Six stages from idea to production
We developed this framework after running dozens of AI projects and noticing where things tend to go wrong: vague requirements, poor data quality, and models that work in a notebook but never reach production. Each stage has a defined deliverable and a sign-off gate before the next one starts.
Free consultation call
A 30-minute video call where you describe the business problem. We ask about your current data sources, the volume of records, the systems you use, and the outcome you want. There is no charge, and there is no obligation. By the end of the call we can usually tell you whether AI is the right tool or whether a simpler approach would serve you better.
Discovery sprint (two weeks)
We access a sample of your data under a non-disclosure agreement and run an initial feasibility analysis. This includes checking data quality, identifying missing fields, and testing whether the signal-to-noise ratio is strong enough for a useful model. At the end of the sprint you receive a written report with one of three verdicts: go ahead, go ahead with data remediation first, or do not proceed. If the verdict is "do not proceed", we do not charge for the sprint.
Proposal and fixed-price quote
Based on the discovery findings, we write a proposal that specifies the model type, the accuracy target, the integration method, the timeline, and the cost. Timelines typically range from four to twelve weeks depending on complexity. The quote is fixed: if we underestimate the effort, that is our problem, not yours.
Build and iterate
Our engineers build the model, train it on your data, and test it against held-out validation sets. We share progress every week through a short written update and a fortnightly demo call. If the model does not meet the agreed accuracy target on the validation set, we keep iterating at no extra cost until it does or until we mutually agree to stop.
During this stage we also build the integration layer: the API endpoint, the data pipeline, or the plugin for your existing software. We test the full chain end-to-end before moving to deployment.
Deployment and handover
We deploy the model to your infrastructure or to a managed cloud environment, depending on your preference. Deployment includes documentation: a technical reference for your IT team and a plain-language guide for end users. We also run a 90-minute training session for the people who will use the system day to day.
Go-live happens on a date you choose. We stay on call for the first 48 hours to handle any issues immediately.
Monitoring and maintenance
Every model drifts over time as the real world changes. Our standard monitoring package runs for six months after deployment. We check prediction accuracy weekly, compare it to the agreed threshold, and retrain the model when performance drops. You receive a monthly one-page report showing accuracy, usage volume, and any anomalies we spotted.
After six months you can renew the monitoring contract, bring maintenance in-house using the documentation we provided, or switch to ad-hoc support at an hourly rate.
What you get at each stage
At the end of the discovery sprint you receive a PDF report covering data quality scores, feature importance analysis, and our recommendation. Clients tell us this report alone has helped them clean up databases they had been ignoring for years.
The proposal document includes a Gantt-style timeline, a risk register with mitigations, and a breakdown of costs by stage. No line item says "miscellaneous" or "contingency". If we think something might go wrong, we name it and explain how we plan to handle it.
After deployment, the handover pack contains the model card (inputs, outputs, known limitations), the API specification, the retraining script, and the user guide. Everything is version-controlled in a Git repository that we transfer to your team.
Questions we hear often
Most projects fall between £8,000 and £45,000. A straightforward predictive model on clean, structured data sits at the lower end. A multi-stage automation system with custom integrations sits at the upper end. The discovery sprint gives us enough information to quote accurately.
Not necessarily. We can work with anonymised or pseudonymised data in many cases. Where personal data is required, we sign a data-processing agreement and follow GDPR requirements throughout the project. We can also train models on-premises if your data cannot leave your network.
We agree on a measurable accuracy target before the build phase starts. If the model does not reach that target on the validation set, we keep working at no additional cost. If after a reasonable number of iterations we conclude the target is not achievable with the available data, we explain why, deliver what we have, and refund the build-phase fee.
Yes. We have integrated models with Salesforce, SAP, Microsoft Dynamics, Xero, and several bespoke ERP systems. If your software has an API or supports webhooks, we can almost certainly connect to it. If it does not, we discuss workarounds during the discovery sprint.
From the first call to a live model, most projects take between six and sixteen weeks. The discovery sprint is two weeks, the proposal takes a few days, and the build phase is typically four to twelve weeks depending on scope. Deployment itself usually takes one to three days.
Ready to start?
Book a free 30-minute consultation. We will tell you honestly whether AI is the right fit for your problem.
Book a call