Two weeks ago we opened the beta to a handful of hand-picked operators, single digits on purpose. I promised field notes over launch theater. Here is the first honest installment: what surprised us, what broke, and why a small first group keeps proving itself the right call.
What the first operators did first (not what we predicted)
We assumed everyone would rush to load lead lists. Volume, after all, is the promise.
Nobody did. Every single member started the same way: they poked the AI. They called their own new phone number to hear their assistant answer with their company name. One member called his own line at a quarter past midnight, pretending to be a seller, trying to stump the assistant. It answered as his company, walked "the seller" through the full qualification, condition, timeline, price, motivation, and booked the callback he asked for. He could not shake it.
Then they read their own transcripts. Then they tested the phone lines again. Only then did anyone start thinking about data.
The lesson landed hard: trust in the assistant comes before volume through the assistant. An operator hands their seller conversations to software only after they have personally tried to break it. So onboarding now leans into that moment instead of hurrying past it: call your assistant, interrogate it, try to trip it. It holds up, and something changes in how people use the platform afterward.
What broke, and how fast it got fixed
A beta with single-digit members exists to catch exactly these. All were found by real usage, root-caused the same day, and fixed within about a day.
The scheduler heard "Friday" and booked "today." That midnight stress test asked for a Friday callback. The date resolver handled "today" and "tomorrow" but quietly fell back to the current date for weekday names. Found in the morning logs, fixed before lunch. Every callback since resolves correctly.
The classifier misread an honest seller. A homeowner answered a follow-up with the facts of their situation: no equity, needs lots of repairs. The reply classifier heard the word "no" and filed it as a brush-off. A human negotiator knows disclosure is engagement, someone describing their problem to you is talking to you, not waving you off. We wrote that rule down, taught it to the system, and the corrected follow-up went out the same day.
An onboarding record never marked itself finished. A member did everything right, and one internal status still read "draft" forever, which made one dashboard lookup come up empty. Nothing user-visible broke, but the record lied, and records that lie eventually cost you. Fixed, backfilled, and the flow now finishes what it starts.
Three fixes the next fifty operators will never see. That sentence is the entire business case for starting small.
What we shipped around the cohort
None of it glamorous, all of it the difference between a demo and a platform: carrier compliance registration in motion for every member (the unskippable, week-scale bureaucracy that legitimate texting requires), usage caps wired to every spend path so nobody can accidentally burn a budget, and daily automated health checks on every pipeline lane, because silent failure is the failure mode that kills operators.
The honest scoreboard
It is small and it is honest: the first operators are inside, the system is catching their edge cases, and the current push is getting every member from touring the system to working their own leads through it. No results theater yet, no cherry-picked screenshots. Results with numbers attached, when the numbers are real. That is the deal this newsletter and blog made with you, and two weeks in, it holds.
We build for part-time real estate investors doing this around a job, without a team. The software runs the persistence. The operator keeps the judgment.
Want to see what your own list looks like through that lens? The free Pipeline Grader ranks the leads you already own and puts a dollar figure on the pile. We never store your list.
People get tired. Computers don't.