There is no shortage of AI tools marketed at B2B marketing teams. There is a real shortage of clarity about which of those tools belong in their actual operations and which are just subscriptions on a credit card.

Matt Wood, Chief AI & Technology Officer at AWS, published a framework this week that explains the problem more precisely than most of the debate I hear at enterprise marketing conferences. Two economic curves are moving simultaneously in AI. Almost nobody discusses both at once. Reading them together explains why most marketing teams are betting on the wrong one.

Two Curves, Two Different Bets

The first curve is the one that gets all the attention: the frontier. Frontier AI budgets keep rising year over year. Not because each unit of capability is getting more expensive, but because the definition of “frontier” never holds still. Each model generation absorbs more reasoning, more context, more tool use, more evaluation. So even as price per token falls, what we ask of leading models expands to fill the cheaper envelope, and total spend doesn’t shrink.

The second curve is the one most marketing teams aren’t watching: for any fixed level of capability, prices are falling at rates that change the underlying economics of what’s worth building. Epoch AI and Artificial Analysis track this by holding a capability threshold constant on standardized benchmarks while measuring how price evolves. The result: GPT-3.5 Turbo-level performance on general knowledge fell roughly 9x per year. GPT-4-level performance on PhD-level science questions fell roughly 40x per year. Models that cost $30 per million tokens in early 2023 had equivalents below $1 by mid-2024 and below $0.10 by early 2025.

That’s not a marginal efficiency gain. That’s a category change in what’s economically viable to build — and almost all the marketing strategy conversations I’m part of are still anchored to the first curve.

Why Marketing Teams Default to Frontier Chasing

Wood calls “the field” the broad space behind the frontier: AI capabilities that are no longer novel but are cheap and available enough to embed in ordinary work. Forrester estimates AI orchestrates less than 1% of core business processes even as it shapes process design, tooling, and data integration.

Marketing teams are probably at or below that number, and there’s a structural reason. Marketing organizations face institutional pressure to look current. A demo with this week’s newest tool generates a slide for the quarterly business review, generates social content for the team’s LinkedIn, generates a sense of visible motion. Mapping the lead qualification workflow and building a behavioral scoring layer on top of it doesn’t generate that slide — even though it’s what actually moves pipeline numbers.

The pattern repeats across most B2B marketing teams: a wide portfolio of AI tools tried fast and abandoned almost as fast, and a real operating surface — the workflows where prospects actually move and where campaign decisions actually get made — that remains entirely manual. The same argument I made about building systems instead of buying chatbots applies with particular force in marketing: durable value doesn’t live in the tools you buy, it lives in the systems you build on your actual operations.

What the Field Looks Like at Current Prices

When the price of an AI capability falls 40x in two years, workflows that previously didn’t justify the cost of systematic deployment start generating positive return. That shift is happening right now in marketing, and the opportunity map looks radically different from what most teams last calculated.

There are marketing AI capabilities that have been technically viable for two or three years but were economically hard to justify at 2022 or 2023 inference costs. Behavioral lead scoring at scale to prioritize handoffs to sales. Dynamic audience segmentation from multi-channel signals. Message optimization in email or WhatsApp flows based on historical response patterns. Content quality analysis before distribution. None of these are frontier ideas. They are field capabilities that, at current inference prices, carry a return profile that didn’t exist eighteen months ago.

The shift Wood describes: when inference was expensive, the model was the system. Now that it’s cheap and falling, the model is a component and the system is the work. For marketing, that means the advantage doesn’t come from accessing the newest model. It comes from having built the systems that connect available models to the real data and workflows of your operation — which is what I’ve called the harness being the moat: what you build around the model is what can’t be purchased.

What IQ Source Does With This

The most common question I get from marketing leaders is: what AI tool should we buy? It’s the wrong question — not because tools don’t exist but because it starts at the solution before mapping the problem.

The right question is: which of my marketing workflows already have data clean enough, processes repeatable enough, and error tolerance high enough that deploying AI against them generates positive return at current inference prices? That question requires a map of actual operations, not a vendor evaluation.

That diagnostic is what we build in the first phase of AI Maestro: mapping the real marketing workflows, identifying which ones have the data and process state required for deployment to actually work, and prioritizing by return at current inference cost. We don’t choose the model. We map the specific field of your marketing team — the work that has to come before anyone buys another subscription.

The AI field in marketing is wide and uncultivated in most organizations. The prices are already where they need to be for the work to be worth doing. What’s missing is the map.

Map your marketing team’s AI field