Machine learning

It depends on the problem, not on a fixed amount. What matters is enough history to cover the patterns the model must learn, including seasons and rare events, and reliable labels for the outcome you want to predict. We check this during the data review and tell you plainly if the data cannot support the target yet, and what to collect so it can.

Yes, and we design for it from the start where decisions affect people, such as credit or claims. We prefer models whose reasoning can be explained, provide reasons for individual results, and document the data, features, validation and limits of each model version. That documentation supports your model-risk and regulatory reviews.

We agree with you which groups and outcomes matter for the decision, then measure error rates and results across them on held-out data. Where differences appear, we investigate the data and features behind them and discuss the trade-offs with your team. Fairness checks are rerun with every retraining, not only at launch.

Monitoring shows it before the business feels it. We track the input data and the model’s results against what it saw in training, and alert when they shift. Your team then sees the evidence and decides whether to retrain, adjust thresholds or fall back to rules. Retraining runs through the same tested pipeline as the original model.

The main drivers are the state and accessibility of your data, whether labels exist or must be created, how many systems the model must connect to, and the level of explanation and documentation your regulators expect. Running costs depend on prediction volume and whether results are needed in real time or in batches. We estimate once the data review is done.

The baseline and first validated model come early, so you can judge the value before the production work. Getting to production depends on data access, integration with the systems that use the result, shadow running beside the current process and any model-risk review. We give you a plan after the data review, with the decision points marked.

Intellectual property in the work we build for you is yours from the first commit — no assignment step at the end. That includes the trained models, feature code, pipelines and documentation produced for you, and your data stays yours. Any XSIS tools that existed before stay ours and are licensed to you. Open-source libraries we use remain under their own licences, which we record for you.