Every marketing conference this year has a session on agentic AI. Almost none of them can tell you what that means on a Tuesday morning, beyond swapping the word “automation” for “agentic” on a slide. The label has arrived well ahead of the operating discipline, and the data backs that up: MIT’s NANDA research found that roughly 95% of enterprise generative AI pilots are producing no measurable P&L impact. That gap, between adoption and actual outcome, is exactly where most marketing teams are stuck right now.
Having spent the last year building Netcore’s own platform around this shift, and watching hundreds of brands attempt versions of it, I’ve come to believe the age of agentic marketing didn’t begin with a model release. It began the moment brands ran out of runway on an older idea: that acquisition is the whole game, and retention is whatever’s left in the budget after acquisition is funded.
Here’s what I think every marketing leader needs to rethink.
Stop treating retention as a cost centre.
For a decade, marketing budgets have quietly encoded a belief: finding a new customer is the real work, and keeping one is overhead. That belief made sense when acquisition was cheap, and customer data was hard to act on at scale. Neither of those things is true anymore. Acquisition costs have climbed for years, and most brands are sitting on more first-party data than they’ve ever used: purchase history, channel preference, lifecycle stage, sitting mostly idle in a warehouse.
The result is an odd, expensive habit: brands pay once to acquire a customer, watch them go quiet a few months later, then pay the same ad platforms again to win back someone whose entire history they already hold. That’s not an acquisition problem. It’s a retention system that was never built to notice the customer had gone quiet in the first place.
One operating layer beats seven dashboards.
The technical unlock underneath agentic marketing isn’t one clever model; it’s sequencing. Break the work of retention marketing into stages and you get roughly four jobs: surface what’s actually happening across the business instead of forcing a human to open six dashboards every morning, segment customers at a resolution no team could build by hand, hundreds of behavioural micro-segments instead of the ten or fifteen a marketer might realistically maintain, generate messaging for each of those segments inside a brand’s own tone and policy, and decide who gets which message, on which channel, and when.
Do those four jobs in isolation, on four different tools, and you’ve rebuilt the same fragmented stack with extra steps. Do them in sequence off one shared context layer, and each stage gets smarter because it can see what the last one just learned. That’s the actual difference between “we added an AI feature” and “we built an agentic platform,” and it’s a distinction most vendors right now are happy to blur.
We’ve watched this play out inside our own marketing team, not just in the product we sell. Skills built for customer journeys and customer stories, living inside the workflow our team already used, saw real, sustained adoption. A standalone app we built for SDR and AE positioning, living outside that workflow, saw none. Same underlying capability, different outcome, entirely explained by whether the work sat inside the sequence or bolted onto the side of it. And not every agent earns its keep just by existing: one person on our team once built an agent that sent me a morning Slack brief of coffee fun facts. Genuinely fun facts. Also genuinely a waste of tokens and hours that could have gone into account research or outreach. Agentic marketing is not “more agents”. It’s agents doing the four jobs above, in sequence, tied to an outcome someone actually asked for.
The model is not the moat. What you feed it is.
Every marketing platform has access to roughly the same handful of frontier models. That’s not a controversial statement anymore; it’s just table stakes. So if a vendor’s entire pitch is “we use AI”, ask the harder question: AI trained on whose judgement?
Differentiation was never going to sit in the model layer. It sits in the proprietary data and decisioning logic layered on top of it, years of which subject lines actually perform, which channels work in which markets, and how a lifecycle differs between a bank and a food delivery app. Combine that with a customer’s own first-party data, and you get something meaningfully more useful than a generic assistant bolted onto a CRM. Whether that advantage survives as foundation models keep improving is a fair question, and one every vendor claiming this advantage, including us, will have to keep re-earning rather than assuming.
This is also why the instinct to rebuild your entire stack around agents is worth resisting. We learned this ourselves, the expensive way: we built our own SDR call quality scorer and a Clay alternative on n8n; both worked as demos, neither scaled, and making them reliable across teams and edge cases turned out to be a full product company’s job, not a side project. Even Anthropic, the company building the models everyone’s racing to use, runs its own go-to-market on Slack and Salesforce rather than replacing them. HubSpot’s Dharmesh Shah has made the same point publicly: agents will use the systems that already exist, and they’ll bias toward whichever platform gives them a great agentic experience, not just a great user experience. Buy the systems of record. Build the intelligence layer on top of them. That’s where the actual differentiation lives, for us and for the brands we work with.
Outcome-based pricing is coming for more than marketing software.
Software has historically been priced against inputs: seats, messages, and API calls, because inputs were what could be measured, and a seat-based tool’s job was productivity, not a business outcome. That logic is breaking down. Customers are increasingly indifferent to inputs and focused on outcomes: did conversions rise, did retention improve, did revenue actually move?
We’ve tied a real share of our own platform fee directly to a client’s business KPIs, not as a marketing line, but as the actual commercial structure. That’s a materially different relationship than a software licence, and it shifts real risk onto the vendor. Outcome-based pricing only works for a company confident its product moves the metric it’s being paid against, which is precisely why most vendors talk about “partnership” while still invoicing you for seats.
The marketer’s job title is changing, whether or not the org chart says so.
The most consequential shift agentic marketing brings isn’t to the software. It’s to the job. A marketer freed from building reports and manually shuffling data between systems doesn’t become less necessary; they become newly available for the work that was always supposed to be the job: positioning, pricing, and understanding customer psychology well enough to know which micro-segment actually deserves a different message. We saw this in our own numbers before we saw it anywhere else: the quarter we combined people cost, tool cost, and AI token cost into a single line instead of three, our combined spend dropped 20%, not because anyone was cut, but because the work itself needed fewer hands to move the same distance. Call it a chief profitability officer instead of a growth hacker, since marketing remains one of the highest variable costs in most consumer businesses, and every rupee not wasted on reacquiring a customer you already had drops straight to the bottom line.
The agents are not the moat. What a team does with the time agentic marketing gives back is.
Share your exclusive thoughts to:
jeevika@thefoundermedia.in
