The strategy work happens before the agent does anything
Most guidance on building an agentic marketing strategy skips to the exciting part: set a high-level goal, let autonomous agents plan and execute, and reassign your team to "strategy." The pattern is everywhere.
Agentic marketing gets defined as the deployment of autonomous AI agents capable of planning, executing, and optimizing customer engagement workflows to achieve specific business goals—operating independently, making real-time decisions across channels to maximize ROI.
The implication is that the hard part is the autonomy.
It isn't. The hard part is everything the agent stands on. An agent that can plan a winback campaign but doesn't know which customers already re-engaged, what offers compliance allows, or which shade of your brand blue is approved will execute confidently and wrongly. The autonomy is the easy 20%. The foundation is the 80% nobody wants to write a strategy doc about.
So treat the term carefully. An agentic marketing strategy is less a plan for what agents will do and more a plan for what agents will know. Get the knowledge substrate right and execution becomes a configuration problem. Get it wrong and you've automated the production of off-target, off-brand work at a speed no human can catch.
Why "set a goal and let it run" quietly assumes a miracle
The reigning playbook has a seductive structure: define guardrails, run a pilot, scale across channels, and convert campaign managers into agent supervisors.
The advice is to establish goal-based guardrails, tell the agent its discount limits, frequency caps, and excluded segments, let it experiment within that perimeter, then expand its mandate across all channels once the pilot proves ROI.
Reasonable steps. They just assume the agent already has trustworthy inputs.
That assumption is where it breaks. Even vendors enthusiastic about autonomy concede the point:
unified behavioral, transactional, and other data has to feed one real-time source of truth, because without unified data, autonomy collapses—agents can't act on incomplete context.
The undercurrent beneath every agentic pilot is the same anxiety that has dogged marketing for a decade—the data is fragmented, and now you're about to let software act on it at machine speed.
There's a second gap the consensus barely mentions. Goal-and-guardrail framing handles who and whether, but not whether it's on-brand. An agent told to "reduce cart abandonment by 15%" can hit that number with a discount code your finance team never approved, a product claim your legal team would kill, and a tone that reads nothing like your brand. Guardrails as commonly described are financial and audience-level fences. They are not brand knowledge.
Agents need two foundations, not one
Here is the criterion that separates a durable agentic marketing strategy from a demo. Useful agent output depends on two distinct foundations, and most teams build only the first.
The first is governed customer data: unified, identity-resolved, and current. This is the foundation the market already talks about, and it's why the data warehouse keeps surfacing as the natural home for agentic systems—it's where the complete picture of the customer already lives.
The relevant signals are each customer's current behavior, lifecycle stage, value, and propensities at the moment of activation
, and an agent reasoning without them is guessing.
The second foundation is operational brand knowledge, and it's the one almost everyone skips. Brand guidelines, approved claims, voice and visual rules, product facts—agents need these structured as a queryable context layer they can reason against in real time, not a 60-page PDF sitting in a shared drive. The reason this matters is concrete. Teams evaluating general-purpose AI for marketing keep hitting the same wall:
the AI gets colors wrong, hallucinates products, and doesn't meet the brand bar.
The two foundations fail in opposite, equally useless directions. Data without brand knowledge produces output that's accurate about the customer but wrong about the company—right audience, off-brand message. Brand knowledge without data produces output that's perfectly on-brand and aimed at the wrong person. A strategy that funds one and ignores the other is a strategy for fast, confident mistakes.
This is also why the organizational design conversation is converging here. The Harvard Business Review's framing for the agentic age centers on what it calls a
"brand code": a machine-readable knowledge base encoding brand strategy, customer insights, and business rules that both people and AI agents can act on.
Different name, same idea: the strategy is the substrate.
What to actually evaluate when you build the stack
Once you accept that the foundations are the strategy, the platform questions change. You stop asking "how autonomous are the agents?" and start asking "what do the agents reason against, and where does that knowledge live?"
A few criteria worth pressure-testing with any vendor:
Does customer data stay in your warehouse, or get copied into theirs? Architectures that ingest and store a separate copy of your data create a second source of truth that drifts from the first and adds a migration before you've launched anything. The warehouse-native alternative avoids this. Platforms such as Hightouch run on a composable CDP approach wheredata is activated directly from your existing cloud warehouse instead of ingesting a separate copy—meaning no duplication, no multi-month implementation, and the warehouse stays the single source of truth.
Is brand knowledge a real, queryable layer or an afterthought? Ask how the system keeps creative on-brand at volume. The serious answers involve grounding generation in your actual assets and grading the output. One approach pairs models with a brand context layer thatlearns from existing assets, uses automated judges to grade outputs, learns from user feedback, and keeps generations on-brand.
Do you have to rip out your stack to adopt agents? Beware migration tolls. The healthier posture treats agents as portable. Hightouch's agents, for instance, are positioned as an independent product line—they operate independently of the CDP, so you don't need the full customer data platform to use them in your existing stack.
Whether or not you choose that vendor, the principle holds: adopting agents shouldn't require a forced platform migration.
Where does AI run relative to your data? Features that require customer data to leave your infrastructure trade governance for convenience. Keep that trade explicit in the evaluation.Notice that none of these are questions about how clever the model is. The model is the commodity. The context is the moat.
What a working loop looks like in practice
Strategy gets abstract fast, so make it concrete with a single lifecycle use case: reactivating lapsed customers.
In the old model, a marketer builds a segment, picks one offer, schedules a send, and waits for an end-of-month report. Everyone in the segment gets treated the same, which is the original sin—
these tactics carry baked-in assumptions that customer behavior is predictable and that everyone in a segment responds the same.
In an agentic loop, the work shifts. The marketer defines eligibility, allowed offers, channels, and the outcome to optimize. From there the system decides at the individual level. A practical illustration:
an agent might decide a lapsed customer should get a 10% winback offer, but only if they haven't already re-engaged elsewhere, only on SMS based on prior response, and only in a time window when they're historically likely to convert.
That's the difference between a rule and a decision.
This is the territory Hightouch AI Decisioning—a capability inside its Lifecycle Marketing Studio—operates in.
It uses reinforcement learning to determine the best message, offer, channel, creative, timing, and frequency for each customer on a 1:1 basis, including whether to send at all.
The control stays with the marketer by design:
you authorize what actions the agent can take, define what content to use, and set thresholds to balance performance with send volume, so the system optimizes within your brand's strategy.
The loop only closes because both foundations are present. The data tells the agent who this customer is and what they've done. The brand and offer rules tell it what it's allowed to say and give away. Remove either and the loop produces noise. Keep both and it compounds—
every decision is measured against a holdout group and your defined metrics, and the system learns from each interaction to improve future decisions.
What good looks like, in numbers and in roles
The point of all this isn't autonomy for its own sake. It's lift you can measure against a control group, and time returned to people for the work machines can't do.
The measurable side shows up when the foundations are solid. One specialty retailer with a large loyalty base put a decisioning agent to work on a single goal and saw results in weeks:
PetSmart, with more than 70 million loyalty members, used AI Decisioning to increase incremental salon bookings by 22% within three weeks.
The number matters less than the shape of it—a narrow, well-instrumented goal on top of trustworthy data, not a sweeping "automate marketing" mandate.
The human side is where the role change everyone predicts actually lands, and it's more grounded than "campaign managers become AI strategists." The realistic description of the future marketer is
a generalist with great taste, judgment, and creativity, who uses agents to execute at light speed.
Taste and judgment are exactly the inputs that build and maintain the brand-knowledge foundation. The strategy work doesn't disappear into the agent. It moves upstream, into defining what the agent knows and what good looks like.
Worth naming a real limit, too: feedback loops are only as fast as your data lets them be. If campaign outcomes have to travel from a downstream tool back into the warehouse before an agent can learn from them, the cycle slows. That latency is a legitimate thing to interrogate during evaluation, not paper over—the speed of learning is part of the foundation, not a detail.
The strategy is the substrate
Building an agentic marketing strategy is the most important infrastructure decision marketing teams will make this cycle, and the consensus advice quietly skips the part that determines whether any of it works. Autonomy is not the strategy. The strategy is the two foundations underneath the autonomy: governed customer data that keeps agents aimed at the right people, and operational brand knowledge that keeps them sounding like you.
Evaluate vendors on where your data lives, whether brand knowledge is a queryable layer or a forgotten PDF, how fast the learning loop closes, and whether adoption forces a migration. Pilot on one narrow, measurable goal where both foundations are already strong. Then expand. The teams that win the agentic era won't be the ones with the most autonomous agents. They'll be the ones whose agents had something worth reasoning from.
For a deeper look, writeup of its agentic marketing platform is a useful reference point.