NOTE: To use this framework, copy the content using the “Copy page” button above, then fill in the worksheets for your organisation.

AI use case prioritisation framework

The default way to start an AI programme is from the bottom: a spreadsheet of ideas collected from across the business, scored, and turned into a string of pilots. Individuals get faster, and the numbers on the board deck don’t move.

This framework starts from the top. You begin with the goals your board already tracks, pick the two or three domains that move them, redesign how work flows in each domain, and only then score the use cases that bring the new process to life.

It draws on two books we recommend to every CxO leading an AI programme: McKinsey’s Rewired (second edition), which sets out how to choose domains, and Andy Grove’s High Output Management, which explains why redesigning a process beats speeding up its parts. For the thinking behind each step, read How to pick the right processes to reinvent with AI and Rebuilding your processes with High Output Management.

The prioritisation ladder

Level	The question	What you produce
Value drivers	What creates value in our industry?	Two or three industry value drivers
Business strategy	Where do we play, and how do we win?	Your strategic choices
Goals	Which numbers are on the first slides of the board deck?	Five to eight board metrics
Domains	Which top-level processes move those numbers most?	Two or three domains
Value levers	Which metric in each domain drives its output?	Levers with baseline and target
Process redesign	How should this domain work with AI?	A redesigned process
Solutions	What capability would move each lever?	Two to four solutions per domain
Use cases	Which agents or workflows deliver each solution?	A scored use case backlog

---

Step 1: value drivers and business strategy

Every industry has two or three drivers of value that apply to any company in it: risk calculation in insurance, supply chain in retail, utilisation in professional services. You don’t choose these, but you need to name them.

Then write down your business strategy. We use the Playing to Win framework from A.G. Lafley and Roger Martin’s book of the same name:

• Where to play: which markets, segments and parts of the value chain you compete in.
• How to win: what you uniquely bring to win there.

AI opens new where-to-play and how-to-win combinations. You still don’t have an AI strategy: you have a business strategy enabled by AI.

Industry value drivers	Where we play	How we win

---

Step 2: the goals on your board deck

List the handful of numbers on the first slides of your board pack. Call them OKRs, KPIs or metrics; what matters is that the board already reviews them.

Board metric	Current value	Target and date	Executive owner

---

Step 3: pick two or three domains

Rewired calls an end-to-end area of the business a domain. You might call it a value chain, a customer journey or a top-level process. A company runs on 15 to 20 of them; pick the two or three that most affect the goals in step 2.

To list your domains, start from the APQC Process Classification Framework. It groups every organisation’s work into six operating processes (from developing vision and strategy to managing customer service) and seven management and support processes (HR, IT, finance, risk and others). Run the support processes well, but look for the biggest opportunities in the operating processes: they’re the ones tied to the board numbers.

Domain	Board metric it moves	Size of the effect (1-5)	Owner’s appetite (1-5)	Data readiness (1-5)	Score (multiply the three)

Choose the two or three with the highest scores, and name a business owner for each. If you’re mapping this in Claude, the Kowalah Plugin has the APQC framework built in; see Build your AI operating model, starting from nothing.

---

Step 4: value levers, with baseline and target

A value lever is the metric at the top of a domain that determines its output. For each chosen domain, pick two to four levers and record where they are today.

Domain	Value lever	Baseline	Target	How it’s measured

If you can’t state the baseline, measure it before you build anything. It’s the number your day-90 board report will compare against.

---

Step 5: redesign the process, not the tasks

Andy Grove describes every team as a production line: inputs go in, labour is applied, outputs come out. The leader’s job is the highest output, at the required quality, for the lowest effective cost.

The tempting move is to give each role inside the process an AI tool: an email prompt for SDRs, a forecasting dashboard for managers. Each step gets faster and the output of the whole process stays the same. That is why individual productivity gains don’t show up at company level.

Instead, the domain owner redesigns the process:

1. Define the inputs and outputs. What comes in from other teams and processes, what goes out, at what quality and cost today.
2. Redesign with AI in mind. Work with someone who knows what current models and agents can do. Look for chances to:

• Reduce the number of steps
• Remove whole parts of the process
• Run steps in parallel
• Combine steps
• Cut the waiting time between steps
• Reduce the number of human roles required

Domain	Inputs	Outputs today (volume, quality, cost)	Redesigned process in one sentence

This is not an IT project. The business leader owns and redefines the process.

---

Step 6: solutions and use cases

For each value lever, list the solutions that would move it, then the use cases (individual agents or workflows) that deliver each solution. Build this list with the people who work in the domain; our use case discovery interview template gives you the questions.

Write each use case as a hypothesis:

We believe that [agent or workflow] will [change to the value lever] for [the people who use it].

Keeping use cases grouped under one solution in one domain pays three ways:

1. Shared data. The agents use the same sources, so you connect them once.
2. Compounding output. Each agent hands its work to the next, so the gains stack.
3. Focused change. The same group of people use them, so enablement is targeted.

---

Step 7: score with ICE

Score every use case from 1 to 10 on three dimensions and multiply them for a score out of 1,000: Impact × Confidence × Ease. Multiplying, rather than adding, means a use case that scores badly on any one dimension can’t hide behind the other two.

Domain	Solution	Use case	Hypothesis	Impact (1-10)	Confidence (1-10)	Ease (1-10)	ICE score	Rank

Scoring guidelines

Score	Impact	Confidence	Ease
10	Transforms a board metric	Proven in a similar context, data ready	Buildable now with existing resources
7-9	Moves a value lever measurably	Strong evidence, good data	Moderate effort, few integrations
4-6	Improves efficiency within the domain	Reasonable hypothesis, some data gaps	Significant effort, achievable with planning
1-3	Minor, limited scope	Speculative, major data gaps	Complex, many dependencies

Before ranking, check each high scorer against three questions: does it share data with the other use cases in its solution, does it hand work to them, and does it serve the same people? Where two use cases tie, pick the one that does.

---

Step 8: sequence into a roadmap

Pick the first release: one solution, in one domain, with a measurable effect on one value lever. Keep the first release small enough to be live in six to eight weeks.

Release	Domain	Solution	Use cases	Value lever and target	Business owner	Target date
1
2
3

To take the first release for approval, use our AI business case template. For the week-by-week plan from there, use our Claude Enterprise first 90 days checklist.

---

Run this framework in Claude

The worksheets work on paper. They work better against a live record of how your organisation runs, which is what the Kowalah Plugin gives Claude: your operating model, its processes and the opportunities already raised, in the conversation where you’re working. You don’t need a Kowalah engagement to start. Sign in with your work email and you get your own organisation to map from empty.

Framework step	In Claude with the plugin
1-2. Strategy and board goals	Claude reads your AI vision and success measures, so later steps are tested against them. Setting them is done in a Vision Map Workshop or in the Kowalah Platform
3. Pick your domains	Map your units and the processes they own (the APQC framework is built in), then run /programme-review to see coverage and where constraints concentrate
4. Value levers	Record timings and measures against each process; say in the conversation which metric each domain is accountable for
5. Redesign the process	/process-diagnostic shows where cycle time goes and what has never been measured; /process-redesign proposes a to-be process beside the original
6. Solutions and use cases	/spot-opportunities works through a team’s real friction and checks what already exists
7. Score with ICE	/raise-opportunity tests an idea against your vision, scores it and adds it to your backlog
8. Sequence into a roadmap	/programme-review checks whether the work in flight is aimed at the gaps that matter

To start from nothing, follow Build your AI operating model, starting from nothing, which takes one function from an empty model to mapped processes in an afternoon.

---

Worked example: marketing and sales

The example below is illustrative; the scores show how the method works, not a benchmark.

Value levers for the marketing and sales domain

Value lever	What it measures
Average deal size	Revenue per closed deal
Close ratio	Share of qualified opportunities won
Sales cycle length	Time from qualified lead to signed contract

Solutions: a lead scoring engine, a proposal automation platform, and a deal acceleration system.

Use cases for the lead scoring engine

Use case	Hypothesis	Impact	Confidence	Ease	ICE score	Rank
Lead triage agent	We believe a triage agent will route every inbound lead within minutes, shortening the sales cycle for inbound buyers	7	8	8	448	1
Lead scoring agent	We believe a scoring agent will focus sellers on the leads most likely to close, raising the close ratio	8	6	6	288	2
Lead contact agent	We believe a contact agent will book first meetings with qualified leads the same day, shortening the sales cycle	7	6	5	210	3

The three agents share the same CRM data, hand work to each other in order, and serve the same sales team, so they belong in one release.

---

Implementation notes

• Do steps 1 to 4 before anyone lists a use case. Ideas collected first anchor the conversation on tasks, not on the numbers that matter.
• The domain owner decides. IT and the AI team advise on what’s possible; the owner of the process owns the redesign and the metric.
• Score as a group. One person’s ICE scores reflect their bias. Score together and discuss the widest gaps.
• Re-score every quarter. Each release changes what you know about your data and your people.
• Keep it in one place. The domains, levers, redesigned processes and backlog belong in your AI Operating Model, so the next quarter starts from the record rather than a blank spreadsheet.

---

APQC Process Classification Framework (PCF) is an open standard developed by APQC, a nonprofit that promotes benchmarking and best practices worldwide. The PCF is intended to facilitate organizational improvement through process management and benchmarking, regardless of industry, size, or geography. To download the full PCF or industry-specific versions of the PCF, as well as associated measures and benchmarking, please visit www.apqc.org/pcf.