Build, Buy or Generate a Business Capability Model
There are four ways to get a business capability model: build it with your own team, hire a consulting firm to build it with you, generate a draft with an AI assistant, or start from a ready-made reference model. Each fits a different situation, and most teams end up combining two or three.
This page sets out what each is good for and what to watch for, what building one actually involves, and what happened when we asked an AI model for the same capability model three times.
Four Ways to Get a Capability Model
| Option | Good for | Watch for |
|---|---|---|
| Build it in-house | A model your own people wrote and own, in your organization's vocabulary, with the agreement that comes from building it together. | Drafting is the short part. Agreeing names, boundaries and definitions across leadership takes most of the time, and that is architect time not spent using the map. |
| Hire a consulting firm | A facilitated build when you lack the people or the method, with the buy-in that workshops create. | The cost of the engagement. Ask which reference model the firm starts from, and what you own when the engagement ends. |
| Generate a draft with an AI assistant | A first draft in minutes, brainstorming names and definitions, and checking your own map for areas you missed. | The same prompt returns a different structure each time, nothing in the answer shows what is missing, and there is no version anyone can cite later. The test below shows all three. |
| Start from a ready-made reference model | A complete, defined and versioned starting point, with definitions to L3 and KPIs from the first day, that you then tailor. | It costs money, and it is a reference, not your answer: you still adapt it to how your organization works. |
In practice the options combine well. A common pattern is to start from a reference model, tailor it in workshops, run either in-house or with a facilitator, and use an AI assistant for drafting and gap-checking along the way. An assistant connected to the reference model answers from it instead of inventing a new structure each time.
What It Takes to Build One
A capability model to L3, with definitions, goes through the same stages however it is built:
- Agree the scope and the levels: the whole enterprise or one function, and how deep to go.
- Draft L1 and L2, then decompose to L3 with a definition for every capability.
- Review the draft with the people who do the work, and reconcile overlaps and gaps.
- Add KPIs, and link the capabilities to value streams and data where you need them.
- Version it, so the people who use it are all working from the same model.
How long it takes depends on the scope, the depth, and how many groups have to agree. The drafting stage is the shortest, and it is the only one an AI model or a reference model removes. Review, agreement and tailoring remain whichever way you start.
We Tested It: The Same Prompt, Three Times
On October 31, 2026 we sent one prompt to a general-purpose AI model (through its API, with web search off) three times, and measured what came back. The prompt asked for a capability model for a hospital and health system, to L3, with a definition for every capability and two KPIs for every L2:
Create a complete business capability model for a hospital and health system, decomposed to Level 3. Give every capability at every level a one-sentence definition, and give every Level 2 capability two KPIs. Return only JSON in this shape: {"capabilities":[{"name":"","definition":"","children":[{"name":"","definition":"","kpis":["",""],"children":[{"name":"","definition":""}]}]}]}| Run | L1 | L2 | L3 | Defined | KPIs |
|---|---|---|---|---|---|
| Run 1 | 10 | 34 | 136 | 180 of 180 | 68 |
| Run 2 | 6 | 20 | 60 | 86 of 86 | 40 |
| Run 3 | 10 | 38 | 152 | 200 of 200 | 76 |
- It followed the instructions. Every capability in every run had a definition, every L2 had exactly two KPIs, and no name repeated within a run. As a first draft, each response was usable.
- It returned a different model each time. The three runs had 6 to 10 L1 capabilities, 20 to 38 L2 and 60 to 152 L3, from the same prompt.
- The names changed too. Any two runs shared only 3 to 4 L2 names exactly, out of 50 to 68 distinct names between them. Exact matching is strict, and many differences are rewordings of the same idea, but a team working from one shared map needs one wording.
- Nothing in any response said what was left out, and each run left out areas another one covered. Only one of the three had capabilities for clinical research or for philanthropy and the hospital foundation. Another had nothing on population health or virtual care, which the other two both included. Each read just as confidently as the others.
None of this makes generated drafts useless. It means a generated draft is a starting point to check, not a model to adopt: test it for completeness against the three checks on our hub page, fix one vocabulary, and version it before anyone plans with it. The three completeness checks.
One model, one prompt, one day. Other models, prompts and settings will give other results, which is the point: rerun the prompt and you get a different model. The scope here is a hospital and health system only, so these counts are not comparable with our Healthcare and Life Sciences package, which also covers pharmaceutical and medical device companies.
Questions
Can I build a capability map with an AI assistant?
You can draft one, quickly. Treat the draft as a starting point: the same prompt returns a different structure each time, and nothing in the answer shows what is missing. Check it for completeness, settle one wording for every capability, and version it before anyone plans with it.
How long does it take to build a business capability model from scratch?
It depends on the scope, the depth and how many groups have to agree. Drafting is the shortest part. Most of the time goes into reviewing the draft with the people who do the work and agreeing names, boundaries and definitions.
Should we buy a capability map or hire a consultant to build one?
Buy a reference model when you need a complete, defined starting point quickly and have people to tailor it. Hire a consultant when you need facilitation and leadership buy-in more than content, and ask which reference model they will start from. The two combine: a consultant can tailor a model you bought.
Is a free or generated capability map enough?
For a one-page sketch or a single workshop, often yes. For a program that will make funding, ownership or AI decisions on the map, you need definitions to L3, one stable wording and a version you can cite, whichever way you get there.