Decide
- AI Strategy & Roadmap
- Use-Case Portfolio & Business Case
- AI Readiness & Governance Assessment / Design
- Process Redesign
The Claude Partner Network has 16 service categories. Most buyers start at the wrong one.
Pexon Group is a Certified partner in Anthropic's Claude Partner Network, Services Track. The track defines sixteen service categories, which sort into four jobs: decide, build, activate, operate. Most companies buy a proof of concept first, then stall on identity, tenancy and cost, none of which a proof of concept answers.
Our position
The cheapest thing to approve is the wrong place to start
A proof of concept is the easiest Claude engagement to get signed, so it is the one most companies buy first. It answers whether the model can do the task — which, in 2026, is almost never the open question. The questions that actually stall the rollout are whose permissions a request carries, which tenant the inference runs in, and which team the tokens are billed to. None of the three is a prototype question, and all three outlive the prototype. None of the three is a prototype question.
The catalogue
Sixteen categories, four jobs
This is the Services Track vocabulary, in Anthropic's own wording so you can match it against their list line for line. The grouping is ours: the categories are published as a flat set, and flat is how they get bought in the wrong order.
Build
- Build Prototype / POC
- Internal Agentic Platform Build
- Customer-Facing AI Product Build
- Connectors, plugins and API integrations
- Industry Solution Accelerator Deployment
Activate
- Rollout & Activation
- Claude Training Sessions
- Platform Migration to Claude
Operate
- Managed AI Operations
- Sustained Adoption / CoE Program
- Ongoing Cowork / Claude Code Scaled User Support
- Claude Consumption / FinOps Optimisation
What each job is for
The question behind each of the four
| Job | The question it answers | What skipping it costs |
|---|---|---|
| Decide | Which use cases, in which order, and who signs off on identity, tenancy and budget? | A roadmap with outcomes and no named owners. The pilot expands into three approvals nobody scheduled. |
| Build | What gets built, against which internal system, and who operates it afterwards? | A prototype that proves the model works and is thrown away because nothing around it was built. |
| Activate | Who gets a seat, under whose permissions, and what are they trained to stop doing? | Licences bought and unused. Adoption measured in logins rather than in work that changed shape. |
| Operate | Who is on call, who reads the invoice, and which team is each token billed to? | One line in a cloud invoice with no team behind it, and no argument available for year two. |
Read the right-hand column first. Each of those four outcomes is a thing we have watched happen, and in every case the category that would have prevented it was on the list and was not bought, because it was harder to approve than the one that was.
Where each category lands
Sixteen categories, and where each one lands
Sixteen categories do not need sixteen URLs, and a page per vendor category would be a sitemap rather than a portfolio. Several of the sixteen are the same engagement described at a different granularity, so they are sections inside one page. What follows is the honest mapping, including the three categories that deliberately have no page of their own and the reason for each.
Decide. Enterprise AI strategy and roadmap carries both the strategy category and the use-case portfolio and business case, because a portfolio without the arithmetic under it is a wish list and the two are never sold apart. The readiness and governance assessment is EU AI Act readiness, which is the form that question takes for a company deploying in Europe. AI business process automation is the process redesign category, and it is in the decide group rather than the build group for a reason the page argues at length: the redesign is the deliverable, and the automation is what you do once it is settled.
Build. Agentic AI platform is the internal platform build and the customer-facing product build in one page, split into two sections, because the architecture is substantially the same and the thing that changes is who carries the risk when it is wrong. AI proof of concept is the prototype category, written around the decision most POC briefs skip. MCP server development and the connector catalogue cover connectors, plugins and API integrations across ninety source systems. The industry solution accelerator category has no page, and that is deliberate: nine landing pages differing only in the sector noun is thin content on a domain with this much authority, so the customer references do that job instead, because they describe work that actually happened.
Activate. Rollout and activation splits by blast radius rather than by product: Claude Code enterprise rollout is a permission problem because it touches repositories and secrets, while Cowork rollout and adoption is a data classification problem because it touches mailboxes and documents. Claude Code training is the training category. Platform migration also has no page of its own — the search demand for it is close to zero and the work itself is two existing pages, the LLM gateway, which is how you move between providers without a rewrite, and Claude in your own cloud tenant, which is where you are moving it to.
Operate. Managed AI operations is the third category with no page of its own, and the reason is different from the other two: it is not a product, it is an operating tier. What it costs is on the pricing page as a published monthly ladder, from monitoring only up to three forward deployed engineers, and what it covers depends entirely on what was built in one of the categories above. AI center of excellence carries both the sustained adoption programme and the ongoing scaled user support, since a support tier with no charter above it becomes a mailing list within a quarter. Claude cost optimisation is the consumption and FinOps category, and it is the one engagement on this list that pays for itself in an invoice you can read afterwards.
Standing
Where the partnership comes from
Pexon Group is a Certified partner in Anthropic's Claude Partner Network (Services Track), granted on 1 October 2026. This site is published by PEXON ROMANIA S.R.L., a separate legal entity in the same group, and the party you contract with here.
What partner status does not tell you
It is a statement about a delivery practice, not about your engagement. Standing in the Services Track is assembled from counts Anthropic can verify — people holding current certifications, customers in production, published stories — and it is re-checked quarterly, so it is a fact with a date on it rather than a permanent property. It says nothing about whether the architecture we would propose for you is right, and it is worth less to you than the two pages on this site that show their working: the gateway page, which publishes the routing decision, and the cost pages, which publish the arithmetic. Buy the architecture call, not the badge. If the badge is the deciding factor, you are choosing on the weakest available evidence, and we would rather say so here than have you work it out later.
What a buyer asks about partner status
What does the Claude Partner Network actually certify?
Standing in the Services Track is assembled from things Anthropic can count: individuals holding current certifications, customers running Claude in production, and published customer stories. It is a statement about a delivery practice rather than about any one engagement, and Anthropic re-verifies it each quarter, so it can change without the partner doing anything.
Is Pexon in Anthropic's public partner directory?
Not yet. The directory entry and the co-marketing materials arrive with the next tier up, not with Certified. We would rather write that sentence than let you discover it yourself: the partnership is real and granted in writing, and the directory listing is a separate thing that follows later.
Which of the 16 categories should a first engagement be?
Almost never a prototype, which is what most companies ask for because it is the cheapest thing to get approved. A prototype answers whether the model can do the task, and that is rarely the open question. The readiness assessment and the use-case portfolio are the two that unblock everything after them.
Do we have to buy the whole catalogue?
No, and the catalogue is not a bundle. Sixteen categories is the vocabulary Anthropic uses to describe what partners deliver, not a sequence anybody works through. A first engagement here is the two-week Readiness Blueprint at €4,900, which decides which of the sixteen you actually need. All prices are net and exclude VAT.
Does partner status change what Claude costs us?
Not the model bill. Consumption is billed to you directly by Anthropic, AWS or Microsoft at their published rates — we do not resell tokens, and the invoice does not pass through us. What a partner changes is how quickly the architecture around it gets decided, and whether per-team attribution exists before the first invoice arrives rather than after.
What if we already have an AI strategy?
Then skip the strategy category and say so at the start. The common version of this is a strategy document that names outcomes but no owners for identity, tenancy and budget, in which case the gap is the governance assessment rather than another roadmap. If the owners are already named, go straight to the platform or connector work.
How to start
Not a sales call. An architecture call.
Thirty minutes with the architect who would actually run the engagement.