Samlytics
A revenue assistant that can explain its own answers
A revenue tool I designed and shipped end-to-end (independent 0→1 build) — an assistant you can ask questions, grounded in your own calls and CRM, plus deal scores that come with the reasons behind them, not just a number. Multi-tenant, BigQuery-first, used by 200+ sales users.
Problem
Revenue teams sit on a pile of signal — calls, emails, CRM history, product usage — and still mostly run on gut. The data's there but hard to actually ask questions of, and the few models that score deals are black boxes nobody trusts enough to overrule a rep.
Insight
A revenue assistant only earns its place if it can answer 'why' — why this deal looks weak, why this account looks shaky — in a way a person can push back on.
Being able to explain itself isn't a nice-to-have here; it's the whole reason anyone keeps using it. A score you can't question just gets overridden and forgotten.
What I built
Three pieces working together: an assistant that answers questions using the team's own calls and CRM data (so it isn't making things up); a layer that structures what's actually said across deals; and a scoring model that, for each deal, shows which factors drove the score.
You can ask 'why is this account at risk?' and get an answer that points at the specific calls, the stalled steps, and the factors behind it.
How it fits together
A retrieval layer over the team's calls and CRM → an assistant that answers questions grounded in that → a pipeline that turns conversations into structured signals → a scoring model that shows its drivers → one place to see it all.
The thread running through all of it: every answer and every score should trace back to something the user can go look at.
Calls I made
Pick the model you can explain over the one that's a hair more accurate but opaque. A score you can argue with gets used; a slightly better one you can't doesn't.
Keep the assistant strictly grounded in the team's own data. A confident, made-up claim about a real account is a trust-ending mistake.
Put the 'what' and the 'why' in the same place, so people aren't jumping between tools to understand a single deal.
What happened
It's in active use by 200+ sales users. Conversations about deals got more concrete — instead of competing anecdotes, people could pull up the actual signals — and the explainable scores earned enough trust to actually shape what reps prioritized.
Shipping it solo, end-to-end, also forced every product call to be one I had to live with — no handing off the hard parts.
What I learned / would do differently
Explainability is what moves a model from ignored to relied-on. It's not a compliance checkbox; it's the adoption strategy.
Most of the assistant's quality came down to curating the data it drew from, long before any model tuning.
Revenue teams don't want a smarter oracle. They want a faster, more honest argument with the data in front of them.