In large consulting projects and enterprise AI programs, too many AI and data science projects follow the same unhelpful pattern. A team of data scientists works for months in notebooks, training and tuning models, carefully optimising precision, recall, and accuracy. Eventually, the results are presented: metrics are up, charts point in the right direction, and the room nods in approval. But then… the project finishes.
The model never leaves the data scientist’s laptop or cloud. It lives in a Jupyter notebook or a Python script on a server, accessible only to a small technical group. Or gets wiped at the end of the project. The majority of people who could actually benefit from the model are not technical, so they can’t interact with it directly. Without an accessible, usable way to bring the model into everyday workflows, it simply can’t deliver ongoing value.
To create actual change, you need a good model AND you need adoption. And adoption comes from empowering people to actually use the thing you’ve built. That means designing and delivering an interface that makes the model usable for non-technical users from day one, not as an afterthought. It means taking the time to understand how people work, what slows them down, and how the AI can make their lives easier.
On a recent banking project at Faculty, where we implemented agents for reviewing held transactions, we made this a priority. We didn’t stop at a model with good metrics; we delivered a simple, intuitive interface that any relevant user could access without technical training. From the very early stages of the project, users were working with the tool in their daily workflow.
Before, reviewing a held transaction could take a human analyst several minutes jumping between multiple systems, pulling up transaction histories, and making judgement calls with incomplete information. After, the interface brought all the relevant data and the model’s recommendation into one place, in a format they could act on immediately. Decisions that once took minutes were made in seconds, and with more confidence.
By the time the project formally ended, the solution was already embedded in their routine. It wasn’t a “new system” they had to remember to use: it was simply how the work got done. The tool saved time, reduced frustration, and made their jobs easier.
The lesson is simple: a model in a notebook might impress in a demo, but a model in a well-designed interface changes behaviour. And it’s that behaviour change: widespread, sustained, and measurable that turns an AI project from a one-off experiment into a lasting success.