Thinking · Guide

Turn AI ideas into an action plan.

Most teams have a list of AI ideas and no honest way to choose. Here's the value-and-risk framework we use in our workshops to sort any idea into a clear next step — build now, pilot, experiment, or wait — plus a scoring checklist and worked examples you can use today.

The framework

Two questions sort any idea.

The industry-standard way to prioritize AI is an impact-versus-feasibility matrix. We use two reads that make it concrete — and, as Gartner notes, feasibility matters just as much as value.

Read 1 · Value

Is it worth building, and how big?

Weigh the value if it works against the cost and complexity to build it. High value and low effort is where you start; high value and high effort is a deliberate bet; low value gets experimented with or parked.

Read 2 · Risk

How much should a human stay involved?

Weigh the type of knowledge it needs — explicit and documented vs. tacit judgment — against the cost of being wrong. Together they tell you how much human oversight a use case needs to be safe.

The value read

Four places an idea can land.

Plot value against cost and complexity, and every idea falls into one of four zones — each with a different move.

High value · low cost

Quick Wins

A simple prompt or small build your team can use right away — usually inside a tool like ChatGPT, Copilot, or Claude.

Move: build it now.
High value · high cost

Big Bets

A larger workflow or connected system — bigger than a prompting tool, with real integration underneath. Worth it when the value is real.

Move: pilot first, then commit.
Low value · low cost

Experimental

Cheap to try and modest payoff. Worth a quick experiment to see whether more value shows up once it exists.

Move: try it yourself.
Low value · high cost

Wait & See

Sounds great, but it's beyond what AI or your data can do well today. Hold a few months and revisit — AI moves fast.

Move: park it, re-check soon.
The risk read

How much human stays in the loop.

Plot the type of knowledge against the cost of errors, and you get the right level of oversight for each idea.

Explicit · low stakes

No Regrets

Clear-cut work where mistakes are cheap and easy to catch — bulk replies, summarizing documents, screening résumés.

Oversight: let it run, monitor lightly.
Explicit · high stakes

Quality Control

Clear-cut, but mistakes are costly — drafting high-value contracts, production code, due diligence.

Oversight: AI produces, a human verifies.
Tacit · low stakes

Creative Catalyst

Judgment-heavy but low-risk — ad concepts, sales scripts, product ideas.

Oversight: AI suggests, a person picks.
Tacit · high stakes

Human-First

Judgment-heavy and high-stakes — setting strategy, enterprise integrations, disciplinary decisions.

Oversight: a person leads, AI assists.
Score each idea

Five quick reads per idea.

Run every idea through these. Be honest, not aspirational — especially on data and effort.

01

Value if it works

Nice-to-have, real impact, or genuinely game-changing?

02

Cost & complexity

A quick prompt, a workflow, or a big connected system?

03

Type of knowledge

Clear rules and data, or human judgment and nuance?

04

Cost of being wrong

Cheap to catch, or costly and hard to undo later?

05

Data readiness

Is the data clean, reachable, and cleared for use?

This mirrors the weighted-scoring approach used across the industry — value, feasibility, time-to-value, reuse, and strategic fit, with risk as a penalty. Score it collaboratively across business, IT, and compliance to keep it honest.

Worked examples

The same idea, sorted two ways.

Run each through both reads, and the next step writes itself.

Draft weekly client recap emails
Quick WinQuality Control

Build it now — small build, real time saved — but keep a human reviewing before anything sends.

Answer routine, bulk customer questions
Quick WinNo Regrets

Automate it — clear-cut and low-stakes — and monitor lightly rather than reviewing every reply.

Generate ad concepts and sales scripts
ExperimentalCreative Catalyst

Let AI generate options and have a person choose — cheap to try, judgment stays human.

Set next year's company strategy
Wait & SeeHuman-First

Keep a person leading — high-stakes and judgment-heavy. Use AI to research and pressure-test, not to decide.

Turn meetings into documented SOPs
Big BetQuality Control

Pilot it — high value, real build — with a human verifying each document against the source.

"Replace our whole support team"
Wait & SeeHuman-First

Park it — beyond what's safe or feasible today. Revisit once the smaller wins are proven.

Build the plan

Five steps to a real action plan.

You don't need a perfect roadmap — you need a prioritized list and a first move.

01

List every idea

Capture everything, from the whole team. Don't filter yet.

02

Score each one

Run all five reads on value and on risk above.

03

Sort Now / Next / Later

Quick Wins first; Big Bets earn a pilot before you commit.

04

Assign oversight

Decide how much human stays in the loop for each one.

05

Ship one Quick Win

Momentum beats a perfect roadmap. Start there.

One thing most people miss: trend beats position. Use cases move — voice agents looked like "wait and see" eighteen months ago and have shifted up and to the left. Re-score your list every few months, because the line keeps moving in your favor.

Want our read on your shortlist?

Bring the AI ideas you're weighing and we'll score them with you against your data and environment — and turn the top corner into a plan you can act on. About 30 minutes, no slides.