The Bristol Practice · Private GP · 07 Jul 2026 · GM-CQC-2026-BP · v2
Prepared for Dr Charlie Marshall & Dr Henry Hardy · The Bristol Practice · following our call, 7 July 2026
What we heard on Tuesday — and a two-month path to evidence that runs itself
Four GPs, a membership list of ~40 and growing, and a disproportionate share of founder time going to the regulatory side: policies done, audits and analytics now starting, and reports that Semble makes slow to get out. This page turns our conversation into a concrete scope, the terms we quoted, and next steps — written so you can review it with Henry.
The 30-second read
You told us
Regulatory work takes a disproportionate share of time. Getting audit and analytics data out of Semble is time-consuming. Much of the CQC evidence already exists — it needs tagging to the categories CQC wants. The same Claude/Gemini prompts are re-run by hand. Data handling must be airtight.
We propose
A two-month starter that stands up your AI Foundation — a private environment in your own UK cloud that runs AI requests internally — plus a Semble data connector, your existing evidence mapped to CQC's quality statements, and your repeated prompts turned into scheduled, anonymised workflows.
Terms quoted
$5,000/month for ≈100 hours (≈£3,700 at today's rate) during the starter; afterwards scale to need (ongoing support from ≈£1,000/month unless we're building new features). One month's notice either side. No software to license — expertise plus infrastructure you own.
Next step
Review with Henry. Alexey is in the Bristol area in about two weeks — happy to talk it through in person over a coffee.
Of the 861 GP practices holding a published rating under CQC's current framework in our dataset, 723 — 84% — are Good. A first assessment lands against a known, and high, bar.
What pulls practices under
2×
Safe (133 practices) and well-led (130) sit below Good more than twice as often as any other key question. Both are evidence-and-oversight disciplines — governance, not clinical surprises.
Direction of travel
Dec2026
As we discussed on the call: the CQC changes landing at the end of the year put still more weight on evidence that is collected and acted on continuously — pipelines, not binders assembled the week of the visit. We wrote up what changes in 2026 →
The CQC application is done; the work now is audits and data analytics as the patient list grows. In your words, it “takes up a lot of brain space” that belongs on clinical care and building the business.
The data bottleneck
Semble holds the data; reports are slow to produce
Clinical records live in Semble, policies in Practice Index. Extracting data for audits and CQC reporting is time-consuming — and much of the evidence already exists; it needs tagging to the categories CQC asks for.
The constraint we build to
Airtight data, anonymised in processing
The less that leaves, the better. Everything below runs inside your own UK-region cloud environment (Azure or AWS) — you hold the access; identifiable Semble data is minimised and anonymised before analysis; nothing trains anyone else's models.
One thing this proposal deliberately does not lean on: AI note-taking in consultations. You told us documenting is part of your clinical thinking, and that today's generated notes read wordy — so nothing here depends on scribing. The same foundation supports it later if any of the GPs wants it (one already uses Heidi), but it is not the plan.
The two-month starter — proposed scope
6 workstreams · sized for a founder-led practice
Each item traces to something you named on the call. The aim of the first two months is visible return — reports you stop producing by hand — plus a foundation every later use case reuses.
01
AI Foundation — a private environment in your own Azure or AWS space (UK region) that runs AI requests internally; you hold access and keys; aligned with GDPR and NHS DSPT by construction
Month 1Foundation
02
Semble connector — the extraction bottleneck removed. We built the same connector for another partner clinic, surfacing Semble data to their AI of choice
Month 1Proven before
03
Evidence map — inventory what you already have (Semble, Practice Index, policy drive) and tag it to CQC's quality statements, so gaps are visible and the pack is assessor-ready
Month 1–2Your words: “tag it in the right place”
04
Prompt workflows — the Claude/Gemini prompts you re-run by hand become scheduled, anonymised workflows on your infrastructure, with the model chosen per task and per cost
Month 2Quick ROI
05
Audit & analytics reports — recurring clinical-quality and audit reporting generated from Semble data: for evidence-based practice first, compliance as the byproduct
Month 2Recurring value
06
Monthly evidence report — collected automatically, actioned visibly, verifiable whenever CQC arrives; doubles as your internal governance review
After the starter, the engagement scales to what you actually need — same named team stays attached, with a weekly check-in with Nadya Tolmacheva, our UK delivery lead.
We've built this before
3 live case studies · greenm.io
Three published projects that map onto this scope — click through for the full write-ups.
ROC Clinic · private clinic group
Unified data platform — reporting without the scramble
One data layer across clinical and operational data: manual reconciliation gone, real-time reporting, secure exports. The pattern behind workstreams 02 + 05.
“Month-end stopped being a scramble. The team now self-serves most answers.”
AI letter-quality assessment with full auditability
AI review of clinical correspondence with clear metrics and governance, built to NHS security standards — the same airtight constraint workstreams 01 + 04 are built to.
“The AI-powered assessment has already helped standardize communication and improve review efficiency.”
$5,000 per month for ≈100 hours of engineering and delivery time, for the first two months — ≈£3,700 at today's exchange rate. On the call we rounded to £3,500; billing is in USD, so the exact GBP figure moves with the rate.
After that
Ongoing support runs ≈£1,000 per month; it rises only when we're actively building new features together. Hours adjust with one month's notice.
Commitment
One month's notice, either side. Engagements typically run on a 6-month frame once rolling — but nothing locks you in beyond the notice period.
What you keep
Everything. The environment, the connectors, the workflows and the evidence base are built in your accounts — if we part ways, it all stays with the practice.
To recap
A two-month starter that stands up your AI Foundation — a private environment in your own UK cloud, running AI requests internally — connects Semble, maps the evidence you already have to CQC's quality statements, and turns the prompts and reports you run by hand into scheduled, anonymised workflows. $5,000 per month for ≈100 hours (≈£3,700 today), one month's notice either side, and everything we build sits in your accounts, so it stays with the practice. After the starter, ongoing support runs from ≈£1,000/month unless we're building new features together. Alexey is in the Bristol area in about two weeks — a coffee with you and Henry works as well as a call.
