180Connect uses third-party AI today, because it is far cheaper while the branch is small. In time the branch could train its own models on its own outreach — but only once it has enough real results to learn from. This page shows how close each model is, roughly how long is left at the current pace, and the steps from here to a model in use.
Each model has two milestones: the first is enough to try, the full is enough to train properly. Careful estimates, not guarantees.
Learns from clients who answered, yes or no — most emails get none, so answers are the scarce part.
0 of 500 emails that got an answer
Every email a CAM checked and sent teaches a model how the branch writes.
26 of 1,000 emails sent
Wins are the rarest result, so this fills slowest and comes last.
5 of 200 clients won
The steps the branch would take, at a high level. Stages 3 onward need a developer and a budget.
CAMs send outreach and record what happened: replies, no answers, and clients won.
Make sure there are enough answers, and that they don't all come from one CAM, one sector or one busy month.
Train on older months and check whether the model would have called the newer months better than today's AI.
Cost out training and running the model. Tuning an existing open model costs hundreds of pounds, not thousands.
How recorded client responses and results break down in our database.
No client outcomes recorded yet. As Client Acquisition Managers (CAMs) record outreach replies and conversions with clients, the breakdown appears here.
The new model makes suggestions next to the current AI, and CAMs compare the two in their normal work.
The new model takes over; the previous one stays ready to switch back to if results drop.