Skip to content

Martech Integration

Frequently Asked Questions: Your Stack Isn’t Smart…Yet

How can I be sure our martech stack is integrated for AI?

Most martech stacks pass data; few decide. An AI-connected stack runs on real-time decisioning: the CDP unifies signals, the CRM exposes outcomes, agents choose the next best move and fire it through your automation—now, not next Tuesday. If your “integration” can’t change a message mid-session, you have plumbing, not personalisation. See how leading vendor Braze are rebuilding around agentic decisioning (Braze + OfferFit).

This acquisition builds on Braze’s previously announced development of a native AI agent codenamed Project Catalyst, which is designed to help brands personalize and optimize experiences with highly relevant journeys and content at scale.

What’s the blunt difference between yesterday’s CDP and tomorrow’s?

Yesterday: warehouse-adjacent silo with exports.
Tomorrow: decisioning layer on live data, converging with orchestration to test, choose and trigger actions across channels. That’s not hype; it’s the direction of market guides and vendor roadmaps. If your CDP can’t reason on streaming signals, you’re redecorating a museum.

Where do AI agents live in my Martech? In the CDP, CRM or MAP?

Wrong question. They live across them. “Agentic marketing” is the re-architecture: agents read the CDP, reconcile identity and intent, update the CRM, and act through email/push/web/ads continuously. Enterprise platforms are formalising this. Salesforce “Next-Gen/Agentforce” is a complete agentic marketing solution designed to help you personalise the right moments. We are simply transitioning from playbooks to prompts.

What proof is there that agents beat headcount?

I love this question as it is so divisive! They don’t. They coexist. Agents are there to optimise headcount, perform consistently without restriction, and deliver unbridled responsiveness and velocity by actively reasoning, planning, and executing actions autonomously. The coexistence model views agentic AI as a fundamental shift toward a hybrid workforce where agents optimise operations and outperform traditional “headcount” efficiency. The reason why they complement rather than ‘beat’ headcount is connected to their fundamental function. Agents are wonderful at prediction (forecasting ‘what next’ and measuring reliability), whilst judgment is a high-value capability reserved for humans. Agents can however, reason, plan, use tools, and, most of all, act. For a superb explaination, watch this video from Tim Sanders, G2 Innovation Officer and Harvard Business School Executive Fellow.

Isn’t my CRM already “AI-enabled”?

If “AI-enabled” means predictive fields nobody trusts or superficial enrichment, sure. Real personalisation requires closed-loop learning: agents propose, automation executes, CRM outcomes update the CDP, and the next agent changes its mind. If your CRM can’t feed decisioning (not dashboards) back into journeys, your “AI” is a screensaver.

What must be true in the data layer before agents stop hallucinating?

Three non-negotiables:

  • Event timeliness (sub-second where it matters).
  • Identity resolution with deterministic anchors.
  • Guarded features: consent state, sensitive attributes masked or computed at the edge.
    Vendors and analysts agree the CDP’s future is AI-first, composable, and live—not batchy Franken-exports. If your “unified profile” updates hourly, you’re already late.

“Personalization “that finally works” requires AI embedded in every step of the customer data-to-execution workflow.” – Simon AI

What does “agentic creative” actually change?

You stop briefing assets once a quarter and start briefing behaviours hourly. Agents don’t just choose who/when; they adapt what/how based on lift. This is why decisioning acquisitions matter: multi-agent testing turns messaging into an always-on experiment, not a brand calendar. If your MAP can’t run 1,000 micro-experiments/week, it’s a bottleneck.

Isn’t this risky for privacy?

It’s risky without privacy. The bar is consented, contextual, and reversible. Edge techniques and consent-aware profiles let agents learn from patterns without moving raw PII everywhere. If your team can’t demonstrate how consent state flows through decisions, you don’t have “personalisation”—you have exposure. (Even the bullish voices on agentic AI keep repeating: it only works when data context is right.)

What are the biggest red flags that my “AI personalisation” is theatre?

  • Batch brain: segments rebuild nightly; messages ship tomorrow.
  • Orphans: CDP, CRM, MAP all “AI-enabled,” none share outcomes.
  • KPI myopia: you track opens, not decision win-rates.
  • Governance shrug: no consent state in the feature store.

If you see any two, pause spend and rebuild the loop.

What metrics prove agents are creating value (not just activity)?

  • Decision Coverage: % of touchpoints where agents choose, not rules.
  • Latency to Lift: time from new signal → deployed variant with measurable lift.
  • Win-rate by Context: agent’s success per audience, channel, moment.
  • ROT (Return on Time): hours saved by agents vs. manual ops; reallocated to creative/strategy.

Platforms positioning around “real-time personalisation” make this measurable; demand to see it in your instance. Salesforce cover the basics in this Complete Guide to Real-Time Personalisation.

If I only copy one move from the leaders, what is it?

Put decisioning where the data lives. That’s the through-line of every serious move this year: mParticle inside transaction moments with Rokt; multi-agent decisioning inside Braze; Slack becoming the agentic control room on top of Salesforce Data Cloud. Stop trucking CSVs. Move the thinking to the data.

What does a no-nonsense 90-day plan look like?

Days 1–30: Instrument & expose

  • Map top 10 decision points (signup, cart, churn signals).
  • Pipe events into CDP with consent flags; expose a read API to agents.
  • Turn on one agentic loop (e.g., abandoned cart or post-demo follow-up).

Days 31–60: Close the loop

  • Write outcomes back to CRM with win/loss reasons.
  • Automate variant testing via your MAP; require lift or kill.
  • Ship an internal Agent Report: decision coverage, latency to lift, ROT.

Days 61–90: Scale & secure

  • Add two more high-value loops (renewal saves, pricing nudges).
  • Bake governance into code: add consent metadata in the feature store
  • Tie incentives to decision win-rate, not volume sent.

If your vendor can’t support this schedule, you’re not under-resourced—you’re over-tooled.

Executive cheatsheet: vendor reality check (ask these in the demo)

  • “Show me a live profile update triggering an agent decision in under 1 second.” (No pre-recorded clicks.)
  • “Prove you can write outcomes back to CRM and change the next journey automatically.”
  • “Where does consent live at decision time? Show the field. Now revoke it.”
  • “How do you calculate decision win-rate—and who sees it besides the sales team?”
  • If you get deckware, leave.

Enjoy this content? Also try Frequently Asked Questions: Agentic AI. Why It’s the Big Shift, Not Just Another Tool.