P8 / AI

AI feature feasibility sprint

A fixed-price, fixed-scope sprint for teams with an AI feature in mind and no proof yet that it can be made reliable or economic — ending in a decision, not a deck.

From £4,500

Last updated:

An AI feature feasibility sprint is a fixed-scope assessment of one proposed LLM feature or workflow before a build is committed: candidate models and architectures are compared against a representative sample of your data, an initial evaluation set scores them, and a working spike settles the riskiest technical question where code is the only honest answer. Revenant Systems delivers a go, no-go, or change-scope recommendation with a recommended production design and an implementation estimate — from £4,500.

Who is the sprint for?

The sprint fits teams that know where an LLM could help but not whether the feature can be made reliable or economic: a bounded workflow in mind, representative data available, and a build decision waiting on evidence. It suits a first AI feature and a proposed addition to an existing product equally — the question is the same.

Who is it not for?

Teams that have already built the feature: the AI production-readiness audit assesses an existing prototype or live feature, and the sprint would repeat work already done. It is also the wrong fit for an organisation-wide AI opportunity review — the sprint answers one engineering question about one feature, in depth.

What does the sprint cover?

One tightly bounded feature or workflow, taken from discovery to a recommendation: suitable model and architecture approaches compared against your sample data, a small evaluation set that scores them, a working spike where one is needed, and estimates of production cost and latency, with security and data-handling observations recorded as they arise.

Model candidates come from Anthropic, OpenAI, Google, and OpenRouter, with local models through LMStudio and llama.cpp where UK data residency rules a hosted API out — judged on your data rather than a public benchmark.

What answer does the sprint end with?

A recommendation you can act on: build it as designed, do not build it, or build a different version of it. Each arrives with the evidence behind it — evaluation scores, cost and latency numbers, the recommended production design — and an implementation scope and budget range, so a yes can proceed without a second scoping exercise.

Discuss your feature idea

UK company and contracting entity · UK GDPR-aware delivery · Vendor-neutral model selection · Private and self-hosted deployment options

What you receive

How the engagement runs

What we need from you

Access and client data are handled under our information security statement.

FAQ

Frequently asked questions

What if the answer is no?

Then the sprint has done its job. A no arrives with the evidence — where quality fell short, what being wrong would cost — and it costs a fraction of discovering the same thing after a build. You keep the evaluation set and the findings either way.

Is the spike production code?

No — a spike exists to settle the riskiest question quickly, and it is written for that. What carries forward into a build is the evaluation set, the model choice, and the recommended design; the build stage then produces code engineered for production.

How is this different from the AI production-readiness audit?

Timing. The sprint runs before anything is built and answers whether it should be; the audit runs after a feature exists and answers whether it will survive production. A sprint that ends in a build tends to meet the audit's checklist on the way through, because the same engineer applies the same standard.

Each package is an audit or assessment offered on its own — the process behind it is covered in how we work.

Have a feature in mind and no proof yet? Let's talk.

Get in touch

Your message is read by the engineer who would scope the work; the reply is a short technical conversation.