Neon One, 2022–2023
Grassroots peer-to-peer fundraising platforms under-serve the top of the donor pyramid. Matching gifts and endurance pledges (a donor pledging against someone's fundraising progress) were a clear competitive gap with large revenue potential, because those dollars were moving offline where the platform earned nothing. Engineering initially estimated ~3 months to build these features, but the deeper the team dug, the more of the payments system turned out to need refactoring or rebuilding that nobody had expected. Three months slowly became twelve.
Monthly go/no-go, taken seriously every time. Every month I met with leadership to review progress and genuinely reconsider cancelling. Every month we deliberately re-affirmed it as top priority based on the evidence at hand. Continuing was not a default we drifted into, but a choice we kept re-earning.
Contain the bet. I isolated one to two developers on the project so the rest of the team kept supporting the platform normally. The overrun never stalled everything else, making the monthly "continue" call sustainable.
Protect the estimators. I don't estimate; I ask the engineers to. When they're wrong, my job is to reassess and shield them from leadership's frustration. At this level of complexity timelines are almost impossible to get accurate. If we ding engineers for an honest estimate, they'll stop giving honest estimates.
Customers deployed matching and endurance pledging, and the platform started capturing donations that it had previously bypassed entirely. The timeline overrun also produced a durable operating principle: I explicitly limit estimates, rarely going beyond ~50 hours for sponsored work, rarely beyond 2–3 months for feature work. The larger the project is, the more it's exposed to unknowns. For anything bigger, I'd rather commit to "this is a multi-month project; let's reassess regularly" than defend a single arbitrary number. This allows us to structure for the uncertainty rather than betting against it.
Running a long, uncertain project without sunk-cost marching or rage-quitting: contain the blast radius, re-earn the go decision on evidence, and keep the engineers honest by keeping them safe.