How our Artificial Intelligence projects work
From the first phone call to a running system, here is every step you can expect. No black boxes, no surprises.
The six stages of an engagement
Most projects follow this arc. Timelines vary with scope, but the structure stays the same because it keeps both sides aligned.
Discovery call
We start with a 45-minute video call. You describe the problem; we ask questions about your data, your team, and what success looks like. By the end we both know whether there is a fit. If there is not, we say so and, where we can, recommend someone better suited. This call is free and comes with no obligation.
Data and workflow audit
If we agree to move forward, we spend two weeks inside your operation. We review your databases, spreadsheets, CRM exports, and manual processes. We interview the people who actually do the work, because the gap between how a task is documented and how it is performed is where most AI projects fail. At the end you receive a written audit report with a scored opportunity matrix and a recommended project plan.
The audit costs £1,800 as a standalone deliverable. If you proceed to a build, we credit that amount against the project fee.
Proposal and scope agreement
Based on the audit, we draft a statement of work covering deliverables, timelines, data-handling obligations, acceptance criteria, and cost. We walk through it together, negotiate where needed, and sign. Nothing starts until both parties are comfortable with the scope.
Build and iterate
This is where the engineering happens. We work in two-week sprints. At the end of each sprint we show you a working prototype, collect feedback, and adjust. You are not waiting months for a reveal; you see progress fortnightly.
Typical build phases last between six and fourteen weeks, depending on whether we are configuring existing tools, fine-tuning a pre-trained model, or building something from scratch. We handle data cleaning, feature engineering, model training, and integration with your systems.
Testing, training, and handover
Before go-live we run the system against a held-out test dataset and measure accuracy, latency, and edge-case behaviour against the acceptance criteria from the statement of work. If anything falls short, we fix it before deployment.
We then train your team. Sessions are hands-on, not lecture-style. Each participant works through real scenarios on their own machine. We leave behind a written playbook and a short video walkthrough for future hires.
Go-live and ongoing support
We deploy into your production environment and monitor closely for the first 30 days. If you opt for a retainer, we continue monitoring model performance, retrain when accuracy drifts, and hold quarterly reviews. If you prefer to manage it in-house, we make sure your team has everything they need before we step back.
What makes this process different
Most AI consultancies start with a technology pitch. We start with your problem. If a simple rule-based script solves it, we will tell you that and save you the cost of a machine-learning project.
We also insist on fortnightly demos during the build. Some firms disappear for three months and come back with something that does not fit. We would rather catch misalignment in week two than week twelve.
The audit-credit model means you never pay twice for discovery. And because every statement of work includes explicit acceptance criteria, there is no ambiguity about what "done" means.
Questions we hear often
How long does a typical project take?
From discovery call to go-live, most projects run eight to eighteen weeks. A straightforward integration of an off-the-shelf tool can be faster. A custom model trained on a large, messy dataset takes longer. The audit report gives you a realistic timeline before you commit.
Do we need a data team in-house?
Not necessarily. We can manage the technical side entirely. That said, having at least one person internally who understands the system makes long-term maintenance smoother. Our training sessions are designed to create that capability.
What if the audit shows AI is not the right solution?
Then we tell you. We have talked three companies out of AI projects in the past year because simpler automation would have done the job at a fraction of the cost. You still get the audit report, which is useful regardless.
Who owns the models and code?
You do. Every statement of work includes an IP assignment clause. Once the project is paid in full, all custom code, trained model weights, and documentation belong to your company.
Can we start small?
Absolutely. Many clients begin with a single-use-case pilot, see the results, and expand from there. The audit is a good low-risk entry point because it gives you a clear picture before any significant spend.