Enterprise · Step 2
One working AI system, in your environment, in six to eight weeks.
Our senior team builds the first project from your roadmap and takes it to production use: integrated with your systems, measured against a bar agreed up front, and documented so your team can run it.
- Duration
- 6–8 weeks
- Commercials
- Fixed scope, fixed fee
- Team
- Senior AI architects and engineers
- Where it runs
- Your cloud or private environment
Is this you?
A Sprint fits when…
- You have one use case with a business owner who wants it and data you are allowed to use.
- A proof of concept exists and has been “nearly ready” for months.
- Your team is strong but has not shipped an LLM system into production before.
- Risk and security need to see guardrails, logging and human review designed in, not added later.
- You want a result this quarter that makes the case for the larger program.
What you get
What a Sprint builds
One of these, chosen in the assessment or in a short scoping phase. Narrow on purpose: a first win should be small enough to finish and real enough to matter.
A grounded knowledge assistant
Answers staff or customer questions from your own documents and data, cites where each answer came from, and is designed to decline when the sources do not cover the question. Built with retrieval or a knowledge graph (GraphRAG) where relationships matter.
A private LLM deployment
A model running inside your network boundary, with access control, logging and usage policies, so teams can use generative AI without sending sensitive data to a public service.
An automated workflow with a human in the loop
An agent that reads, classifies, drafts or routes work, such as claims intake, document review or service requests, with people approving the steps that carry risk.
A machine-learning score inside a process
A model that ranks or flags cases (risk, priority, likely outcome) where your team already makes the decision, with monitoring for drift.
How it works
Week by week
Week 0
Scope lock
We write down what the system does, the data it may use, the acceptance bar, and who signs off. If you came through an assessment, this already exists.
Weeks 1–2
Foundations
Environment access, data pipelines, the evaluation set and the first end-to-end slice. You see something working by the end of week two, even if it is rough.
Weeks 3–5
Build and evaluate
Weekly demos. Each one is scored against the evaluation set, including questions the system should refuse, so progress is a measurement and not an impression.
Weeks 6–8
Harden and hand over
Guardrails, access control, logging, monitoring and runbooks. Your team is trained to operate it. We close with a value report: what was built, what was measured, what to do next.
Clear boundaries
What is and isn’t in a Sprint
Included
- Architecture, build and integration for one agreed use case
- An evaluation set written before the build, and scores against it each week
- Guardrails, access control, logging and human-review steps
- Deployment into your environment with documentation and runbooks
- Handover training for the team that will operate it
- A closing value report with measured results and recommended next steps
Not included
- Several use cases at once
- Rebuilding core platforms or large data migrations
- A promise of a specific accuracy figure before we have measured it
- Compliance certification or legal sign-off, which stay with your teams
AI systems can be wrong. We design for that with evaluation, refusal behaviour and human review, and we report what we measured rather than promise a number in advance.
Questions
What people ask us first
Can we start with a Sprint and skip the assessment?
Yes, if the use case is already chosen, has an owner and has data behind it. We run a one-week scoping phase first to lock the acceptance bar. If the use case turns out to be the wrong first project, we say so before the build starts.
Where does our data go?
It stays in your environment. We build inside your cloud account or private network, under your access controls. Which models are used, and whether any data leaves your boundary, is a design decision you approve in week zero.
How do you deal with wrong or made-up answers?
By designing and measuring for it. Answers are grounded in your sources and cite them, the system is tested on questions it should decline, and risky actions go to a person for approval. We report the measured results; we do not claim errors can be eliminated.
What do you need from our engineers?
Access, a named technical contact, and ideally one engineer who pairs with us part-time so the knowledge stays in your team. We work in your repositories and your delivery tools.
What happens when the Sprint ends?
Your team can run it on their own from the runbooks. From there you can continue with a program of further sprints, or with AI Sustain for monitoring and steady improvement.
Next step
Talk about a Sprint
Describe the use case in a sentence or two and where it stands today. We aim to reply within one business day.
