Most AI pilots never reach production. Ours are scoped to.
The gap between a working demo and a system your business depends on is data access, integration, security review and someone to run it at 3am. We build across all of those — which is why our AI work ships.
The honest framing
Where AI projects actually stall
Not on the model. On the data that lives in four systems with no clean join key. On the security review nobody scheduled. On the integration into the workflow people already use. On the question of who owns it once the consultants leave.
We assess for those first. Sometimes the finding is that the highest-value work isn't AI at all — it's the data platform underneath. We'll tell you that before you spend a year finding out.
Start here
AI Readiness Assessment
Two weeks. Fixed scope, fixed fee. We interview your teams, inventory your data, and score candidate use cases against value and feasibility.
- Findings document with current-state data and platform inventory
- Prioritised use-case roadmap with cost ranges
- A recommended first build, and why it is first
- Governance framework: acceptable use, data handling, human review
Yours to keep whether or not you continue with us.
Services
Six things we build, and the engineering behind each
AI advisory & readiness
Where AI creates measurable value in your operation, what your data and infrastructure can support today, and a prioritised roadmap with costs and honest sequencing. Includes the governance framework: acceptable use, data handling, human review, model selection.
Agentic automation
Agents that carry out multi-step work inside your systems — triaging requests, reconciling records across platforms, preparing documents, monitoring exceptions and escalating what needs a person. Built with audit trails and hard boundaries, so you can see what the agent did and why.
Knowledge assistants (RAG)
Assistants grounded in your own documents, contracts, manuals, policies and tickets — answering with citations back to the source, and respecting the permissions of the person asking. The hard part is never the model; it is retrieval quality and access control, and that is where the engineering goes.
Document intelligence
Structured data extracted from invoices, contracts, drawings, inspection reports and scanned records, with confidence scoring and a human review queue for anything below threshold. Usually the fastest measurable return in the practice.
Predictive analytics & forecasting
Demand, capacity, asset failure, cost and consumption models built on your operational history — delivered into the dashboards and systems your team already opens, not a portal nobody logs into.
Data platform & governance
The foundation: pipelines, warehouse or lakehouse architecture, data quality, lineage, cataloguing and access control. Unglamorous, and the reason everything above works.
Operations
Running it is the job
An AI system without evaluation is an unmeasured liability. We instrument accuracy and spend from day one.
Pipelines & versioning
Deployment pipelines with version control for models, prompts and retrieval configuration — so a change is reviewable and reversible.
Evaluation harness
Test suites that catch regression before users do, run on every change, with results your team can read.
Cost & drift monitoring
Inference spend tracked and alerted, output quality monitored for drift, and a documented rollback path.
How an engagement runs
From assessment to a system you own
Readiness assessment
Interviews, data inventory, use cases scored on value and feasibility. Findings document, costed roadmap, recommended first build.
2 weeks · fixed feeProof of value
One use case, real data, controlled environment, success criteria agreed before we start. It clears the bar or it doesn't, and we report which.
4–8 weeksProduction
Security review, integration into live systems, monitoring, evaluation harness, documentation and training.
8–16 weeksOperate
Managed support, model and prompt updates, cost optimisation, quarterly performance review — or clean handover with runbooks.
Ongoing
Responsible AI
How we handle your data
Every line below is a commitment we will put in a contract, not a values statement.
- Your data trains nothing. Enterprise model endpoints with zero-retention terms — and we will show you the contract language.
- Permissions are inherited, not reinvented. An assistant shows a user only what that user could already access.
- Consequential decisions keep a human. Review steps are designed into anything affecting people, money or safety.
- Everything is traceable. Inputs, outputs and sources logged, so an answer can always be audited back to where it came from.
- Deployment location is your call. Cloud, your own tenancy, or fully self-hosted where regulation or policy requires it.
Questions we get asked
Straight answers
How long before we see something real?
Two weeks to an assessment with a costed roadmap. Six to twelve weeks to a working system in production for a well-scoped first use case.
Our data is a mess. Are we too early?
Almost every organisation is, and it is rarely disqualifying. The assessment tells you which use cases your current data actually supports — there is usually at least one, and the roadmap sequences the rest behind the platform work that unblocks them.
Can this run without sending data to a third party?
Yes. Open-weight models in your own tenancy or on your own hardware, at some cost in capability. We will be straight with you about the trade-off rather than pretending there isn't one.
What does it cost to run?
We model inference cost during the proof of value and instrument it in production with alerting on spend. No surprises at the end of the month.
What if AI isn't the right answer?
Then the assessment says so, and recommends what is. We would rather lose the AI project and keep the client.
Who owns what you build?
You do — code, prompts, configuration, pipelines and documentation. Handover with runbooks and training is a deliverable in every engagement, not an upsell.
Next step
Two weeks to know where you stand.
Fixed scope, fixed fee, findings you keep regardless of what you do next.