Atlas Robotics — Intent signals that doubled meeting rate
Stitching six fragmented intent sources into one account-priority score lifted AE meeting acceptance from 18% to 41% and influenced $62M in pipeline within six months.
Atlas Robotics had invested in every major intent platform on the market — Bombora, 6sense, G2, ZoomInfo, LinkedIn signals, plus their own product-usage telemetry — yet the sales team was acting on none of it. Each tool produced its own ranked list, scores didn't reconcile, and AEs had stopped opening the dashboards. Marketing was generating record MQL volume, but the meeting-acceptance rate had quietly collapsed from 31% to 18% over four quarters. Leadership knew the data existed; what was missing was a single, defensible answer to the question 'who should I call this week?'
The intent paradox in B2B
Enterprise B2B teams now sit on more buying-signal data than any generation before them — and convert less of it. The average revenue stack in 2026 ingests signals from 7–9 distinct platforms, but Forrester finds fewer than 18% of sellers trust any single one enough to prioritise their week around it. The result is signal fatigue: more dashboards, fewer decisions.
The underlying problem is methodological, not technological. Each vendor weights surge differently, scores at different cadences, and resolves accounts against different ID graphs. Stitched naively, the lists contradict each other. The teams that win are the ones that build a single account-priority score with clear, sales-readable provenance — not the ones that buy a tenth signal source.
Models alone don't move pipeline. What moves pipeline is a model that is trusted by the AE on Monday morning, that ranks fewer than 50 accounts, and that explains itself in one line per account. Anything more complex gets ignored.
Average intent / signal platforms in an enterprise revenue stack
Forrester Revenue Tech Census 2025
Of sellers trust any single intent platform enough to act on it weekly
Forrester
Higher conversion when a single composite score replaces multiple ranked lists
Bain GTM Benchmark 2025
"Every dollar Atlas was spending on intent was already in the building. The opportunity wasn't a tenth signal source — it was making the nine they had agree."
How we actually ran it.
Signal inventory and provenance audit
Map every signal currently flowing into the GTM stack and what it actually means.
Sat with each platform owner, documented refresh cadence, account-resolution method, scoring scale and known false-positive rate. Built a single signal-provenance matrix — six columns wide, 47 rows deep.
A one-page signal map signed off by ops, marketing and sales. Three platforms were retired in week 2.
Composite score design
Replace nine conflicting lists with one ranked account view.
Weighted each retained signal by historical correlation to closed-won. Layered in product-telemetry weight (the strongest single predictor for Atlas). Capped the output at 50 accounts per AE per week and forced an explanation string for each ('Why this account, why this week').
An account-priority score that ranked exactly the accounts AEs were willing to call.
Sales activation rhythm
Wire the score into the seller's existing workflow — no new tab.
Pushed the top 50 into Salesforce as a 'This Week' list, with a 90-second Loom from marketing explaining the top three. Added a one-click reject button that fed back into the model.
Adoption hit 86% in the first two weeks. Reject reasons became the most valuable training data we had.
Closed-loop retraining
Make the model better every week without a data-science queue.
Weekly retrain pulling in AE accept/reject feedback, meeting-acceptance outcomes, opportunity creation and stage progression. Drift monitoring alerted ops when any single signal's predictive weight moved more than 15%.
Within six weeks the model's precision on 'will accept meeting' rose from 0.34 to 0.61.
Buying-group expansion
Move from account-level to buying-group-level signal.
Used LinkedIn engagement plus product-usage at the user level to identify the 3–7 actual humans inside each prioritised account. Routed personalised outreach by role, not by account.
Per-account meeting yield jumped from 1.1 to 2.6 stakeholders engaged.
Pipeline attribution and governance
Prove the score is the cause, not the correlation.
Held out a control cohort of 10% of accounts that the team worked from legacy lists. Tracked downstream opportunity rate, deal velocity and ACV against the scored cohort.
Scored cohort produced 2.3x more opportunities at 1.4x average ACV. Board approved 2027 expansion.
The chronology of the work.
- Day 0
Kick-off with CRO, CMO and RevOps lead. Goal: triple weekly accepted meetings inside two quarters.
- Day 14
Signal inventory complete. Three platforms retired, $340K annual saving redirected to model build.
- Day 30
First composite score in Salesforce. AE adoption monitored daily.
- Day 60
Meeting-acceptance rate crosses 30% for the first time in 14 months.
- Day 90
Closed-loop retraining live. Precision on 'will accept' hits 0.55.
- Day 120
Buying-group routing rolled out across the top 200 accounts.
- Day 150
Meeting-acceptance hits 41%. $24M influenced pipeline crosses the line.
- Day 180
Programme transitions to BAU. Total influenced pipeline: $62M.
The three places most programmes die.
Buying a tenth signal source instead of stitching the nine
Diagnose first. Most enterprise stacks already contain enough signal — the missing layer is provenance and weighting, not data.
Building a beautiful dashboard that AEs never open
Deliver the score inside the workflow already in use (Salesforce, Outreach, Gong) with a one-line explanation per account. If it needs a new tab, it loses.
One-and-done model build with no retraining loop
Wire AE accept/reject signals back into a weekly retrain from day one. Model decay is the single largest reason composite scores lose credibility.
Six months in, Atlas had a single account-priority score trusted across marketing, sales and ops; a closed-loop retraining engine that improved every week; and a sales team that started Monday with a 50-account list they actually believed. Influenced pipeline crossed $62M, meeting-acceptance more than doubled, and the company retired three redundant intent platforms in the process.
AE meeting-acceptance rate
Influenced pipeline in 6 months
Model precision on 'will accept meeting'
We didn't need more data. We needed one number we could defend. Once we had it, the sales team stopped arguing with marketing and started calling.
Lessons we'll carry into the next programme.
Signal stitching beats signal buying — the next platform is almost never the answer.
Account-priority scores live or die on AE trust. If you can't explain the rank in one sentence, retire the model.
Cap weekly lists at 50 accounts. Any more and the list becomes wallpaper.
The reject button is the most valuable training data you will ever capture.
Hold out a control cohort or you'll never know whether the model is causing the lift.
Intelligence is not a dashboard. It is a single, defensible answer to 'who do I call this week?' — delivered into the workflow the seller already uses, and improved every week by what the seller does next.
Want a programme like this one?
Tell us where you want pipeline to come from next quarter — we'll show you how the next 90 days could look.