Testimonials
Clients in their own words. Names withheld under confidentiality, roles and sectors are real.
They told us our data could not support the model we asked for. It was the most useful invoice we have ever paid.
Ten more, unedited
Every number they gave us had an interval around it. Our previous vendor gave us four decimal places and no test set.
Our planner beat their first three models. They said so in the report rather than quietly changing the metric.
They recommended switching off a model we were paying them to maintain. I did not know that was allowed.
The model card told us where the system would fail before it failed there. That is the whole point and nobody else does it.
We asked them to fine tune a language model. They talked us out of it and built something cheaper that we can audit.
They found the target variable leaking into a feature in a system that had been live for two years.
They evaluated a vendor's model for us and told us the claim did not hold. That conversation saved a seven figure decision.
The annotation protocol they wrote was more valuable than the model. We use it for everything now.
Two years old as a firm and more rigorous than the research group we worked with before.
They refused to build something we asked for, and explained why in a way I could take to my board.
We will put you on a call with any of them
Including the client whose model we recommended switching off, and the one we told not to proceed. Those two conversations will tell you more than this page does.
Request a reference callSend us your data problem. We will tell you whether it is one.
A working data scientist reads every enquiry. You will get an honest read on whether your data can support the model you have in mind, before anyone quotes you a number.