Data Department worked with a technology group over several years to build its own ML (machine learning) and AI (artificial intelligence) capability. We mentored the squads of a new R&D (research and development) centre through the full machine learning lifecycle, ran group-wide ML and AI training, and helped grow the function into a department of more than 30 people that built 20+ models across 7 industries.
The decision
The group had a choice that many companies face. It could keep buying machine learning projects one at a time from outside, or it could build the capability in-house.
It chose to build. That decision brings its own risks. New teams often produce promising notebooks and demos that never reach production. Hiring alone does not solve this: people need to learn how to frame a business problem, get data into shape, deploy a model and keep it running.
Our role was to shorten that learning curve, so the new teams built models that reached production rather than experiments.
What we built
This project was about people and practice, not a single model. The work had three parts.
| Workstream | What we did |
|---|---|
| Squad mentoring | Mentored a new R&D centre's squads through the full ML lifecycle |
| Domain squads | Four squads, each focused on one area: hospitality, energy, tourism and automotive |
| Group-wide training | Ran ML and AI training across the wider group |
| Department growth | Helped grow the function into a 30+ person ML and AI department |
Mentoring across the full lifecycle. The machine learning lifecycle runs from framing the business problem, through data preparation, modelling and evaluation, to deployment and monitoring. We worked alongside the squads at each stage, so they learned by working on real projects in their own domains.
Domain-focused squads. Each of the four squads worked on one area of the group's business: hospitality, energy, tourism and automotive. Domain focus keeps teams close to the decisions their models are meant to support.
Training beyond the core team. Machine learning projects succeed when the business side understands what models can and cannot do. Group-wide training built that shared understanding outside the technical teams.
Results
- 4 squads mentored through the full ML lifecycle: hospitality, energy, tourism, automotive.
- Group-wide ML and AI training across the wider group.
- 30+ person ML and AI department built up.
- 20+ models built across 7 industries.
What made it work
Real decisions as the measure of progress. A team's maturity shows in how many models it builds for real business decisions, not how many prototypes it produces. More than 20 models across 7 industries is the number that matters here. Our spare parts forecasting case shows the same principle: a model designed to fit into the tools people already use.
Learning on real projects. Mentoring squads through real projects in their own domains taught more than classroom training alone. Training then reinforced the practice across the group.
Organising around business domains. Squads aligned to hospitality, energy, tourism and automotive kept the work tied to real business decisions.
Building for independence. The aim of enablement is a team that no longer needs outside help for routine work. That is the same principle behind our AI strategy and fractional Chief AI Officer advisory: senior guidance while your own team grows. If you are deciding whether to build or buy, the free 30-minute AI opportunity review is a good place to start.
Frequently asked questions
What does mentoring an ML team involve?
It means working alongside a team through the full machine learning lifecycle, from framing the business problem to deploying and monitoring models. The team does the work; we guide decisions, review approaches and share practice.
How long does it take to build an in-house ML department?
In this case it was a multi-year effort that grew the function to more than 30 people and 20+ models across 7 industries. The timeline depends on the starting point, hiring and the pace of the first projects.
Should we build an in-house team or work with consultants?
Many companies do both: outside experts deliver early projects and coach the in-house team, which then takes over. The right mix depends on how central machine learning is to your business and how many use cases you have.