Your pipeline is not in the CRM. It is in a messenger, and it is a week behind.
Twenty-four live threads on screen and none of them entered anywhere. The follow-up nobody sent because nobody was watching. The proposal PDF hunted for in Finder again. Below is what that work looks like today, what it looks like once Silow has mapped it, and what it gives back in hours, in money, and in capacity.
How the work looks today. How it looks after.
Seven processes, in the order we would take them. The left-hand column is not a caricature of a sales team — it is what recorded teams were observed doing. Where a row is tagged Measured, the number comes from real recorded work. Where a row is an extrapolation, it says that too.
Up to 66% of screen time in the go-to-market team we recorded was a messenger. Not the CRM, not the dialler, not the sequencer — the messenger. The qualification, the objection, the price conversation, the promise to send something by Friday: all of it happens in threads, and none of it is anywhere a manager can see on a Monday.
The conversation is the source of record, and the systems are downstream of it. What was said, what was promised, what is now owed and by whom — read out of the work itself rather than typed in afterwards by the person least inclined to type.
Twenty-four live threads were on screen. Not one had been entered into the CRM. One person spent 190 minutes over three weeks moving them across by hand — roughly an hour a week, on one ritual. Elsewhere in the same capture, the team was hand-building a browser extension to scrape their own leads into the CRM. Building, by hand, the thing the recording already filled in.
Conversations become rows a human confirms. We are deliberate about what this is, because the honest verdict from our own report is unflattering: the trace fills an activity board, not a deal board. In that recording roughly four rows were genuinely hard — the largest "deal" the naive version found was a travel booking. We would rather tell you that now than after you have bought it.
The follow-up gets sent when someone remembers, three days late, written from memory. The thread that went quiet stays quiet because nothing is watching it. The messages that actually move a deal forward live in one rep’s head and in their sent folder, and leave the company when they do.
The next step is drafted from the language your team already uses at that stage — captured from real threads, not invented by a model. The quiet thread surfaces on its own. A human reviews and sends; nothing auto-sends to a customer.
Before a call, someone spends twenty minutes in six tabs assembling what is already known: the last thread, the last proposal, the site, the notes, the person. It is preparation work, done identically every time, by the person you hired for the conversation rather than the assembly.
The brief is pre-assembled from what already exists across the team’s tools, and the rep starts at the conversation. Observed in the trace across the teams we recorded — the same assembly shape, every time a call is prepared.
One PDF was manually located in Finder and re-attached across two messengers on five separate occasions in three weeks. No shared link for it existed. Proposals start from an empty page rather than from the last one that closed, because nobody can find the last one that closed.
The document has one retrievable place, and the proposal starts pre-filled from the deal that actually converted. The five-times-attached file is not an anecdote — it is a category, and it is the cheapest hour on this page to get back.
Across 105 reconstructed tasks, work was moving a median of 3.4% of the time it existed. The other 96% it sat in a queue. And the queue has a shape: work that needs a second human — a reviewer, an approver, a customer reply — waits a median of 7 days, while work one person can finish alone waits 1.1. Two thirds of all waiting, 67.4%, sits across a handoff.
Every handoff gets a defined shape: what is attached, what "done" means, who is waiting and since when. The queue becomes visible, which is most of the fix. Making the rep faster optimises the 3.4%. Your cycle time lives in the other 96%.
You report adoption from licence seats and app names. Measured that way the recorded team used AI 11.7% of the time. Measured from the actual work: 14.9%. One person was undercounted sevenfold — 0.9% against a real 6.2% — because his assistant runs inside his code editor and the telemetry only ever sees the editor.
The share of real work that actually ran through a model, measured from the work itself. Not seats sold, not logins — the version a CRO can defend when the board asks what the AI budget bought the revenue team.
We have recorded two teams, and neither was a standalone sales floor. One was a 9-person go-to-market team — marketing, sales and operations together — 5 weeks, ~325 hours of work, 748 tasks, ~88,000 scenes. The selling function inside it is where the CRM, messenger, follow-up and proposal findings on this page come from. The other was an 8-person cross-border operations team, 8 days, ~95 hours. Both anonymized. Rows tagged Measured come from that recorded work; rows that are extrapolations are labelled as extrapolations and carry no number we did not see. Every figure was computed before a model was allowed to write a sentence about it, then each sentence was checked back against the figure.
Time, money, and the people you already have.
The rule for this section: not one invented percentage. Every figure below is either measured in a recording or supplied by you — because a saving you cannot defend in a board meeting is not a saving, it is a slide.
We do not quote you an industry percentage, and we will not show you another vendor’s case study. Every candidate above is scored in hours against your recording, then ranked by payback — so the first thing you build is the one that gives back fastest, not the one that demos best. The anchors we can already point at: 190 minutes over three weeks retyping threads into a CRM, by one person, on one ritual. A file re-found and re-attached five times. A median seven-day wait every time work crosses a desk.
The first is hours × your loaded cost — your rate, not a benchmark we made up. The second is the one nobody counts: the AI you already pay for and nobody opens. In the operations team we recorded, the company had built its own internal assistant and recommended it to the team. Recorded usage across eight people: seven at zero percent, one at one. That licence line is being paid every month right now.
No new system to adopt, no data to move, no process change while we look — Silow runs on the tools the team already uses, including the messenger the deals actually live in. The capacity comes back inside the same team, selling the same way with the assembly work removed. What you then do with that capacity is a decision for the people running the business, taken with better information than they have ever had.
A deal is moving 3.4% of the time it exists. Making the rep faster optimises that 3.4%. Two thirds of the waiting sits on the far side of a handoff — and no vendor sells that, because it is not a tool category.
From a pipeline you guess at to a layer you own.
The automations are the visible part. The thing that makes them possible — and that keeps paying after they ship — is the record of how your revenue team actually works.
The CRM reflects what somebody found time to type. The real state of the pipeline is in threads, in a sent folder, and in one person’s head. Adoption of the AI you bought is reported from seats.
The real work, reconstructed from the recording: where the hours go, where the deal is waiting, what the team actually does all day — ranked by impact, effort, risk and payback against your own numbers.
The top of the ranking ships: the thread-to-record fill, the follow-up draft, the handoff contract. Each one arrives as a specification drawn from observed steps, not from a workshop guess. Every draft is reviewed before it reaches a customer.
The agents are consumers of something more valuable: a private, structured record of how your company actually sells. Onboarding that teaches a new rep from what the best one really does. Search that answers from it. Whatever you build next, on data only you own.
Where AI actually works in sales — and where it stalls.
Everyone selling AI to revenue teams has a case study with a number in it. We are not going to quote you someone else’s, because we cannot verify it. Here are the patterns instead, including the failures.
The most reliable win, and the least exciting one. Where it works: filling the record from conversations that already happened. Where it stalls: teams that expect a clean deal board to fall out of it. What a trace honestly gives you first is an activity board.
Personalisation at scale is real, and it is also how inboxes died. The teams that win derive the message from what already converts; the teams that stall generate from nothing, and the output reads as generated and is deleted as generated.
High-frequency, low-judgement assembly in front of every conversation. Pays back where there is an existing library of context to retrieve from, and produces confident noise where there is not. The differentiator is retrieval, not generation.
Mechanical, repetitive, and stubbornly manual — because the last good version cannot be found. The bottleneck is almost never the writing. It is that nobody knows which proposal actually closed.
The loudest category, and the one we will not claim. Our own capture cannot measure meetings honestly — three quarters of what our naive metric called "meeting time" had no room attached. We killed the report rather than ship the chart.
Everyone has a CRM. Almost nobody has a CRM that reflects what was actually said in the messenger. The gap between the two is where pipeline quietly dies — and it is a data problem long before it is a model problem.
Pilots do not die because the models are bad. They die because nobody could say which of forty candidate processes to do first, or what "correct" looked like when it was done. That ranking is the thing Silow produces; the agents are what you build on top of it.
Nobody automates the waiting. It is not a tool category, so no vendor sells it — yet two thirds of the elapsed time in the team we recorded sat in a queue between two people. It is the largest number on this page and the cheapest one to move.
The part other vendors leave out.
Two of the hypotheses we most wanted to be true were killed by our own data, and one of them is the single loudest category in sales AI. You should hear that from us here rather than discover it in month three.
The recording sees activity, never outcome. Whether the deal closed, whether the quarter landed — that is not in the data. Join one column of your outcomes to our trace and it becomes provable. Until then, anyone quoting you an ROI number on a sales team is guessing, and so would we be.
Three quarters of what our naive metric called "meeting time" had no room attached — one person’s 666 "meeting minutes" were a parked video-call landing page. That is a capture problem, not a metric problem, so we cut the meeting-ROI report instead of shipping the chart.
The pitch was "N people doing the same thing, here is the saving." The real answer across nine people over five weeks was 44 recoverable minutes — 0.22% of working time, with a median duplicate of fifteen seconds. There is no product there, so we do not sell one.
Numbers are computed before the model is allowed to write a sentence about them, so no figure gets invented — and in the SOP report all thirty supporting quotes were verified against the raw scenes, not against a summary. But a true number can still be attached to the wrong claim. We have caught exactly that twice — both times with eyes, not with the gate.
Weeks, not quarters.
Silow runs on the team’s existing machines and tools. No migration, no new system to adopt, no process change while we look. You choose the deployment mode — including fully air-gapped.
Scenes become tasks, tasks become work directions. Numbers are computed before a model writes a sentence about them, and every quote is checked back against the raw scene it came from.
Every candidate scored by impact, effort, risk and payback against your real hours — so the first thing you build is the one that pays back fastest, not the one that demos best.
Each opportunity arrives as a specification an engineer can build from: the procedure, the evidence, the scenes it came from. Not "this looks automatable" — the actual steps, observed.
What you do with the hours is yours.
Silow maps work, not workers. No productivity ranking, no rep scoring, no performance evaluation, no automated decision about anyone — not as a policy written afterwards, but because the unit of analysis is the process. Two weeks of recording, and the first ranked roadmap lands with the hours attached.