Case study 02 / AI trust and safety

From a list engine to an activation engine, in four months

The client is an AI trust and safety platform. Security, compliance and AI teams use it to evaluate, monitor and enforce the behavior of AI agents. A technical product sold to buyers who leave no signal on forms and punish generic outreach.

4 months
from the April audit to a standing GTM asset the team operates
100%
of 67,728 contacts source-attributed, funnel visible end to end
0.8%
bounce across 35 warmed domains and 104 mailboxes, zero spam flags
15–20 hrs
recovered per week, about half a GTM hire never made
Before / after

A CRM full of contacts nobody had ever touched

The April audit found roughly 59,000 CRM contacts, 56,000 of them with no recorded activity and no source-to-revenue attribution. The starting point is not our claim; it is what the audit measured.

Before / the April audit
After / four months later
Attribution
56,000 of 59,000 contacts with no recorded activity, no source-to-revenue path
67,728 contacts, 100% source-attributed, funnel visible end to end
Scoring
None. Every account looked the same on the way to a rep
4,176 buyers scored on awareness by readiness, 51 routed to sales as sales-ready
Sending
Cold email going out from the core domain, the company's own reputation at stake
35 warmed domains, 104 mailboxes, 0.8% bounce, zero spam flags, core domain protected
Signals
Manual checks in a sales prospecting tool, one person at a time
A live signal layer: web de-anonymization, competitor and hiring signals
The motion
Operator-dependent: prospecting tool, then CSV, then CRM, by hand
A documented campaign system with copy guardrails. The motion no longer depends on any one person
Deliverability is the slowest asset to build and the easiest to burn. It is built, and the core domain never carried a cold send.
Selectivity

27,800 qualified accounts, 51 worth a rep's Monday

Scoring on awareness by readiness is what turns a list into a queue. 45 of the 51 sales-ready accounts match the enterprise and regulated CISO to CIO profile.

Qualified TAM
~27,800
Scored buyers
4,176
In-market
747
Sales-ready
51
The funnel, now visible
Leads67,057
MQL296
SQL67
Opportunities300
Customers6
LinkedIn layer

~1,790 qualified prospects across two reps, 41 accepted connections in the first weeks and roughly 386 pending.

How many of your CRM contacts have never been touched? The diagnostic counts them in 5 to 10 working days.
If it doesn't surface at least 3 pipeline leaks you'd pay to fix, you don't pay.
Get the diagnostic
The offer test

The infrastructure passed the test. The offer did not.

Two cohorts ran on the same infrastructure, in the same weeks, at the same 0.8% bounce. An event invitation to security leaders returned 14 positive replies, 6 of them inside 48 hours. A product pitch to the broad technical ICP returned zero. On the 95-name CISO list the invitation ran at 50% opens and 46% clicks.

CohortSendsRepliesPositive
Security conference invitation
Security leaders
2,907
39
14
6 within 48 hours
Product pitch
Broad technical ICP
5,521
15
0
0 meetings
Twice the volume, half the replies, none of them positive. The control group that proved relevance beats reach.
Why we publish the zero

We read 22 case studies from the platforms and agencies in this category. None publish the campaign that failed. We publish the zero because the finding is the product: a system that tells you what to retire is worth more than a vendor that hides it.

What we do not claim

Two variables moved together, offer and segment, so we do not overclaim a single cause. None of this is booked pipeline yet. It is early evidence the direction is right, produced on infrastructure the client already owns.

What the four months produced

What changed was not activity. It was uncertainty.

Four months ago there were five open bets and no evidence for any of them. Each one now has a verdict or a cheap way to get one.

Ruled out
Product-led cold outbound

5,521 sends answered the question. The budget stops here.

Early traction
Event-led outreach and LinkedIn warming

14 positive replies and 41 accepted connections say keep going.

Queued
AI-mature ICP and gated inbound

Cheap, fast tests on infrastructure that is already paid for.

15 to 20 hours a week came back to the team, about half a GTM hire never made, and the campaign system is documented so the motion survives any one person leaving.
Client quote
Head of growth, AI trust and safety platform
How we measure

Every figure is pulled from the operating systems themselves: the CRMs, the sending platforms, the engagement tools, the Demand Compass™ ledger, with baselines drawn from the initial audits, and was shared with each client before being shared publicly. Reply classifications are human-reviewed; positives exclude auto-replies. Scored records retain per-lead evidence links (the signal, the post, the source), so claims can be audited row by row. We publish the cohorts that produced zero. The finding is the product. We do not report "influenced pipeline" and we do not publish multiples we can't show the math for.

See the demand you can't see. Then the leaks that lose it.

The audit shows you the demand you can't see. The diagnostic shows you why your system misses it. The engine makes sure you never miss it again.