How we build AI: our development process
Our AI development process is built to get one use case into daily use quickly, then grow from there. Every stage has a clear output, a decision point for you and no surprise costs.
Start with an assessment- 350+clients helped with AI, software and automation
- Since 2018building AI for businesses
- 30 daysto a first working AI release
- 100%refund promise on the first release
Our AI development process, stage by stage.
This AI development lifecycle is the same whether we are building a generative AI assistant, an AI agent, a machine learning model or an AI app. What changes is the detail inside each stage.
- Stage 1
Assess the use case
We review the task, the people who do it today, the systems and data involved, and what success would look like. You learn whether AI is the right answer before spending on a build.
- Stage 2
Scope and price
You receive a written scope: deliverables, success measures, review steps, a fixed price for the first release and an estimate of monthly running costs.
- Stage 3
Design data, model and integrations
We plan data access and preparation, choose the model on accuracy, speed and cost, and design how the AI connects to your systems and where people review its work.
- Stage 4
Build and evaluate
We build the solution and test it against a set of real examples agreed with your team, measuring accuracy, edge cases and cost per task.
- Stage 5
Integrate and launch
We connect your systems, launch with one team, train users and hand over code and documentation.
- Stage 6
Monitor and expand
We track accuracy, usage and cost, improve prompts and models, and scope the next use case once the first one has proved itself.
What makes this AI implementation roadmap different.
Many AI implementation processes spend months on discovery and slides. Ours is designed around a working release your team can use, as early as possible.
The same roadmap scales from a single workflow to a company-wide program, one step at a time.
One use case first
We avoid big-bang programs. One focused use case proves value and builds the foundations for the next.
Real data from the start
AI is tested on your actual documents and records, not a polished demo set, so results hold up after launch.
Decision points for you
Each stage ends with a clear output and a go or no-go decision, so you are never locked into the next phase.
Running cost designed in
Model choice and architecture are made with your monthly AI bill in mind, not added as an afterthought.
The AI lifecycle doesn’t stop at launch.
AI systems need care after go-live: data changes, models improve and new use cases appear. This loop keeps your solution accurate and affordable.
- ↓01 / MONITOR
Track quality and cost
Accuracy, usage, errors and cost per task are tracked from day one.
- ✧02 / IMPROVE
Tune prompts and models
Prompts, retrieval and models are updated and re-tested on your evaluation set.
- ✓03 / REVIEW
Check results with you
We review results against the success measures agreed in the scope.
- ↗04 / EXPAND
Scope the next use case
The next workflow reuses the same connections, so it usually goes faster.
Results this process has delivered.
Each of these projects followed the same stages, starting with one clear use case.
Challenge: Engineering proposals took a full day of manual work.
Engineering proposals, without hours of manual work.
Proposal work reduced from 8 hours to under 30 minutes.
Challenge: Each new ERP record in Fishbowl took around 30 minutes to set up.
ERP updates that take seconds.
Fishbowl setup reduced from 30 minutes to 3 seconds.
Challenge: Answers were buried across millions of content items.
Answers from a large knowledge base.
An AI knowledge agent built to find information across millions of content items.
Challenge: A growing client base needed a platform built to scale.
A SaaS platform built to grow.
Platform development and AI features that support a growing client base.
Challenge: The platform was unstable after a migration.
A stable platform after migration.
Migration and engineering work that restored platform stability.
Challenge: AI responses needed higher quality and clear guardrails.
Better AI responses, with clear guardrails.
An emotion-aware AI governance layer built around response quality.
Results are specific to each project. Your scope and success measures will be agreed for your project.
What we’ll need from your team.
A fast first release depends on a few things from your side. We agree these during the assessment.
A decision maker
One owner who can make calls on scope, priorities and launch.
System access
Access to the systems involved, or a test environment, set up early.
Sample data
Real examples of the documents, records or requests the AI will handle.
Reviewers
A few people who do the work today to review AI output during testing.
Your first AI release.
Live in 30 days.
We define the use case, data controls and expected running costs before we write any code.
- 1Days 1 to 3
Define the use case
Map the task, confirm access and agree on the scope, success measures, data controls and expected running costs.
- 2Days 4 to 10
Build the core solution
Select the right model and architecture, and build the AI solution around your inputs, rules and required output.
- 3Days 11 to 22
Connect and test
Connect the systems and test the workflow, permissions and expected AI usage with your team.
- 4Days 23 to 30
Launch it
Put the AI solution to work, confirm the running setup and hand over the code and documentation to your team.
100%REFUND
Working AI in 30 days. Or your implementation fee back.
We agree on one use case, a clear scope and what “working” means before we start.
The 30-day delivery window begins once the agreed access, data and other requirements are ready. Refund terms and acceptance criteria are defined in your signed agreement.
Is this a good fit?
Strong fit
- One clear use case or business process
- A problem that costs time, money, capacity, or customer satisfaction
- A measurable outcome
- A project owner who can make decisions
- Access to the employees who understand the process
- Access to the required data and tools
- A real need to launch soon
Not a fit
- A company-wide transformation in 30 days
- No clear process or business problem
- No measurable outcome
- No access to the required data, tools, or users
- An open-ended research project
- A generic chatbot with no clear business purpose
- A project that depends on unavailable third-party access
Start your AI assessment.
Tell us what you want AI to do for your business. We’ll review the use case, systems, data requirements and whether a first release fits a 30-day build.
- 1. We review the use case and the systems involved.
- 2. We confirm whether it fits a 30-day build.
- 3. You get a clear scope and price before deciding.
Straight answers on how we work.
What are the stages of AI development?
Our AI development process has six stages: assess the use case, scope and price, design data, model and integrations, build and evaluate, integrate and launch, and monitor and expand.
How long does the AI development process take?
A first working release for one focused use case is planned for 30 days once access, data and requirements are ready. Later use cases follow in stages.
What is an AI implementation roadmap?
A plan for which AI use cases to build, in what order, with what data and systems. Ours starts with the use case that pays back fastest and adds the next once the first has proved itself.
How do you test AI before launch?
We agree a set of real examples with your team and measure accuracy, edge cases and cost per task against it before launch, then keep using it after launch.
What happens after launch?
We monitor quality, usage and cost, update prompts and models, and scope the next use case when you’re ready.
Can we pause between stages?
Yes. Each stage ends with a clear output and a decision point, so you choose whether and when to continue.