Not every AI asset earns the same scrutiny. A skill published for internal use and a product sold under licence carry very different risk, so we hold them to different bars. This page sets out, in working checklists, exactly what the AI Design Authority expects of a Demonstrator, an internal Claude Skill, an Accelerator, and a Product — and how the asset's tier, and where it lives in Beacon, raise or lower that bar.
Asset type — Demonstrator, internal Claude Skill, Accelerator, Product — sets the baseline depth of what we expect. It reflects maturity and exposure: how far the asset is built to be reused, sold, and maintained.
Tier 1–4 — the AI Asset Taxonomy, from AI for Us to AI from Us — then flexes that bar up or down. The closer an asset gets to the client and to standalone revenue, the more of our recommended expectations become mandatory.
Keep three lenses distinct. Asset type and tier describe the asset. Risk Class 1–3 describes how much harm it could do, and sets how deeply we review. A low-risk product still gets the full product checklist; a high-risk demonstrator gets a heavier review than its type alone implies.
At Step 4 · Approvals, an asset is cleared across four lenses. The Authority owns the technology lens. These checklists are our expectations for that lens — drawn from the four standards pillars — plus the qualifying criteria an asset must meet before it is worth reviewing at all. Content, commercial, and marketplace-readiness lenses sit with the Product Owner, sponsor, and legal.
The Authority is custodian of both homes in Beacon, our single AI portal: the AI Library, where internal Claude skills are reviewed before they are published for consultants, and the Asset Exchange, the marketplace where client-ready accelerators and products are listed for our people to discover and take to clients.
Each item is marked Mandatory — met or formally waived before approval — or Recommended — expected good practice, deviate with reasoning recorded.
The same asset type carries different weight depending on its tier. This is the shape of it — the checklists below make each cell concrete.
| Asset type | Tier 1–2 · AI for Us (internal) | Tier 3 · Client-facing IP offering | Tier 4 · Productized advisory |
|---|---|---|---|
| Demonstrator | Light — self-certified floors | Client-setting data & confidentiality rules apply | — not a product path — |
| Internal Claude Skill | AI Library review before publication | — graduates to accelerator — | — n/a — |
| Accelerator | — internal reuse → see Skills — | Client-ready — listed on the Asset Exchange; full gate plus security, IP, marketplace readiness | — promote to product — |
| Product | — rare — | Full gate + run & commercial controls | Highest bar — full Cyber Approval + ASLM gates |
Light Full technology gate Full gate, raised for client exposure & revenue
Skills live in the AI Library; client-ready accelerators and products live in the Asset Exchange — both inside Beacon.
A quick, local build used once — often vibe-coded — to prove a use case live in a client conversation or an internal test. Not maintained, not reused, not billed. Its job is to make an idea tangible in the room.
Typical tier: 1–2 (AI for Us). The Authority's stance: lightest touch — self-certified against a short set of hard floors. No formal gate, because nothing here is meant to survive the meeting.
The floors are non-negotiable. A demonstrator is light precisely because it stays disposable. The moment it runs on real data, gets shared, or heads toward reuse, it is no longer a demonstrator — re-enter it at Prioritization as an accelerator candidate and apply that checklist.
How the tier shifts this. A demonstrator shown in a client setting — edging toward Tier 3 — inherits the client environment's data-handling and confidentiality rules; treat the client's AI guidelines as taking precedence. An internal-only Tier 1 demo stays at the lightest bar.
A packaged, reusable Claude skill that encodes a Capgemini Invent method or workflow, published to the AI Library in Beacon for consultants to use during delivery. Internal — it runs on our enterprise instances and is not sold to clients. This is where much of our day-to-day productivity gain is codified.
Typical tier: 1–2 (AI for Us). The Authority's stance: we own the review. Every submitted skill is QA'd against the checklist below before it is published to the Library — lighter than a client product, but real, because one skill is reused across the firm and can touch client data in delivery.
The AI Library review. Skills reach the Library through a submission-and-review process the Authority runs: a consultant submits a skill, we QA it against this checklist, and only then is it published for others to use. A skill that grows into something client-facing or sold is no longer a Library skill — it re-enters the process as an Accelerator, bound for the Asset Exchange.
If it fails these, it is a personal script, not a firm-wide skill — keep it local.
How the tier shifts this. A Tier 1 foundational skill — part of Invent Essentials, taught to every consultant — is held to the bar above at its lightest. A Tier 2 domain-specific skill that encodes one practice's or industry's know-how raises the content-accuracy and IP bar: the review leans harder on whether the method is right, not just whether the skill runs. The moment a skill is meant for a client's hands or for sale, it leaves the Library and re-enters as an Accelerator, bound for the Asset Exchange.
A pre-packaged, reusable component with real IP at its core that speeds up selling and delivering to clients. Client-ready and listed on the Asset Exchange, reused across engagements — proven, supported, and usable by someone who did not build it. Not billed as a separate line item; its value is blended into the engagement.
Typical tier: 3 (AI from Us). The Authority's stance: the full technology gate. This is the core of what we approve for client use, because reuse in front of clients multiplies both the value and the risk.
If it fails these, it is a demonstrator — send it back before spending a gate on it.
How the tier shifts this. An accelerator is a Tier 3 client-facing offering by definition — listed on the Asset Exchange in Beacon — so the client-exposure bar always applies: client-environment security, demonstrable IP defensibility, and marketplace readiness (licence & T&Cs), with accessibility and brand alignment mandatory and the Cyber Approval Process triggered before it is listed. Something reused only inside delivery is not an accelerator — it belongs in the AI Library as an internal Claude skill. If it grows into a standalone sold offer, it graduates to a Product.
A fully maintained, commercialised standalone thing the client buys — via licence, subscription, or usage-based pricing. The client pays for the output or access itself, not for a bespoke build. Owned by a product function with its own roadmap, run team, and lifecycle.
Typical tier: 4 (AI from Us). The Authority's stance: the highest bar — everything expected of an accelerator, plus productization, run, commercial, and external-exposure controls.
Start from the accelerator checklist. A product must meet every mandatory accelerator expectation first. The items below are what a product adds on top — the difference between something reused inside delivery and something sold and run as a standalone offer.
How the tier shifts this. A product is Tier 4 by definition, so the tier axis mostly stops flexing here — the variation is the deployment target. Own infrastructure, a hyperscaler marketplace listing, and deployment into a client environment each raise their own residency, tenancy, and listing obligations on top of the checklist above.
One thing we will not do: approve an asset for reuse or sale on the strength of a good demo alone. A demonstrator that impresses in the room is not an accelerator — without the moat, the clean documented build, and an owner who can hand it over, reuse just spreads fragile, unmaintained code across engagements. If it has not met the qualifying criteria, we send it back rather than wave it through. Route exceptions through Governance & Waivers.
The readiness checklist mirrors these expectations and tells you exactly what to close before a review.
