emails at scale

Time leverage: 3/5 — SDR team is small; impact moderate.
AI fit: 3/5 — model can write competent emails, but voice and personalization need work.
Data availability: 2/5 — only ~5 past outbound emails that converted, not enough to anchor voice.
Owner availability: 4/5 — SDR manager would own it.
Low risk: 2/5 — direct-to-prospect, low review threshold = brand risk.
Total: 14/25. Backlog. Worth doing later after voice library is built.

Common mistakes

Where prioritization goes wrong

Scoring on coolness instead of leverage. Voice agents are cool. Most companies do not get 5/5 on Time Leverage from voice. Score on the math, not the demo.
Ignoring Owner Availability. No matter how high the leverage, a workflow with no owner will not ship. Owner Availability is a multiplier — not just a tie-breaker.
Skipping the Low Risk score. The fastest way to kill an AI rollout politically is a single embarrassing customer-facing mistake. Pick low-risk workflows first to build trust.
Doing 7 workflows at once. The top 3 method is not a suggestion. With 7, owners get diluted and nothing ships.

The ranked backlog

Where everything else goes

Workflows scoring 17-19: second wave, days 60-90 of the rollout, after the top 3 are stable.

Workflows scoring 13-16: third wave, quarter 2, often with adjusted scope (e.g., add a human-review step to raise the Low Risk score).

Workflows scoring below 13: parked. Revisit when conditions change — owner appears, data becomes available, model capability shifts.