Performance marketers are being pushed to let algorithms manage execution while still proving that every dollar drives measurable business value. At the September MarTech Conference, three practitioners — Maria Corcoran of Jiffy.com, Anthony Tedesco of Cisco Systems and Jiaxi Zhu of Google — shared how to do that without giving up accountability.
Clear objectives become the real control layer
Giving up granular control does not mean giving up direction. Corcoran said AI needs more than campaign settings; it must understand how the target audience, product catalog, website and surrounding content connect. Zhu added that algorithms cannot maximize every metric at once, so defining the primary objective is the critical step.
“As long as you’re meeting that goal, with AI or not with AI becomes a secondary question,” Zhu noted.
For B2B programs, Tedesco flagged a nuance: the final conversion event, such as a closed deal, may occur too rarely to train data-hungry models. The fix is to identify proxy signals that fire frequently enough for optimization while keeping campaigns tied to pipeline and revenue.
Where automation works — and where it needs a leash
Some tasks are natural fits for automation. Real-time bidding and creative asset assembly let models evaluate thousands of contextual signals faster than any human. But even strong tools require watchful oversight.
Corcoran described a Google Performance Max test at Jiffy.com that delivered solid overall ROI but shifted budget disproportionately toward one of the company’s four business lines — one that hadn’t funded the initiative. Rather than abandon the tool, the team recognized that Jiffy’s site architecture wasn’t clearly differentiating its service lines to AI models. A campaign trial became a site-structure improvement project.
Proven metrics still anchor AI decisions
As search evolves, new visibility signals are emerging. Tedesco said Cisco tracks AI visibility metrics to see how large language models interpret and cite its content. But traditional funnel metrics still serve as reliable anchors.
“You don’t necessarily need to reinvent the wheel,” Tedesco noted.
Corcoran continues to rely on LTV:CAC and cost per acquisition while using AI to speed cross-channel analysis, spot attribution gaps and assess how influencer or user-generated content affects performance.
Operational gains may come first
For many teams, the most immediate AI payoff is removing repetitive work. Examples from the panel:
- Ad trafficking: a rules-based task that can become a push-button process.
- Self-service analytics: natural-language tools surface insights that previously required SQL expertise or long analytics queues.
- Reporting: Corcoran uses Claude to unify financial, ad, site and sales data, eliminating about three hours of daily reporting.
- Advanced analysis: running n-grams or correlation checks no longer requires a data science degree.
Adoption is not success
A conference poll showed 58% of attendees are experimenting with AI for performance analysis, 23% are exploring use cases, and just 11% have fully implemented it. Zhu cautioned against measuring progress by tool adoption or platform log-ins. The real indicator is whether AI applications improve actual business outcomes — so establish clear benchmarks before you scale what works.
Set guardrails, not unlimited access
Tedesco said the goal is to find “that balance of automation and autonomy that makes sense for your business.” Clean data taxonomy gives AI systems the structure they need. Corcoran recommends starting AI in background analytics before letting automated tools touch active, market-facing budgets.
The core discipline hasn’t changed: reach the right audience, deliver relevant messaging and drive growth. AI processes data faster, but marketers still set direction, validate data and define boundaries.
Source: MarTech
