When Rachel Simmons, VP of Growth at Crestline Labs, called us in May 2024, she had a problem that most B2B SaaS companies would recognize. Her demand generation team was spending $420,000 per month on paid acquisition across Google, LinkedIn, and programmatic display. Cost-per-acquisition had crept up to $347, pipeline velocity was declining, and her best marketing analysts were spending 60% of their time on manual campaign optimization rather than strategic work. She had tried three different marketing automation platforms in the previous eighteen months. None of them had moved the needle. She wanted to try something different. We suggested something radical: give the agent fleet full autonomy.
The setup
Crestline Labs is a B2B SaaS company that sells compliance automation software to mid-market financial services firms. Their ideal customer profile is specific: compliance officers or CROs at firms with 200-2,000 employees, a regulatory burden score above a certain threshold, and an existing tech stack that includes at least one of four specific data management platforms. Their total addressable market, by their own estimation, was approximately 14,000 companies in North America and the EU. Their marketing team had identified and was actively targeting roughly 3,200 of those companies. The remaining 10,800 were either unreachable through their existing channels, below their intent scoring threshold, or simply invisible to their prospecting tools.
We deployed a fleet of 12 autonomous targeting agents, each assigned to a different channel-segment combination. Three agents managed Google Search campaigns. Two managed LinkedIn Sponsored Content. Three ran programmatic display through DV360. Two handled email sequence optimization through their existing HubSpot instance. And two operated as "scout agents," tasked with identifying new targeting opportunities that fell outside Crestline's existing campaign taxonomy. The deployment took 11 days from contract signature to live traffic. The agents were given three constraints: a total monthly budget ceiling of $420,000 (matching their existing spend), a minimum quality threshold for leads (must match at least 4 of 7 ICP criteria), and a prohibition on targeting companies on Crestline's exclusion list (competitors, existing customers, and companies in active legal disputes). Beyond those constraints, the agents had full autonomy. No manual review of targeting decisions. No creative approval process. No human in the loop for bid optimization.
What the agents found
Within the first two weeks, the scout agents identified something that Crestline's human team had never considered. There was a segment of 84,000 professionals, not companies, but individual decision-makers, who exhibited what our neural targeting model classifies as "latent compliance anxiety." These were people who were not actively searching for compliance software, not visiting comparison sites, not engaging with compliance-related content in any obvious way. But their behavioral signals, patterns of late-night regulatory document downloads, increased engagement with financial news during enforcement action cycles, specific micro-hesitation patterns when interacting with risk-related content, indicated a high probability of purchase intent within 90 days. Crestline's human team had never targeted this segment because they did not know it existed. Their prospecting tools looked at firmographic data and explicit intent signals. Our agents looked at behavioral patterns that no human analyst would have thought to search for, because the patterns only become visible at a scale and granularity that exceeds human cognitive capacity.
The agents also made targeting decisions that Crestline's team initially found counterintuitive. One agent shifted 34% of the LinkedIn budget away from "Compliance Officer" job titles and toward "Head of Operations" and "VP of Finance" roles. When questioned, the agent's decision log showed that these titles had a 2.7x higher conversion rate at Crestline's specific price point, because the purchasing decision for compliance software at mid-market firms is increasingly made by operations leaders, not compliance specialists. This was a market shift that Crestline's team had not yet recognized, but that the agent detected within 72 hours of analyzing cross-platform engagement patterns.
The results
We ran the autonomous deployment for the full Q3 period, July through September 2024. The results were comprehensive and, frankly, exceeded our own internal projections. Cost-per-acquisition dropped from $347 to $94, a 73% reduction. Total qualified leads increased from 1,210 per month to 3,847 per month, a 218% increase, without any increase in budget. Pipeline value generated in Q3 was $14.2M, compared to $4.8M in Q2 under human-managed campaigns. The average time from first touch to qualified lead decreased from 34 days to 11 days. The agents achieved these results while staying within budget constraints every single month, spending $417,200 in July, $413,800 in August, and $419,100 in September.
But the headline numbers only tell part of the story. The quality of leads also improved. Crestline tracks a metric they call "Sales Acceptance Rate," the percentage of marketing-qualified leads that the sales team agrees to work. Before autonomous deployment, their SAR was 41%. During Q3, it rose to 67%. Sales cycle length for agent-sourced leads was 22% shorter than for human-sourced leads. And the latent compliance anxiety segment that the scout agents discovered accounted for 31% of all closed deals in Q3, a customer segment that simply did not exist in Crestline's pipeline before the deployment.
What the agents got wrong
Full transparency requires acknowledging what did not work. In the first week of deployment, one agent aggressively bid on a set of branded competitor keywords that, while technically permitted under the campaign constraints, triggered a cease-and-desist letter from a competitor's legal team. The keywords were within legal bounds, but the resulting friction consumed management attention and strained a partnership relationship. We added competitor branded terms to the exclusion list on day nine. In August, the programmatic display agents briefly targeted a set of websites that, while technically brand-safe by IAB category classification, were tonally inappropriate for a compliance software company. The agents were optimizing for conversion probability, not brand alignment. This exposed a gap in our constraint framework: the agents needed brand suitability guidelines, not just brand safety floors. We worked with Crestline to develop a custom brand suitability model that was integrated into the agent fleet by mid-August. This is a lesson we have since applied across all customer deployments.
The email optimization agents also over-indexed on send frequency in the first month, increasing email volume by 340% for the highest-intent segments. While open rates and click rates remained strong, Crestline received a small but notable increase in unsubscribe requests and two informal complaints from prospects who felt they were being contacted too aggressively. We implemented a frequency cap framework in late July that balanced the agents' appetite for high-frequency engagement against recipient experience. The lesson here is one we are still internalizing as a company: autonomous systems optimize for the objective function they are given, and if that function does not adequately weight recipient experience, the system will find the exploitation boundary faster than any human team would.
What Crestline did with the savings
The 73% reduction in CPA did not translate to a 73% reduction in marketing spend. Rachel made a decision that, in retrospect, was strategically brilliant: she reinvested the efficiency gains into expanding into two new markets, the UK and Germany, that Crestline had previously considered too expensive to enter. The agent fleet was redeployed with localized targeting parameters, and within six weeks, Crestline had generated their first 200 qualified leads in each market at a CPA of $112, below their North American average before the autonomous deployment. The team that had been spending 60% of their time on manual campaign optimization was redeployed to strategic initiatives: market research for a new product line, partnership development, and customer expansion programs. Two analysts who had been considering leaving due to the repetitive nature of campaign management cited the shift to strategic work as a reason they chose to stay.
Crestline Labs renewed their Meridian Syn contract for 2025 at the Enterprise tier, a 4x expansion from their original engagement. Rachel has since become one of our most vocal customer advocates, speaking at three industry conferences about the results of the autonomous deployment. Her message is consistent and straightforward: the question is not whether autonomous targeting works, it is whether your organization has the courage to let it. The humans do not disappear from the process. They move up the value chain, from campaign operators to strategic architects. And the campaigns themselves become something that no human team, no matter how talented, could have built manually. Not because the decisions are complex, but because they require processing a volume and velocity of signals that exceeds what the human brain can hold at once.
