I didn't set out to start a company. The product eventually required one.
tactica. began as a football model on my laptop. Four months later, the practical realities of building it meant the product needed a company around it.
tactica. Journal
Football intelligence,
built in public.
Building, auditing and learning from tactica. — a football intelligence platform built around evidence, transparency and accountability.
Why we're creating the Founding 100.
Why we're creating the tactica. Founding 100 Kickstarter, what early supporters receive and what we're still building.
Read the story →From the Journal · 9 stories
A beta isn't supposed to prove you were right.
tactica.'s Closed Beta wasn't built to prove we'd built the right product. It was built to expose where we hadn't and those lessons are now shaping tactica. 2.0.
At some point, you have to start treating it like a company.
tactica. started as a project on my laptop. Now I'm thinking seriously about investment, ownership, distribution and what it would mean to build it as a company.
Building a model that knows when not to trust its data
tactica. found that having football data wasn't enough. The model also needed to understand whether evidence was valid, complete, timely and when not to trust it.
What should a football model understand before it makes a prediction?
We're rethinking what should happen before tactica. makes a prediction from historical team evidence and expected lineups to matchup intelligence, uncertainty and the decision to make no recommendation at all.
An 80% probability didn't mean we were more confident than 64%.
tactica. discovered that an 80% protected-market probability wasn't necessarily stronger football conviction than a 64% outright probability. Our first attempt to fix it didn't survive independent review.
We fixed what our model was saying. We didn't change what it predicted.
tactica. found 415 cases where an explanation attributed part of a prediction to something that hadn't affected it. The evidence told us to fix the explanation—not change the model.
Our model was profitable. We audited it anyway.
tactica. returned +4.96% across its first meaningful observational Performance sample. Instead of treating that as proof the system was healthy, we audited the intelligence underneath it.