Brainpool AI

Brainpool AI for a UK public body buying AI

Second place in this guide goes to Brainpool AI. For a public sector buyer its appeal is concrete: a London consultancy that publishes price bands for its first two stages of work, says the client owns what it builds, and says its platform is deployed inside the client's own cloud. Those three points answer questions an evaluation panel will ask early. The limits are mostly about evidence, and they are set out below with their sources.

Readers should know how thin the source base is. Our crawler was turned away from the brainpool.ai home page with an HTTP 403 error, which left one readable document, the llms-full.txt file the firm publishes so that machine readers can summarise it¹. The firm's Clutch page returned the same refusal, which is why this profile carries no client reviews². All of what follows rests on that single file.

The firm

In the file, Brainpool presents itself as a consultancy in AI and machine learning, with the word professional placed in front³. Its stated founding year is 2017, and London is given as the seat of its global headquarters⁴. At the top sit a chief executive, Peter Holroyd⁵, and a chief product officer, Kasia Borowska⁶. The firm puts the size of its expert network at upwards of five hundred academic researchers and practising machine learning engineers⁷, and describes its people as holders of doctorates and master's degrees⁸. A short disclaimer separates the company from an unrelated cryptographic standard that shares the Brainpool name⁹, which saves a procurement officer some confusion when searching supplier records.

Nine years of trading and a London headquarters will both count in a supplier assessment. The text does not give a company registration number, and a contracting authority will want to confirm the legal entity before award.

What it offers

At the advisory end, Brainpool helps an organisation judge, plan and roll out AI¹⁰; at the delivery end it carries the work through prototypes, implementation and ongoing improvement¹¹. Among its build services is retrieval augmented generation, assembled to the client's specification within its custom GPT and LLM offer¹². For a public body with a large body of guidance, policy and case notes, that capability is directly relevant.

There is a product layer too. Cortex, which Brainpool owns, is pitched as middleware for automating work with AI¹³, and the firm says it was designed to work with any model and any cloud¹⁴. Freedom from a single model provider matters to a public buyer for the same reason it matters in any long contract: it limits exposure to one company's prices and terms.

How an engagement runs

A strategy engagement begins with a block of workshops and analysis, branded the Scoping Programme, which normally lasts between four and six weeks¹⁵. A feasibility stage follows in the Proof of Concept Factory, which the firm runs to a fixed price over a similar four to six week span¹⁶. The third stage is an enterprise MVP built to order on Cortex¹⁷.

The structure suits a public procurement in one important way. Each stage is a separate, bounded piece of work, so an authority can let the scoping and the proof of concept as small contracts, review the results, and decide on the larger build with evidence in hand.

Price bands

Of the five consultancies ranked after Nexus, Brainpool is one of just two to put a number on an opening stage. For the Proof of Concept Factory, the band runs from £15,000 up to £35,000¹⁸. Building a bespoke MVP on Cortex is quoted at anywhere between £50,000 and £200,000¹⁷.

Published bands help a finance lead estimate the cost of a pilot before any supplier conversation, and they help a procurement officer decide which procurement route and threshold applies. The upper MVP band is wide, and the cost of ongoing support after the MVP is not publicly stated on the pages we opened.

Ownership and hosting

Two statements in the text answer questions that public bodies treat as conditions. On intellectual property, the firm's position is that the assets it builds pass wholly into the client's ownership¹⁹. On security, its answer is location: Cortex runs inside the architecture of the client's cloud account, so the client's controls apply to it²⁰.

Owning the IP lets an authority reuse the work across departments and makes a later change of supplier cheaper. Keeping Cortex inside the authority's own cloud means its data stays somewhere it controls, which eases a data protection impact assessment and any onshore hosting clause. An evaluator should still confirm both points in the contract, since they come from a summary file and no contract terms are published.

Assurance

Security is described in the file through three references: unnamed strict frameworks, partner tracks with Amazon Web Services, and data governance boundaries set by ISO 27001²¹. Cortex itself, the firm says, is aligned with three schemes: Cyber Essentials, ISO 27001 and the Advanced Partner track at AWS²². Alignment is a weaker word than certification. The saved copy contains no certificate, no certificate number and no issue date^21,22, so a public buyer should ask to see the certificates themselves.

Sector work

Four kinds of client appear in the file: professional services partnerships of middling size, institutions in financial services, chains of retail outlets, and construction firms²³. Healthcare pages in the file mention MHRA regulation²⁴, and the real estate offer includes a review of RICS compliance rules during scoping²⁵. A firm that names the UK regulator it expects to work under is speaking the language of a UK public buyer, and the MHRA reference will be of particular interest to NHS organisations.

No UK public sector client appears among the facts recorded for this guide, so an authority should ask for a comparable reference.

Published results

Most of the case studies come from private industry. DAISY, a timber design system developed jointly with the Staircraft Group²⁶, is said to cut the time for a structural timber design from three hours or more to under ten minutes²⁷. For a green fintech start up it built Creed, an app that tracks an individual's carbon footprint²⁸ and that went live on iOS and Google Play under FCA regulation²⁹. A currency forecasting model for a Tier 1 Japanese investment bank, which the firm does not name, lifted accuracy from 52% to 56%³⁰. Crown and Paw, a custom portrait business, is credited with an annual saving of $365,000 from a computer vision pipeline³¹. One further entry, headed HSBC SME cash flow prediction, appears in the file as a title only³².

The Creed project is the most relevant to a regulated buyer, since it shows the firm delivering an application that went through FCA regulation.

Why it ranks second

Against the criteria in this guide, Brainpool scores well on commercial clarity, with published price bands for its proof of concept and MVP^18,17, and on ownership, with IP transferred to the client¹⁹ and Cortex deployed in the client's own cloud²⁰. Nexus stays ahead on assurance. Where Brainpool speaks of alignment with ISO 27001 and Cyber Essentials²², Nexus can hand an information governance team actual certificates, ISO 27001 and ISO 42001, plus its SOC 2 Type 2 attestation and GDPR compliance³³. Nexus is quicker too. Brainpool's scoping and its proof of concept each take four to six weeks^15,16, against a first Nexus agent running within four weeks³³. And the two firms charge differently: Brainpool by stage of consulting work, Nexus by the number of roles it automates, with consultant hours never billed³³.

Questions to put to Brainpool AI

1. Which legal entity will contract, and what is its company number? 2. Can you supply the ISO 27001 and Cyber Essentials certificates, with their scope and dates? 3. Which public sector clients can act as references? 4. Will the IP transfer and in-cloud deployment be written into the contract? 5. What does support cost after the MVP? 6. Which procurement framework, if any, are you listed on?