We Cut Costs 80%. It Still Wasn't Enough.
A regulated startup, an offshore team in crisis, a six-month rewrite, and the accreditation clock that beat us anyway.
My first day at TURION, an online-degree startup trying to become an accredited university, I didn't get an onboarding doc. I got a queue of people with complaints about the product team and no clear answer to "what are we actually building toward." The company was operating in one of the more regulated corners of software — accredited higher education — with an offshore development team that was new to building anything this complex, spread across time zones, and reporting to an engineering manager who was, understandably, at the end of his rope.
I spent the first stretch just building a product function that didn't exist yet: a real roadmap, weekly status calls with the rest of the company, mockups people could actually react to, and eventually a weekly development demo directly with the CEO. Visibility fixes a lot of things by itself — most of the "product direction" complaints were really "nobody shows us what's happening" complaints.
It wasn't enough to fix the underlying delivery problem. One evening it came to a head: the engineering manager lost it at the offshore team over another missed target, and in the fallout, we had people resign. I ended up directly leading both the dev team and the design team just to keep things moving while we figured out what came next.
What came next was the harder call. The offshore team kept missing the new targets even after I stepped in. At some point "coach the existing team back to health" stopped being the responsible plan, and I brought in a new development team from my own network to do something I wouldn't recommend lightly: a complete rewrite of the platform, on a six-month clock, to hit our launch date.
While that was underway we ran a cost audit and found a second problem hiding next to the first one — we had an in-house machine learning team building custom models we didn't actually need. The new dev team could get equivalent functionality out of open-source models instead. We let the ML team go and re-hosted the application on that leaner stack. Between the new dev team's rebuild and that infrastructure change, we cut combined infrastructure and development spend by roughly 80%.
We hit the launch date. The rewrite worked. The cost structure worked.
We still didn't get accredited in time to close the next funding round, and without that round, there wasn't a path to keep going.
I think about that ending more than I think about the 80% number, honestly. Hitting the deadline you control doesn't guarantee the outcome you need, when the outcome actually depends on a clock you don't control — a regulator's timeline, in this case, not an engineering one. The instinct in product is to treat "we shipped" as the win condition, because it's the part you can actually engineer. Sometimes it's necessary and still not sufficient, and the honest version of the story includes that part too.