Beyond AI Content Generation: Neil Tewari on Context, Infrastructure and the Agentic Marketing Stack
ICYMI: Neil Tewari catches up with 10 Minute Martech on why AI marketing starts with context, not content.
“The hardest part in marketing specifically is giving agents access to context.”
In this episode of 10 Minute Martech, Neil Tewari, Co-founder and CEO at Conversion, joins host Sara Faatz to explore what separates marketing teams successfully adopting AI from those struggling to move beyond experimentation.
From the infrastructure required to give agents meaningful context to the growing technical expectations placed on marketing leaders, Neil argues that the biggest opportunity in agentic marketing isn’t simply producing more content. It’s building systems capable of understanding enough about customers, companies and campaigns to make better decisions.
And as those systems mature, he expects another significant shift: marketing organizations may eventually scrutinize token consumption as closely as they scrutinize advertising spend today.
Neil’s Big Idea: Agentic Marketing Is an Infrastructure Problem
Ask an AI model to write an email and it can produce something in seconds.
Building a system that knows who that email is for, what conversations have already happened, what the company cares about, which signals matter and how all of that information should influence the message is a very different challenge.
That’s the distinction Neil sees between teams experimenting with AI and those beginning to build genuinely agentic marketing operations.
The strongest teams aren’t expecting a model to “one-shot” the perfect campaign. They’re investing in the plumbing underneath it: connecting customer information, company data, conversation transcripts and other signals so agents have the context required to make useful decisions.
Humans haven’t disappeared from that process either. Neil says teams doing this well still use human QA, but the underlying systems can help them move five or even ten times faster.
For marketing leaders, that changes the job. Understanding where data lives, how systems connect and where context gets lost is quickly becoming part of what it means to lead a modern marketing organization.
Neil Tewari’s 10 Memorable Moments on 10 Minute Martech
1. AI Marketing Is About More Than Generating Content
The most advanced teams aren’t simply asking AI to produce emails, ads or other assets. They’re building systems where agents can contribute across the funnel, from selecting creative and determining what gets shown to a customer to supporting landing page creation and campaign execution.
2. Context Is the Hard Part
Giving an agent a prompt is easy. Giving it access to the customer, company and campaign context required to make the right decision is considerably harder.
That context might include CRM information, company attributes, previous conversations, call transcripts and other signals distributed across the martech stack.
3. The Best Teams Get the Plumbing Right First
Teams seeing stronger results have invested in the infrastructure connecting those signals.
Neil contrasts that with teams frustrated because a general-purpose AI model didn’t immediately produce exactly what they wanted. The difference isn’t necessarily the model. It’s what the model knows when it’s asked to act.
4. Agents Can Touch the Entire Funnel
Agentic marketing doesn’t mean AI needs to create every piece of creative.
An agent might instead choose between approved templates, determine which ad should be surfaced, support landing page creation or make decisions based on customer context.
The important shift is that agents can increasingly participate throughout the end-to-end marketing process.
5. Marketing Leaders Need to Understand Their Stack at a Deeper Level
Neil believes CMOs and VPs of marketing increasingly need to understand how their technology operates beneath the surface.
Questions about where information lives, how Salesforce objects connect, why certain context isn’t available in Marketo or where data gets trapped are no longer purely technical concerns. They can directly affect how effectively an organization uses AI.
6. Legacy Martech Needs to Be Questioned
Understanding existing infrastructure is only part of the challenge.
Neil argues marketing leaders should also be willing to ask whether tools that have been considered standard for years are still the right tools for an AI-driven marketing organization.
Engineering has already experienced a significant AI-driven transformation, with sales and revenue operations following. Marketing risks becoming the organizational bottleneck if legacy technology prevents it from doing the same.
7. Composability Becomes Even More Important in an Agentic World
For agents to work effectively, context needs to move between systems.
Neil argues that data should be able to pass freely throughout the martech stack rather than remaining locked inside individual platforms. Systems that restrict access to their data can make it significantly harder to provide agents with the context they need.
In that environment, composability isn’t simply an architectural preference. It becomes an important part of whether a marketing organization can take advantage of agentic technology at all.
8. Token Spend Could Become a Marketing KPI
Neil compares today’s uncertainty around token consumption with the early days of cloud computing.
Cloud costs initially introduced a different way of thinking about technology budgets. Eventually, organizations became comfortable with costs scaling because greater consumption could be connected to greater capacity and business value.
He expects marketing to undergo a similar transition with AI.
9. Token Spend Could Eventually Outpace Advertising Spend
Neil’s boldest prediction is that token spend will outpace advertising and marketing spend over the next five years.
Today, many marketing organizations remain cautious about variable AI costs. Neil believes that changes once teams can clearly connect increased token consumption with better marketing performance.
Rather than asking why AI usage is costing more, organizations may eventually ask why they aren’t using more of it.
10. Marketing Jobs Will Change More Than Marketing Teams Shrink
Neil doesn’t expect agentic marketing to simply result in dramatically smaller teams.
Instead, he expects the work itself to change.
Marketers will need to become more technical, experiment with more tools and understand how context is collected and used. Teams may spend considerably more time thinking about agent training, token consumption and the signals feeding their systems.
The expectation won’t simply be to use AI. It will be to understand how technology can help the organization move faster, make better decisions and ultimately market more effectively.
Neil’s Martech Hot Take
“I actually think that token spend will outpace advertising and marketing spend over the next five years.”
Neil sees token consumption following a similar adoption curve to cloud computing.
Marketing organizations may initially be uncomfortable with variable AI costs, particularly while the connection between token consumption and marketing performance remains difficult to measure.
But as the infrastructure improves and organizations can see that increased AI consumption is helping them make better decisions, personalize experiences or execute more effectively, Neil expects that mindset to change.
The result could be a fundamentally different marketing budget—one where the resources powering agents become as important as the media budgets used to reach customers.
What Does the AI-Forward Marketing Team Look Like?
Neil expects context to become a much larger organizational responsibility.
Just as teams today scrutinize advertising spend, brand performance and campaign results, future marketing organizations may have people focused specifically on how agents consume resources, what context they’re trained on and which signals influence their decisions.
AEO and GEO are already beginning to introduce another version of that question: marketers aren’t only thinking about their own AI systems, but about how other companies’ AI systems understand and represent their brands.
That means the technical bar across marketing is likely to rise.
The marketers who thrive won’t necessarily need to become engineers. But they will need to understand their systems well enough to ask better questions, experiment with new tools and recognize where technology can improve the speed or quality of marketing decisions.
And for Neil, that transformation starts well before an agent writes its first email.
It starts with the infrastructure and context underneath it.
Listen to the 10 Minute Martech Episode
Next Up in the 10 Minute Martech ICYMI Series
Harley Helmer, SEO Team Lead at Americaneagle.com, joins Sara Faatz to discuss tracking AI search visibility, building brand authority beyond your website and why marketers shouldn’t bet against Google.
Want to Keep Reading?
Here’s the full transcript so you can explore every insight from Neil’s conversation with Sara.
Sara: I’m Sara Faatz, and I lead community and awareness at Progress. This is 10 Minute Martech.
Neil: Right now, there seems to be this slight aversion towards spending a lot on tokens and stay away from variable-based pricing in marketing. I do think that there’s gonna be a change in the adoption curve there.
We saw it in engineering. We saw it in sales. I think that there’s gonna be a big change in the mindset of CMOs and VPs of marketing. I actually think that token spend will outpace advertising and marketing spend over the next five years.
Sara: That’s Neil Tewari, Co-founder and CEO at Conversion. Let’s get started.
So Neil, every day you and your team are actually helping modern organizations move from manually managing campaigns to building modern intelligent systems. So you really have this front row seat to the transformation that’s happening in the martech space. When you look across those journeys, what separates those who are succeeding from those who are struggling, and what is surprising them along the way?
Neil: Yeah, it is really cool to have that front row seat, like you’re mentioning, where we can actually see what are the fastest growing teams in marketing doing, what are some teams that are maybe lagging behind doing, how is that affecting these companies? And I think it’s really interesting when you’re a AI marketing automation company.
I think some folks assume that you’re just mass producing LinkedIn slop or something like that on behalf of businesses, et cetera. I think that the teams that are doing things really well understand that the hardest part in marketing specifically is giving agents access to context, is giving agents access to the ability to know actually what’s going on to actually produce the correct things.
It’s not just sitting down and ripping emails and subject lines based on what it knows about things. They have the plumbing and infrastructure set up correctly. We can take an example: let’s just use email. If we’re trying to email you, Sara, right? You know?
It’d be really easy for me to just throw a prompt in Claude and have it email you, for example. Right. It’s much more complex for me to set up a system that knows all the signals about Sara, that knows everything that, hey, someone from our team actually spoke with Sara, that Gong is passing through and actually influencing the decision, the Gong transcript is influencing the decision.
Everything we know about Sara, the company, the company size, the company prerogatives all feed into actually kind of get that first draft, that first version out. We notice that some of the teams that are doing this really well have invested a lot in getting the plumbing and the infrastructure correct, and they still, even with that, have a human QA things, but they’re still outputting 5, 10x faster.
And some of the slower teams are actually those that are frustrated with, “Hey, Claude just didn’t one-shot this correctly for us.” Right. And that’s what we’re seeing some of the best teams doing.
Sara: Yeah. Yeah, that’s great. And are there things that people are, when they come to you and start working with you, they have either preconceived notions that are incorrect or something that surprises them along the way as they’re modernizing their campaign management systems?
Neil: I think they’re surprised with the capability set that’s possible to achieve, both in helping them speed things up and in actually how much better you can market to folks if you get that infrastructure correct. I think the correct folks coming in and saying, “Hey, I can probably do this and I can probably help out with email.”
And we’re seeing some phenomenal teams are actually having every single ad that’s surfaced to every single person has an agentic component. It doesn’t mean that it’s necessarily making the creative. It’s actually choosing among a large template of folks to actually go in and do things, or it’s actually even creating landing pages, and maybe there’s still a human in the loop that are QA-ing things.
But the entire process end-to-end, full funnel, can be touched by agents. And like I said, it’s not always creating every single creative, every single bit of asset, but I think people are really surprised with how far along agents have come. I think they are surprised because when they type in something to be done in Claude, it doesn’t do it all into perfection.
But with that, combine that with the context required, I think people are really surprised with the end results that are possible. We try and use our own product a lot to be able to show those results to folks. Right. And I think once people are involved, I think they get really excited about it.
Sara: Yeah. Yeah, that’s great. We’ve been doing these roadshow events, the Progress MartechNEXT Roadshow, and one of the things Scott Brinker talks about all the time is context. He said if there’s one word for 2026, it’s context, whether it’s context engineering or some of the other things that you said.
If somebody is listening to this who’s saying, “OK, yeah, makes sense. Infrastructure, plumbing, all sounds good. Where do I start?” What would you tell them?
Neil: Yeah, of course, of course, I have a strong bias to setting that foundation right with Conversion. And this is a big part of what we do as a product is how can we enable things to get deep and actually understand the context at a robust scale.
I think that this is a very interesting time where I think CMOs and VPs of marketing—it’s almost a requirement now to be very in the weeds of how your actual tech stack operates end-to-end. I think for a long time it wasn’t actually a big part of the job. I think everybody just chose Salesforce, everybody chose Marketo, everyone chose 6sense, and they all kind of got to work, and there was—members of the team that go in and use those tools to the fullest extent.
And there wasn’t that much else to be kind of thought about when it came to actually executing. Now it’s like when I say plumbing and I say infrastructure, it means actually really robustly understanding, OK, like, why do we not have this context inside our Marketo? And it says, “Oh, it’s on a Salesforce custom object, and that’s related to these objects this way.”
That’s not stuff I believe that CMOs or EVPs of marketing, and very fairly were using their time on before. But now actually can be the difference between marketing well and marketing poorly compared to your competitors. I think there’s two things really is you’ve really got into the weeds of that infrastructure and the plumbing to understand where are the faults, where are the gaps.
And then second is I think there also now needs to be a questioning of some of these legacy tooling is, should we even be using these things that have been the standard for so long?
Sara: I had been talking for a while about data not being clean, but I think you actually have gotten even more precise.
It’s not necessarily even that data’s not clean, it’s that you have to have high level understanding or maybe a deep understanding of what you’re working with to begin with.
Neil: That’s right. That’s right. Exactly. And I think if we can agree that this is can transform the end result for a marketing company you could make the argument this is now becoming one of the most crucial things to deeply understand as a marketing leader.
Sara: If you’re in a legacy organization, an organization that’s been around for a long time and has legacy systems that, whether they would admit it publicly or not, might be Band-Aid together on the back end when it comes to the martech stack, what advice do you give them to think about how to modernize in a way that doesn’t break the business?
Neil: It’s a great question. Let’s look at the other departments of a company, right? Right. Let’s look at engineering—I studied engineering. It was the first thing that was dramatically disrupted, right? It’s like, oh my God—it is 10 to 100 times easier to engineer things that we previously had to do.
And there was that aha moment, and there was a sharp adoption curve from CTOs, and I think there was also a little bit of pushback on folks that were a little bit slower to adopt. I think that we wanted to be, you know, see engineers were eager to be jumping in and seeing how could this revamp the entire stack.
And now we actually measure how many tokens are we consuming as an engineering team, as a metric of success, which is a little silly. It’s basically how much are we spending? Why are we not spending more? And I think that we started to see this curve in certain parts of sales and rev ops as well. It’s where we see Clay really coming in, and whether it’s Clay or any other enrichment tools, how can we get as much data and as much context on each person we’re reaching out to, et cetera.
There hasn’t been the same level of scrutiny and adoption in marketing, and I think it’s because the legacy systems are so ingrained, the data is so complex that, of course, there’s a little bit of a higher bar to hit before you can actually go ahead and rethink some of the structure in this old tooling. But I think that in the next one, two, three years, the marketing leaders that are considered really ahead of their time will be the folks that said, “How can we actually go ahead and safely start thinking about how we can break down this legacy solution?” We don’t want marketing to become the bottleneck of a whole company as everyone else starts to adopt AI really well.
Sara: Yeah. We’ve talked for years about composability and having composable systems, and that’s really the way to go so you can always have a modern approach to your martech stack. But I think it’s even more critical now. We’re talking internally about AI your way. Every customer that we talk to is on a different journey, right?
And so making sure that the products that you’re delivering are composable enough that, that they can use AI in the way that they want, and they can replace a legacy system easily. It’s just an interesting time. And I think that the composability aspect of martech systems right now—I think if people built that way and they’re not in these walled gardens, are probably gonna have an easier time transforming. But I would love your thoughts on that.
Neil: I think it’s a really great take. I think that data should be freely passing among systems as a requirement for whatever it is that we use now. If we have data that cannot get out of a system, let’s just take Marketo’s APIs are quite limited in how much data can come out of it.
And whether that’s because it’s old infrastructure, or whether that’s because Marketo wants to keep data siloed in its own data just to increase its own moat for businesses. Either way, it’s prohibitive. And so if we want to be able to have access to data that lives in Marketo, in any other system that’s determining which emails or ads or landing pages we get, that’s not possible to do.
And so I think there’s this very high technical bar for marketing products because they process such a high level of data to be as composable as possible. Data should be as free as possible among these systems in order to actually be able to achieve what we should be achieving as a marketing team, which is having free context, infinite context to be able to market at a high level.
Sara: If you’re talking to somebody in the martech stack, right, who maybe is more leaning towards the marketing side and not necessarily as technical, what advice would you give them when it comes to thinking about token consumption and protecting that investment and budget?
Neil: You drew such a good analogy to the cloud economy as well. There was kind of a learning curve to figure out how to do that correctly. And I’m sure that even cloud providers took some time to really figure out exactly how their pricing should look as well.
And I think I wanna draw a similar analogy, which is eventually those businesses were able to get to a place where as you scaled, the bill scaled, but you felt confident it was scaling with you. You were using more resources, but it was providing more value as you did so. Hey, now I can, instead of 100,000 users on my platform, I can hold 500,000. And yes, the bill went up 5x to represent it, but also so did the rest of the core functionality of my company.
And right now there seems to be this slight aversion towards spending a lot on tokens and stay away from variable-based pricing in marketing. I do think that there’s gonna be a change in the adoption curve there. We saw it in engineering, we saw it in sales. Like with engineering, it’s very clear and obvious—hey, you can write X lines of code or it can accomplish Y projects. We just haven’t gotten there with marketing yet. Primarily, the big focus of today’s conversation is that infrastructure isn’t there just yet.
But once we can feel like, oh, we’re actually marketing better, once you feel like we’re getting to that place, I think that there’s gonna be a big change in the mindset of CMOs and VPs of marketing. I actually think that token spend will outpace advertising and marketing spend over the next five years.
Sara: Interesting. That’s a good hot take, and I haven’t even asked you for a hot take, but I like that one. Along those lines, curious what you think—if you had a crystal ball, five years from now or even a year from now, because everything’s changing so quickly, what do you think a true AI-forward or agentic marketing team looks like from a team structure perspective?
Neil: Yeah. With how things are shaping, I feel like you’re right. We try and only think one to two years ahead—[Sara: Right. Yeah …] instead of a half decade ahead as well. And I think I kind of started the ball here, where I do think that there’s gonna be as much scrutiny around how are we thinking about ad spend? How are we thinking about brand image? The way that we have teams fully focused on that, I do believe that half of the team is gonna be thinking about how do we think about token spend? How do we think about training our agents on the right context? We’re starting to see a little bit. We’re starting to see a lot more incorporation of complex signals and things like that. That is context, and that is context that helps humans, and that will therefore be context that helps agents as well. I think that’s gonna be the next big component. We’re seeing this with AEO and GEO. We’re seeing a bigger focus on not our token spend, but other companies’ token spend to figure out where do we fit in the mix?
And so I think there’s gonna be as much scrutiny on that component. I think there’s gonna be an increased bar. Everyone in a marketing org is gonna be required to get more technical, be expected to be tinkering around with more tooling, and be thinking about how can something either make the team faster or the team actually market at a higher and higher level, and make faster decisions. I actually am not one of those folks that believes that teams will dramatically shrink or a bunch of folks will be let go. I don’t think we’ve seen that in other orgs, and if anything, I think we’ve seen a lot more folks trying to hire a lot more. But I do believe that people’s jobs will change a lot, and it’ll be focusing more on technicals, focusing more on context.
Sara: That’s awesome. Well, Neil, I feel like I could talk to you for a lot longer, but I need to stay true to our 10 Minute Martech promise, so thank you so much for your time today. I truly appreciate it.
Neil: Thank you, Sara. Thanks so much for having me on.
Keep reading: 10 Minute Martech recaps.
