Signal Analysis: Agentic Search Is Now the Top‑of‑Funnel — Your SFMC, Braze, and Iterable Journeys Must Start in Chat

Salesforce just quantified what your metrics have hinted at all quarter: a 200% year‑over‑year surge in agentic search as the first step of a purchase journey (Fourth Edition State of Commerce, July 28, 2026). Shoppers now begin with a question to an AI — in an LLM or a retailer’s assistant — not a search box or nav bar. That’s not a vibe shift; it’s a funnel rewrite. Source: Salesforce Newsroom.

What changed — and why it matters

Agentic search collapses discovery and consideration into one conversational moment. The LLM filters what you used to earn with category pages, ad groups, and product grids. If your lifecycle stack still treats “session start” as web traffic or email click, your attribution and journey logic are already outdated.

• The top‑of‑funnel is a chat window. That first question captures intent, constraints, and objections in natural language — data your CDP and journeys rarely see today.
• Buyers expect resolution, not redirects. If your assistant can’t quote availability, compare SKUs, and issue incentives in‑thread, you’ll leak to a marketplace agent that can. TikTok, for example, is building agentic ad flows to transact in‑stream (Adweek, Jul 29, 2026).
• Platforms are moving to agent KPIs. Salesforce has been steering customers to agentic service and marketing metrics; see its Missionforce AELA announcement on Jul 24, 2026 — a multiyear bet that agentic flows are the operating model, not a pilot (Salesforce Press Release).

The stack impact for SFMC, Braze, and Iterable teams

You don’t need a new platform. You do need to refactor events, identity, and content so your assistant and lifecycle tools make the same decision at the same time.

1. Treat the first chat as your primary entry event

• Create a canonical “AgenticSessionStarted” event in your CDP or event bus with fields for question, inferred intent, constraints (budget, size, timeline), and channel (LLM host vs. on‑site assistant).
• In SFMC, map this into Contact Builder and Journey Builder entry with a data extension keyed on a durable ID (not just email). In Braze, use a custom event with properties to start a Canvas. In Iterable, use a custom event to trigger a Workflow with a metadata guard.

2. Right‑size identity from day one

• Map anonymous chat to known profiles with deterministic hooks early: phone for OTP in SMS‑led flows, account SSO for B2B, or consented email capture when the assistant offers value (e.g., save configuration). Failing here breaks suppression and compliance downstream. We’ve seen opt‑outs misfire when teams tie consent to the first email send instead of the first agent interaction.

3. Move offers into structured, agent‑addressable content

• Long‑form modules won’t survive a chat UX. Store offers as atomic, API‑addressable snippets with validity windows, eligibility criteria, and copy variants. SFMC Content Builder with data‑driven blocks, Braze Content Blocks, or Iterable Catalogs are all viable — but only if the assistant can call them.

4. Measure resolutions, not sends

• If your dashboard tops out at opens and clicks, you’ll optimize the wrong moment. Track “Agentic Resolution Rate” (e.g., problem solved, product matched, appointment booked) as the north‑star KPI. Salesforce service benchmarks already cite resolution outcomes across millions of agentic conversations — marketing is next.

5. Build guardrails for model drift and compliance

• Marketing ops must own prompt governance, PII handling, and fallback paths when models change. Analysts are warning about model churn risk for marketers (Marketing AI Institute, Jul 9, 2026). Bake observability and rollbacks into your journeys before an incident.

Play the channel chessboard the way agents do

• Search: Expect fewer SEO landings, more “best‑for‑me” chat queries. Feed assistants structured product knowledge, availability, and policy. Schema beats meta description.
• Email/SMS/Push: Use these to confirm, checkpoint, or re‑engage stalled threads, not to re‑pitch. State‑aware nudges (“We saved your size 9 trail shoe; black is low in stock”) beat generic promos.
• On‑site: Make chat feel like your PDP: specs, bundles, promos, service slots — all callable in‑thread.

What good looks like in your platforms

• SFMC: Journey Builder entries from the AgenticSessionStarted DE, Decision Splits on intent, Content Builder blocks sourced from a promotions DE with eligibility rules, and Event Notification Service back to the assistant when the journey updates state.
• Braze: Canvas start from custom event, Connected Content for pricing/availability, Intelligence Suite for send‑time and channel mix, and a Canvas Exit when the agent resolves to prevent channel collision.
• Iterable: Workflow triggers on custom event, Catalog for atomic offers, in‑app and push tied to agent state, and Journey Guardrails to cap touches if the assistant is live with the user.

Common failure modes we’re fixing now

• Identity gaps: Anonymous chat stitched to the wrong profile, causing double messaging. Fix with deterministic keys and conflict policies.
• Content mismatch: Assistant recommends SKU X; email promotes SKU Y. Solve with a single offer_id flowing from assistant to lifecycle platforms.
• KPI myopia: Teams celebrate click‑through while the assistant already closed the sale. Route resolution events into marketing analytics so attribution follows the outcome.

Key takeaway

Agentic search moved the first click to the first question. If your SFMC, Braze, or Iterable program can’t read, respond to, and measure that first question, you’re optimizing yesterday’s funnel. Start by defining an agentic entry event, normalizing identity, and making offers callable by an assistant.

For more on how agentic orchestration is reshaping lifecycle KPIs and architecture, see: Salesforce’s AI Foundry Is the Real Release Note — Model Wars Are Over, System Design Wins and Salesforce’s ‘Help Agent’ Pricing Will Force Marketers to Measure Resolutions — Not Sends.

If your journeys still assume traffic starts on web or email, that’s the migration and guardrail work we do. If your SFMC, Braze, or Iterable instance is hitting agentic entry‑point headaches, we’ll map the fixes in a working session.