the build. we built a claude-based follow-up assistant for a b2b sales team. every morning it surfaces the deals that have gone quiet and drafts the next touch in the rep's own voice. reps review and send. in one quarter it reactivated 22 stalled opportunities and recovered about $310k in dormant pipeline.
the problem was not lead volume. it was memory. deals stalled after a good second call, the rep moved on to a hotter thread, and three weeks later nobody remembered to circle back. the pipeline looked healthy in the crm. it was quietly dying inside it.
the number the vp of sales cared about: ~$310k in pipeline that was already sitting in the crm, marked open, and going nowhere. no new leads bought. 22 of those deals came back to a live conversation because a rep touched them at the right moment with the right message.
#why do so many deals die in follow-up?
how does ai help with sales follow-up on stalled deals?
ai reads your crm daily, flags deals that have gone quiet past their normal cadence, and drafts the next message in the rep's voice using the real thread history. the rep reviews and sends. it recovers pipeline you already have instead of chasing new leads, and volume stays in human range so deliverability holds.
a rep with 40 open deals cannot hold 40 follow-up clocks in their head. so the loudest deal wins the day and the quiet ones drift. the drift is invisible until quarter-end, when a forecasted deal turns out to have been silent for six weeks. this is the exact pattern our ai for sales work is built around.
#what did we actually build?
three parts, none of them exotic. a step that scores which open deals have gone quiet past their normal cadence. a drafting step that writes the next touch using the real thread history. and a daily list the rep opens instead of combing the crm by hand.
- surface: reads open opportunities in the crm each morning and flags the ones past their expected next-touch date, weighted by deal size and last real reply.
- draft: writes one short follow-up per flagged deal in the rep's voice, citing the last thing that actually happened on the thread. no generic 'just checking in'.
- review: drops the draft into a 'who to nudge today' list. nothing sends until the rep reads it and hits send.
the draft step is the part that changed reply quality. a follow-up that says "you mentioned the renewal lands in q3, wanted to line the pilot up before then" gets answered. "just checking in" gets ignored. Claude reads the thread and pulls the specific hook, so the rep is editing a real message instead of writing one from a blank box.
we are
a daily list that hands each rep the quiet deals worth a nudge and a drafted first version to edit, with a human reading and sending every message.
we aren't
an auto-sequencer that fires templated 'just following up' emails at every open deal on a timer and trains your prospects to filter you out.
volume never left human range. reps send the same number of follow-ups a person could reasonably send, from their own warmed mailboxes. the assistant changed which deals got attention and what each message said, not how many went out. spam complaints stayed flat and deliverability held. that human-in-the-loop line is the same discipline behind everything we ship on the /platforms side.
#how long until it paid for itself?
the first two weeks were noisy. the surfacing step flagged deals that were quiet for good reasons, reps got a list they did not trust, and a few stopped opening it. we spent those weeks tuning the cadence logic and teaching it to ignore deals a rep had deliberately parked.
the recovery showed up once the list got trustworthy. by week three the daily list was short and mostly right, reps opened it out of habit, and the reactivated deals started landing. the ~$310k was the quarter's total, not week one. this is the pattern across our build-in-public case studies: the model is the easy part, the boring loop around it is the win.
i stopped losing deals to my own inbox. the list tells me who went quiet and hands me a message i can actually send.
we are a young studio and we build in public because the receipts are the pitch. if you are a vp of sales watching forecasted deals go silent in the crm, the pipeline to recover is already yours. tell us what you're building and we will scope the follow-up loop against your actual deal data.