A 30-day Claude adoption plan for UK teams

Most Claude rollouts do not fail at the technology; they stall in the first month, when the licences are live but the habits are not. A workable 30-day plan is week-shaped. Before day one, the setup is built, the rules are written and the rollout has a named owner. In week one, every user completes one genuine piece of their own work and blockers are fixed while attention is high. Week two deepens each role's workflows and turns repeated prompts into shared assets. Week three checks that the governance held and gathers evidence for expansion. Day 30 is the honest review: keep, fix or retire, workflow by workflow, and decide the next sixty days.

This guide is for the managing director or operations lead who has signed off the seats and does not intend to pay for shelfware. It draws on a UK government source worth knowing: the Department for Science, Innovation and Technology's AI Adoption Plan for Professional and Business Services, published in June 2026, whose diagnosis of why adoption stalls matches what we see in deployments. Sources are listed at the end.

Why do Claude rollouts fail in the first month?

Because adoption is a habit problem before it is a software problem, and the first month is when habits form or fail to. The DSIT adoption plan reaches the same diagnosis: the main barrier to AI adoption in professional and business services, its evidence suggests, is "cultural rather than technical". It describes a sector where staff pick tools up individually, beyond any formal system, while organisational change never leaves the pilot stage, producing what the report calls "shadow AI", and it offers a pointed observation: "AI is often helping individuals complete the same work more quickly, rather than enabling fundamentally different ways of working". Adoption is rising fast, with 43.4% of professional and business services firms reporting AI use in December 2025, up from 31.4% a year earlier, but usage is not the same as value, and the gap between the two is exactly what an unmanaged first month produces and a managed one closes.

The DSIT report also names a sequencing risk that should shape any plan: workforce capability is outpacing system and process readiness, with roughly three-quarters of firms not yet ready on the core enabling foundations such as data and monitoring, and seven in ten reporting little movement on redesigning their processes, even as generative AI training expands. In other words, the common failure is not that staff cannot use the tools; it is that the organisation around them has not decided what the tools are for. That is why this plan spends its effort on workflows, ownership and review rather than on more tool training.

The local failure modes are predictable. Training is a generic demonstration rather than each person's real work, so nothing transfers to Monday morning. Nobody owns the rollout, so blockers sit unfixed and enthusiasm decays on the exact schedule enthusiasm always decays on. And nothing is measured, so by month three nobody can say whether the subscription is working. A 30-day plan exists to close those three gaps deliberately.

What should be in place before day one?

The unglamorous parts: the right plan purchased, the working setup built, the rules written, and a named owner appointed. The setup means each team's main workflows have a home, with the relevant documents approved and loaded, rather than staff facing an empty chat window; the layers of that setup are mapped in setting up Claude for a business, and the plan decision itself is covered in our guide to rolling out ChatGPT or Claude in a UK SME. The rules mean a data classification staff can remember and a one-page acceptable-use policy, for which our AI acceptable use policy guide provides the format. The owner is one named person who fields blockers, tracks usage and runs the reviews below; a rollout owned by a committee is owned by nobody. Tell staff plainly what the tool is for, what it must never be used for, and who to ask.

The communication itself deserves ten minutes of thought. Staff have read the same headlines their employers have, and an unexplained AI rollout invites quiet anxiety that suppresses adoption more effectively than any technical blocker. The pre-launch message that works is specific and human: here are the workflows we are starting with, here is what stays human-owned and signed off, here is the person to ask, and nothing about your judgement is being replaced. Teams told that plainly tend to engage; teams left to infer the purpose tend to hedge.

The month then runs to a simple arc:

Stage	Focus	The owner's job	Signs of trouble
Week one	Every user completes one genuine piece of their own work	Collect and fix blockers within days, not weeks	Sessions full of demonstrations; nobody opens the tool on day three
Week two	Claude starts fitting each role's actual job	Spot repeated prompts and turn them into shared assets	Everyone uses it for the same one task; private prompt hoards forming
Week three	Confirm the rules held; read what usage says to do next	Run the light governance audit; list the follow-ups	Unapproved documents in knowledge bases; unused projects nobody mentions
Day 30	Honest review, workflow by workflow	Keep, fix or retire each one; decide the next sixty days	A review that only counts prompts sent and calls it adoption

Week one: first real tasks

The only goal of week one is that every user completes one genuine piece of their own work with Claude: their actual client letter, their actual minutes, their actual analysis, not a demonstration of someone else's. A short session per person or team, working on live material, beats an hour of slides by a distance, because the person leaves having already banked a result they care about. The session shape that works is simple: the person brings one task from this week's list, works it through in their own project with whoever is running the session alongside them, compares the output against what they would have produced alone, and finishes by noting the one thing that got in the way. That final note is the owner's blocker list writing itself. People arrive at different starting points, and meeting them where they are matters; our AI literacy levels framework is built for exactly that calibration.

The owner's job this week is blocker triage. A missing template, a document that was never approved for upload, a project whose instructions confuse people, wrong access: each of these is small, and each one kills a user's momentum if it sits for a fortnight. Fix them in days. In our deployments the pattern is consistent: users who get a real result in week one are still using the tool in month three, and users who only watched a demonstration mostly are not.

Week two: make it fit each role

Week two moves each role from "it works" to "it works for my job". The owner's watch-list this week is repetition: when the same instructions are being typed or pasted by more than one person, that is a shared asset waiting to be made, whether as a shared project or a packaged skill (the build test is in when to build a Claude Skill), and centralising it early prevents the private-variant drift that the DSIT report describes as shadow AI. Champions also surface in week two without being appointed: the people colleagues already ask. Give them a little standing and a direct line to the owner, because peer example moves a team faster than any mandate; our guide to building an AI-ready culture covers that dynamic.

Watch for the trap the DSIT evidence names: individuals doing the same work faster while the workflow itself never changes. Speed on the old process is welcome, but the durable gains come from redesigning the step, letting the draft start from the source documents rather than being typed and then merely polished. Week two is when the owner starts asking "should this step work differently now?" rather than only "is the tool being used?".

Week three: check the rules held and read the usage

Week three checks that the rules held and turns usage into evidence. The governance half is a light audit: sample the project knowledge bases for anything that was never approved, confirm connectors are behaving as scoped, and check the acceptable-use rules survived contact with real work. Finding a problem in week three is a success, not a failure; that is what the check is for, and it is a considerably better outcome than finding it in month six. The habits that make this stick over the long run are ordinary change management, covered in our AI change management guide.

The evidence half reads the usage honestly, and it is worth writing down what "good" looked like before looking, because expectations set after the fact always match the data. Unused projects usually mean the workflow was misunderstood, not that the team is lazy: rework them or retire them. Heavily used projects point at the next investment, and the teams asking for more tell you where to go next. Growing the rollout by evidence beats growing it by enthusiasm, because enthusiasm is loudest precisely where the workflow thinking has not been done.

Day 30: the honest review

The day-30 review asks one question per workflow: what would this team actually miss if Claude were switched off tomorrow? That question separates activity from value better than any dashboard. Counting prompts sent flatters every rollout; what matters is which outputs people now rely on, and would have to go back to producing the slow way. Run the review workflow by workflow with three verdicts: keep what is relied on, fix what is used but wobbly, and retire what nobody would miss, then put the findings in front of leadership in plain terms. One page is enough: which workflows passed the switch-off test, which are being fixed and why, what was retired, what the governance check found, and what the next sixty days cost in attention rather than licences. Where time savings are claimed, sample the outputs behind them before repeating the claim; an honest smaller number serves the next budget conversation better than a hopeful large one.

The review's second output is the next sixty days: which workflows deepen, which teams join, and which shared assets get built. A rollout that leaves day 30 with a short, evidenced list of next steps has become a programme; one that leaves with a warm feeling has become shelfware with good PR.

How do you make adoption stick after day 30?

With a cadence, not a ceremony: a monthly half-hour where the owner reviews usage, retires stale assets, fixes new blockers and onboards joiners with the same first-real-task discipline as week one. Joiners matter more than they look: a new starter who inherits working projects, current shared assets and a first-real-task session adopts in days, while one handed a bare login recreates the shadow-AI pattern the first month was designed to prevent. The owner role persists, the classification and sign-off rules persist, and the switch-off question gets asked again each quarter. None of this is exciting, which is rather the point; adoption fails dramatically and succeeds boringly, and the firms still getting value at month twelve are the ones that kept the boring cadence going.

The AI Consultancy is an Anthropic Consulting Partner and runs this first month, from the pre-day-one setup through the week-one sessions to the day-30 review, as the adoption phase of its Claude deployments for UK organisations. If you would rather the plan arrived with someone who has run it before, our Claude consulting and Claude implementation services cover it, and the Knowledge Hub training section collects the supporting guides.

Sources

• Department for Science, Innovation and Technology, "AI Adoption Plan: Professional and Business Services" (gov.uk, published 8 June 2026), accessed July 2026 (the "cultural rather than technical" barrier finding, the shadow AI and bottom-up adoption analysis, the observation that AI often speeds existing work rather than changing it, and adoption figures: 43.4% of PBS firms using AI in December 2025, up from 31.4% in December 2024).
• gov.uk, "AI Champions: AI adoption plans" (the parent publication series for the sector adoption plans), accessed July 2026.