Summary of Greylock Presents: The Intelligent Marketer

Overview of Greylock Presents: The Intelligent Marketer

Episode hosts Mike DeBow (Greylock) and Rishabh Jain (Fermat) interview Eric Seufert (Mobile Dev Memo / advisor & investor) about how AI, privacy changes, and new platform tooling are reshaping marketing, measurement, commerce and ad formats. The conversation covers post-ATT measurement, first‑party signal engineering, personalization and dynamic bundling, how LLMs will become ad networks, new AI-native ad formats, and the evolving roles of media buyers and agencies.

Key themes and high-level takeaways

• ATT catalyzed a fundamental shift from deterministic last‑click attribution to probabilistic measurement and lift-based approaches — and that shift is here to stay.
• Platforms and their tooling (e.g., Meta Advantage Plus, Google PMAX) are the main accelerants of adoption for probabilistic measurement — advertisers will follow where the tools simplify operations.
• First‑party signal engineering (fewer, higher‑density signals) is the frontier for performance gains.
• AI enables down‑funnel personalization (dynamic pricing/bundling/merchandising), which will be a major growth lever for commerce.
• LLMs/chat interfaces are inevitably going to become ad platforms — the right approach is ads alongside high‑quality answers, not ads that degrade trust.
• Agencies and media buyers won’t disappear; their roles will shift toward product, economics, orchestration and strategy while AI automates routine operational work.

Measurement & the post‑ATT landscape

• What changed: ATT made deterministic attribution unreliable. That pushed teams to lift tests, geo holdouts, MMMs and probabilistic approaches.
• “Privacy shuffle”: platforms are calibrating privacy moves (e.g., Google & cookies), but the industry’s move to probabilistic measurement is durable because it’s simply better.
• Winners: legitimate sources of incremental traffic; losers: fraudy traffic and teams that cling to last‑touch metrics.
• Adoption stage: early adopter phase, but platform tools that obfuscate channels (Advantage Plus, PMAX) force advertisers to adopt probabilistic measurement to evaluate performance.

Notable insight (Eric): “If measurement gets better broadly, everyone can invest more — it’s not zero‑sum.”

First‑party data and signal engineering

• Best practice: focus on “signal engineering” — send high‑density, predictive signals back to platforms rather than lots of low‑value noise.
• Implementation work: CAPI (Conversions API) rollout and integrations are valuable but complex (multiple CAPIs, CDP/engineering work, data leakage concerns).
• Organizational implication: marketing teams need engineering support or partnerships to operationalize high‑quality server‑side signals.

Practical note: quality of inputs >> quantity; better inputs let platform models make much better predictions (compounding benefit).

Personalization, pricing and bundles (the personalization economy)

• AI makes real-time, individualized bundling and dynamic pricing much more feasible (especially for digital goods with zero marginal cost).
• Example: personalized bundles for games (different bundles/prices by predicted willingness to pay) — higher conversion and yield.
• Technical/opportunity implication: imagine a future where ad systems can trigger landing‑page A/Bs or generate bundles on the fly — platforms need APIs and two‑way signals to enable this.
• Recent regulatory/market moves (e.g., Apple injunctions, off‑platform payment options) create more ability to transact/offload commission and enable on‑site personalization.

Investor perspective: bundling and dynamic offers are a hugely underexploited growth lever; AI reduces friction of merchandising.

LLMs, AI overviews and “everything is an ad network”

• Current reality: getting surfaced in AI overviews (Google AI summaries) is already driving traffic; directly influencing models (e.g., ChatGPT training) has slow feedback loops.
• Inevitability: major LLM providers (OpenAI, others) will build ad platforms; hiring of ad-product people is an indicator.
• Best UX for ads: keep ads clearly separate from core answers — ads alongside best answers preserves trust. Don’t let ad optimization change the objective answer.
• Monetization strategy for new chat apps: ad monetization typically yields higher ARPU than subscriptions if you can do it well.

Eric’s advice: “It would be a horrible mistake to raise the perception that you might be curating the content to serve an ad — that will poison the well.”

New ad formats & what ad units might look like

Potential AI‑native ad formats discussed:

• Full‑screen, non‑skippable interstitials for freemium queries (YouTube-style takeover).
• In‑stream real‑time display insertions integrated into answers (akin to shopping cards between blocks of text).
• Fixed tower/sidebar units that remain visually distinct from answers to preserve trust.
• Video/avatar ads as chat interfaces evolve to persistent, real‑time video/voice agents.

Design principle: prioritize formats that preserve or enhance trust in the answer (separate but relevant ad placements).

Conversational/contextual signals vs. identity signals

• Existing ad networks are identity‑signal centric; AI/chat apps produce rich conversational/contextual signals that are valuable if consumed carefully.
• Companies with direct conversational engagement will retain that signal internally (it’s “gold”) rather than passing it to third parties.
• Programmatic infrastructure (DSP/SSP) can still function for AI ad inventory, but it must support conversational/contextual inputs and new formats.

Role of agencies, media buyers and teams

• Media buyers will shift away from tactical toggling to strategic orchestration: product integration, measurement strategy, economics and end‑to‑end growth.
• Agencies can gain by adopting AI tooling to surface trapped domain knowledge, automate junior work and deliver higher‑value strategy and customer economics.
• AI will agentize many operational tasks, increasing margins for agencies that incorporate AI as a control/coordination layer.

Eric’s perspective: platforms providing “tell us your goal, we’ll do everything” (Meta Advantage/PMAX) are changing media buyer workflows — the human role becomes more product/economics oriented.

The website and the commerce funnel

• New buyer journey: LLMs/answer engines push discovery and research earlier; by the time a user clicks through, they are often “sold.”
• Website role shifts from information‑first to transaction/checkout‑first; UX should be optimized for fast conversion, personalized bundles and being “transactable” by agents.
• Implication for merchants: prioritize being transactable (APIs, landing page optimization, payment flows) and prepare for commerce inside answer platforms (e.g., Shopify ↔ ChatGPT integrations).

Practical recommendations / action items for marketers

• Measurement: invest in probabilistic approaches, lift tests, geo holdouts and MMMs; stop relying on last‑click as ground truth.
• Signal engineering: audit your incoming/outgoing signals; build fewer but higher‑predictiveness signals back to platforms (server‑side events, CDP integrations).
• Implement CAPI and coordinate across engineering; avoid one‑off partial setups that leak value.
• Experiment with dynamic pricing and personalized bundles for high‑value cohorts — use AI to generate and test offers.
• Prepare for AI ad inventory: design ad units that sit alongside high‑quality answers (don’t compromise trust); prototype full‑screen/video ad experience.
• Protect and own conversational signals if you have them; don’t indiscriminately pass rich user context to third parties.
• Agencies: adopt AI tooling, automate repetitive tasks, and double down on strategy/product‑economics roles that tie acquisition to monetization.
• SEO/AI overviews: maintain baseline SEO plus consider paid strategies — if you need measurability and control, paid buys remain more direct and testable.

Notable quotes & concise insights

• “Everything is an ad network.” — encapsulates the view that any high‑attention surface (LLMs, retail media, CTV) becomes monetizable.
• “Wittgenstein’s ruler” for ads — deterministic attribution can be self‑referential and misleading; probabilistic measurement is healthier.
• “The best possible way to implement this is an ad alongside the best content.” — keep ads separate to preserve trust and answer quality.
• “By the time the users clicked on the link, they've been sold.” — LLMs/answer engines shift the funnel earlier; sites are often transactional endpoints.

Bottom line

The industry has moved: privacy changes accelerated a more rigorous, probabilistic measurement mindset and forced marketing to integrate product, monetization and measurement more tightly. AI (LLMs) will create new discovery surfaces and inevitably become ad networks; the smartest companies will (1) engineer high‑value first‑party signals, (2) experiment with AI‑enabled personalization and dynamic offers, (3) design ad experiences that preserve trust, and (4) reorient agency and media buyer skill sets toward orchestration, product economics and integration.