Enterprise software sales, twenty years of it. Complex deals: long cycles, several parties with different incentives, and a buying committee where the person who wants it is almost never the person who signs. Most of that time has gone on building selling motions rather than working existing ones.
Which means sitting between three things that rarely agree. A CRM process that costs sellers time and hands little back. The reality of a deal, which is messier and more changeable than any field set can hold. And the reporting a C-suite legitimately needs. That gets framed as a tug of war and it is not one: sellers and executives want the same three things, which are control of the deal, deals that close, and a forecast that turns out to be right. The argument has only ever been about the instrument.
Tawkshop is my attempt at a better one.
CRM was designed for internal communication and tracking. It describes what your company does with a deal rather than what is true about the customer, and thirty years on we still mistake one for the other. A model can now read the two hundred emails, so the compression has stopped being the valuable part.
Reading was the easy half. Deciding what is true is not. When two stakeholders contradict each other, a model has no principled way to tell whether one is wrong, whether one outranks the other on that question, or whether both are true about different parts of the business. So Tawkshop runs a diagnosis rather than asking a model to adjudicate: it measures what is genuinely understood against what is merely asserted, holds conflicts open with both sources until a person closes them, and states its gaps instead of filling them. Fields still exist, because reporting and comparison need them, but they are derived from evidenced context rather than being the record.
Python and FastAPI, React, Postgres on Supabase. Twenty-plus releases, a written constitution, and no model writing to canonical state. The code is private, the thinking is not.
pytest-dbguard, a small thing I extracted after finding my test suite had been writing to production for four months.
British, living in northern Spain. Software development is the field I have spent a career next to without ever getting into: selling to developers, working alongside them, translating between product and customer. What I could never do was build the thing myself.
That has changed, and the governance in Tawkshop is a direct consequence. I needed a way to keep a system honest that did not depend on my reading every line.
If correctness matters more than throughput where you are, or you want to argue with any of the above, I would like to hear about it.