Podcast episode

21 Touches: Modern Performance Marketing

by Ken Rona ·August 6, 2026 ·Aperiam

ai-in-adtech orchestration performance-marketing programmatic walled-gardens

TL;DR

Anna Stepura, co-founder and CEO of Adlyse (a recent Aperiam Ventures investment), joins co-hosts Corey Ferengul and Joe Zawadzki to argue that performance marketing has hit a complexity wall — 21 cross-platform touchpoints to convert a customer versus the old rule of seven — and that an agentic OS sitting above Google, Meta, TikTok, and OpenAI is the only viable response. The episode is essentially a founder pitch in podcast form, with genuine depth on agent architecture, creative optimization loops, and the emerging OpenAI ads channel, but listeners should weight claims accordingly given the host-investor relationship.

What was covered

The "agentic OS" thesis: Adlyse positions itself as a neutral optimization layer above individual ad platforms — Google, Meta, TikTok, OpenAI, and Microsoft Advertising — rather than inside any one of them. Anna Stepura's core argument is that each platform's native AI is structurally incentivized to maximize spend on that platform, not to maximize return across all channels.
Platform-native AI bias: Adlyse deliberately ignores recommendations served by individual platforms, running instead more than 20 proprietary reasoning models and generating roughly 100 competing hypotheses before surfacing a single optimization recommendation to a human approver.
The 21-touch figure: Stepura cited a jump from the traditional "seven touchpoints to convert" benchmark to 21 cross-platform touches today, which she expects to keep rising — framing this as the structural driver for why single-platform optimization is no longer sufficient.
Human-in-the-loop to autopilot arc: Adlyse described a roughly 24-day trust-building period (illustrated with a Silicon Valley eyelash brand) during which clients move from reviewing every recommendation manually, to approving recommendations without double-checking, to enabling autopilot with guardrails (e.g., flag decisions above a cost threshold).
OpenAI as an emerging ad channel: Adlyse recently launched OpenAI ads as its fifth channel alongside Google, Meta, TikTok, and Microsoft. Stepura reported that CPMs (cost per thousand impressions, the standard rate card for ad pricing) on OpenAI are declining and targeting is improving; every existing client asked to activate the channel immediately.
Agent-to-agent advertising: The episode touched on a near-future scenario where AI buying agents communicate directly with AI selling agents — a structural shift that Stepura and Zawadzki agree the platforms are leaning into by opening, not closing, their APIs and MCP/CLI tooling.
Creative optimization loop: Adlyse analyzes why a creative asset is or isn't working (not just that it isn't), then uses underlying LLMs (large language models, the AI systems that power tools like ChatGPT) to remix variations and immediately A/B test them, with the loop running faster than a human could action it.
Entry threshold and scale: Stepura cited ~$50,000/month in ad spend as the practical floor where Adlyse's value becomes tangible, with larger teams spending millions per month per platform being the core target.

Notable claims & predictions

Anna Stepura: "It now takes 21 different touches to convert a customer, up from seven, and I believe that number will only increase." — Frames the entire product rationale; no primary source cited.
Anna Stepura: "We run about a hundred different hypotheses on what could be done in a scenario before we select the winning one and ask a human to approve it." — A specific architectural claim about Adlyse's reasoning pipeline.
Anna Stepura on platform APIs: "There is no need for platforms to close their APIs — it's actually vice versa. New platforms come already with super-structured APIs designed specifically so agents can interact with them easily." — Argues agentic buying is welcomed, not threatened, by the walled gardens.
Anna Stepura on OpenAI ads: "CPMs inside OpenAI are getting lower and lower, and results are more segmented. I still believe it's a new channel of opportunity — it's not better than other places, it's just different." — Early-stage signal from priority access, not a large sample.
Joe Zawadzki: "I could see somebody trying to make a run — we at Meta have tuned the most sophisticated agents; you should use your Meta agent when you're running on Google. I don't think it works, but I wouldn't be surprised if you had skirmishes in two years where someone tries to go closed." — A structural prediction about platform consolidation attempts in agentic buying.
Anna Stepura on autopilot timeline: "On average, we see about 40% of manual tasks optimized by Adlyse today; we are striving to achieve 90%." — A product roadmap target, not current capability.

Fact check

Claim — Stepura: "Seven touchpoints to convert" has risen to 21. Unverified and context-stripped. The "rule of seven" is a decades-old marketing heuristic with no single authoritative source; it was never a rigorously measured empirical finding. The "21 touches" figure is similarly unattributed — Stepura offers no study, survey, or data source. The directional argument (fragmented media = more touchpoints needed) is widely accepted, but the specific numbers should not be treated as established fact. Stepura is also talking her own book: a higher touchpoint count directly justifies the need for a cross-platform OS like Adlyse.

Claim — Stepura: Platform-native AI recommendations are biased toward higher spend. True as an incentive observation, but overstated as an engineering fact. It is accurate that platforms' business models benefit from higher advertiser spend, and there is documented practitioner skepticism about the neutrality of platform-native optimization tools (e.g., Meta's Advantage+ campaigns, Google's Smart Bidding). However, Stepura presents this as near-certain bias rather than a contested empirical question. Platforms argue their algorithms optimize for advertiser ROAS (return on ad spend) because retaining effective advertisers is also in their long-term interest. The incentive concern is real and worth flagging — Stepura's characterization is reasonable — but "they will never give you a recommendation that fully serves your business" is stronger than the evidence supports.

Claim — Stepura on OpenAI CPMs declining. Unverified — early-stage, small-sample signal. OpenAI's ad product is nascent; Adlyse had "priority access" and saw "only three buttons" in the first weeks. Any CPM trend from that data set is not statistically meaningful. Readers should treat this as directional anecdote, not market-rate intelligence.

Disclosure gap — host/investor relationship. Corey Ferengul and Joe Zawadzki are co-hosts of the Aperiam Ventures podcast, and the episode explicitly identifies Adlyse as "a recent Aperiam investment." Neither host challenges any of Stepura's product claims, market-size assertions, or benchmark figures. All three speakers are financially aligned. Listeners should apply normal pitch-deck skepticism to unverified performance claims (e.g., "improvement of 6k potential new leads or saving 10k" from a generic example).

Why this matters for ad-tech operators

Agentic buying is becoming a channel-design constraint, not just a marketer tool. Stepura's report that new ad platforms are launching with APIs pre-structured for agent interaction — not retrofitted — suggests SSPs (supply-side platforms, software publishers use to sell ad inventory)

Full analysis

Anna Stepura, co-founder and CEO of Adlyse, went on the Aperiam podcast with Corey Ferengul and Joe Zawadzki to pitch a thesis: performance marketing now takes 21 cross-platform touches to convert a customer, up from the old "rule of seven," and the only sane response is an agentic operating system that sits above Google, Meta, TikTok, Microsoft, and OpenAI and optimizes across all of them. This is a founder pitch on a friendly stage. Zawadzki and Ferengul are co-hosts and Adlyse is an Aperiam investment, so nobody in the room is playing defense.

What's actually being decided for the reader isn't "should I buy Adlyse." It's whether cross-platform agentic buying becomes a real layer in the stack, who owns it, and whether the walled gardens let it live. That's a slow-moving, Type 2 question for now (easy to test with a pilot, easy to walk back), but it hardens into Type 1 if platforms start playing gatekeeper. The forcing function is the OpenAI ad channel going live and every early client asking to switch it on.

The Market Analyst. Follow the incentive Stepura names: every platform's native AI optimizes for spend on that platform, not return across all of them. That's true, and it's the whole business case for a neutral layer. For an operator, this is the same fight independent ad tech has run for fifteen years, now dressed in agent language. The catch is timing. Adlyse claims OpenAI CPMs are falling and targeting improving, but that's a few weeks of priority-access data with "only three buttons" to press. Not a market read. The durable signal isn't Adlyse's numbers. It's Zawadzki's own prediction that within two years a Meta or Google tries to extend its agent across rivals and go closed. That's the risk that decides who wins.

The Skeptic. The 21-touch figure is unsourced, and so was the seven it replaces. The rule of seven was never measured; it was a slogan. Stacking a made-up 21 on a made-up seven and drawing a trend line is talking your own book, because a higher touch count is exactly what justifies buying a cross-platform OS. Strip that number out and the pitch still stands on the incentive argument, so why lead with a stat you can't defend? The "100 competing hypotheses" and "20 proprietary reasoning models" are architecture claims no client can audit. And "40% of manual tasks optimized today, striving for 90%" means the autopilot everyone's excited about is roughly two-fifths built.

The Operator. Here's what breaks Tuesday morning. Stepura's own example is a 24-day trust curve: clients review every recommendation, then stop double-checking, then flip on autopilot with a cost-threshold guardrail. Twenty-four days of a human babysitting an agent is a real integration cost, and it repeats per client. The $50,000-a-month spend floor tells you who this is for, and it isn't the mid-market. The second-order problem shows up at 90 days: when the cross-platform agent reallocates budget off Meta, Meta's own algorithm degrades on the starved campaign, and now you're fighting the platform's learning phase every time the OS moves money. Neutral optimization and platform learning phases don't like each other.

The Customer / End User. Put yourself in the agency or in-house performance seat. You already distrust Advantage+ and Smart Bidding for exactly the reason Stepura names, so a layer that ignores platform recommendations and arbitrates across channels is something buyers actually want. That's real demand, not projection. Every Adlyse client asked to turn on OpenAI ads immediately, which tells you appetite for new channels is high when someone else handles the plumbing. But the buyer is also handing budget-allocation logic to a black box run by a startup, on top of the black boxes they already resent. You've added a layer of trust, not removed one.

The CFO. The economics only work above real scale. At $50k a month minimum, and a core target spending millions per platform, this is an enterprise tool with an enterprise sales cycle and a 24-day onboarding drag before value shows. The payback question is whether the cross-platform lift beats what you'd get letting each platform's native AI run, and nothing in this episode measures that against a control. The strategic cost is worse than the license fee: if agentic buying consolidates and one platform goes closed, everyone who standardized on a neutral OS eats a switching cost. You're paying for optionality that a walled garden can revoke.

Where they part ways. The Analyst and the Customer see genuine, durable demand for neutrality. The Operator and the CFO see a tool that fights the platforms' own learning phases and only pays back at scale. And the whole thesis hinges on one belief the Skeptic won't grant and Zawadzki himself put a two-year clock on: that the walled gardens keep their APIs open. Stepura says new platforms ship agent-ready APIs by design, so openness is the trend. Zawadzki says someone tries to go closed within two years. They can't both be the base case.

What it comes down to. Two things decide this, and neither is the 21-touch stat. First, do the platforms stay open to third-party agents, or does one of them weaponize its agent across rivals and force a choice. Second, does cross-platform reallocation actually beat single-platform native AI net of the learning-phase penalty, measured against a real control. An operator testing this should ignore the founder's benchmarks and run a clean holdout: let Adlyse manage a slice of spend against a native-AI-managed slice, same budget, same window, and read the difference. The demand is real. The measurement is missing.

Prediction: Before the end of 2026, at least one of Google or Meta will ship an agent-buying or MCP-style interface that favors its own inventory or restricts third-party cross-platform agents from acting on equal terms, confirming Zawadzki's "someone goes closed" call ahead of his two-year window.

Confidence: Medium. Platform incentive to protect spend is structural and already visible in Advantage+ and Smart Bidding.

Why: The core mechanism the episode names, that each platform's AI optimizes for its own spend, is exactly the incentive that pushes a dominant platform to privilege its own agent once agentic buying carries real budget. Google and Meta have a long record of opening APIs to build adoption, then adding terms that advantage house tools once the channel matters, which is the same arc header bidding and Smart Bidding followed. Stepura's "APIs are opening, not closing" read is true for new, subscale platforms like OpenAI that need advertisers; it does not hold for incumbents defending mature revenue, and Zawadzki, an investor in the neutral-layer thesis, still flagged the closed-move risk himself. The opposite outcome, both giants staying fully neutral to third-party agents as budgets scale, runs against every incentive in their own business model.

Revisit by 2026-12-31: We're right if Google or Meta ships agent/MCP tooling with terms or defaults that favor its own inventory or limit equal third-party cross-platform buying. We're wrong if both keep agent APIs fully open and neutral to independent optimization layers through year-end.

Listen to the episode →