AI consulting that ends in a decision
Which projects are worth building, whether your data can support them, and what they will cost to run.
Starting with sales? See inbound lead use cases by industry.
What we do before anything is built
Engagements range from a short readiness assessment to a pilot. We agree the questions to answer and the deliverables at the start.
Use-case selection
A workshop-led review of where AI could help, then a ranked shortlist scored on value, feasibility, data and risk, with the reasons shown.
Data readiness
Where the data lives, who owns it, how clean it is and whether you may use it for this purpose, with the gaps in priority order.
Build, buy or fine-tune
For each shortlisted case, a comparison of a ready-made product, a custom build, retrieval over your documents and fine-tuning.
Proof of concept
A small build on your real data, judged against a pass mark and a test set agreed in advance.
Cost modelling
Projected running costs at your expected volumes, and how model choice changes them.
Governance and roadmap
A view on data protection and AI regulation for each use case, and a phased plan from pilot to production.
How an AI programme is run
The standard path from ideas to production AI: assess where AI can earn its keep, prove it with a measured pilot, then scale it with the governance to keep it safe.
Assess
Before any buildProve
A measured pilotMisses the target: change the approach and test again.
Scale and govern
OngoingDiscovery
Workshops with the teams doing the work to find where time, errors and backlog really sit.
- In practice
- Process discovery interviews
- You get
- A long list of use cases
Which approach fits your problem?
Five yes-or-no questions. You get the pattern that usually fits, why, and the page that explains how we build it.
Your suggestion
Answer all five questions to see the approach that usually fits first.
How we keep advice honest
A recommendation is only useful if you can check it.
- Reasons written downPersonEvery ranking and recommendation comes with its scoring and assumptions, including ideas we advise against.
- Pass mark agreed firstPersonA proof of concept is judged against criteria set before any code is written.
- Vendor-neutral comparisonTestedModels and products are compared on your tasks and data, not on reputation.
- Plan that anyone can deliverOutcomeThe plan is written so that your team or another supplier could carry it out.
Tools we commonly work with
A recommendation is only as good as its evidence. Naming a tool is not a partnership or an endorsement.
- Model providers
- Claude
- GPT
- Gemini
- Mistral
- Llama
- Azure OpenAI
- AWS Bedrock
- Evaluation
- Braintrust
- LangSmith
- promptfoo
- Ragas
- Data platforms
- Snowflake
- Databricks
- BigQuery
- PostgreSQL
- Governance frameworks
- NIST AI RMF
- ISO/IEC 42001
- EU AI Act risk tiers
- DPIA
- Prototyping
- Python notebooks
- Streamlit
- OpenTelemetry
Common questions about this work
Keep reading related pages
- AI agent developmentAgents that use tools, with people approving risky actions.
- Generative AI developmentDrafting and content features that are checked before they ship.
- Use cases by industryInbound lead handling for SaaS, financial services, logistics and real estate.
- Pilot fit checkSix questions to suggest which agent stages to pilot first.
Tell us what you want built. We will scope it with you.
Start an AI project