Services
Four things, done properly.
Two of these we have shipped in our own products and can point at. Two we offer on method rather than a public case study — and we say which is which, below, rather than blurring the line.
AI products & agents
Agents that do real work, with a human holding the gate.
Most AI demos collapse the moment they touch production data, real money or a real user. We build the version that survives: scoped tools, grounded retrieval, evaluation you can run in CI, and an approval boundary wherever the model would otherwise act unsupervised.
What you get
- Agent architecture and tool design
- Retrieval grounded in your own corpus
- Approval gates and reversible actions
- Evaluation harness and regression suites
- Token and latency cost controls
- Monitoring, tracing and failure reporting
Shipped in our own products
Automation & conversion
Find the leak, fix it, then prove the fix earned its keep.
Automation is easy to sell and hard to justify. We instrument the process first so there is a baseline, automate the part with the clearest payback, and measure the result against a holdout rather than against optimism.
What you get
- Behavioural instrumentation and funnel analysis
- Workflow automation across existing tooling
- Experiment design and holdout measurement
- Automated deployment of changes
- Attribution reporting leadership will believe
Shipped in our own products
Small language models
Smaller models, lower cost, data that never leaves your estate.
Frontier models are the right default until cost, latency or data residency says otherwise. When they do, a smaller fine-tuned or distilled model often matches task performance at a fraction of the spend — and can run inside your own boundary. The work is proving that on your data before committing.
What you get
- Task benchmarking against a frontier baseline
- Model selection, fine-tuning and distillation
- Quantisation and on-device or in-VPC inference
- Evaluation harness with accepted quality floors
- Cost and latency budgets, measured before and after
Offered on method — no public case study yet
- 01
Start from your task, not from a model. We build the eval set first, so "good enough" is a number agreed up front rather than a judgement call later.
- 02
Benchmark a frontier baseline, then compete against it with smaller candidates on your actual data and your actual quality floor.
- 03
Ship only if the smaller model clears the floor and the economics are real. If it does not, the honest recommendation is to stay where you are — and you keep the eval harness either way.
ERP & systems integration
Make the systems you already paid for talk to each other.
Most operational pain is not a missing system. It is five systems holding different versions of the truth and a person in the middle retyping between them. We map the flow, fix the data model, and automate the handoffs — starting with the one that is costing the most.
What you get
- Process and data-flow mapping across systems
- Integration architecture and sync pipelines
- Data modelling, cleanup and migration
- Automation of manual handoffs
- Reporting layer over unified data
- Phased rollout with fallback at each step
Offered on method — no public case study yet
- 01
Map before you build. The first deliverable is an honest diagram of how work actually moves today, including the spreadsheets nobody mentions in the kickoff.
- 02
Fix the data model before the integrations. Syncing two systems that disagree on what a customer is just spreads the disagreement faster.
- 03
Roll out in phases with a working fallback at every step. Operations cannot stop while the plumbing changes underneath it.
Not sure which of these you need?
That is a normal place to start. Describe the problem and we will tell you which one it is — or that it is none of them.