What Just Died in Martech, and What Dies Next
The 2010–2019 SaaS cohort lost an architecture fight, not a feature fight. ASI can name the next casualty.
Executive Summary
The 2026 contraction of the martech landscape was an architectural disqualification, not a competitive defeat. The casualties from the 2010–2019 SaaS cohort failed to meet the architectural conditions that allow firms to compound under AI competition. Learning-Loop Economics names those conditions. The AI Susceptibility Index identifies where structural exposure concentrates next, including in categories the dataset currently records as growing.
Architecture decisions made in 2014 determined which firms could compound in 2026, before AI made the question visible.
How This Works
The mechanism operates at two levels: architectural conditions that govern compounding under AI competition, and element-level dimensions that locate where structural exposure concentrates.
Learning-Loop Economics specifies three preconditions for compounding: exclusive data capture, tightly integrated architecture, and rapid feedback tempo. When any precondition is absent, interaction data accumulates without producing self-reinforcing advantage.
AI capability delivered through shared foundation-model APIs is non-rivalrous. Loop position is firm-specific and depends on prior accumulation of exclusive interactions, integrated systems, and active feedback cycles.
Incumbent suites holding the customer’s data, workflow, and contract absorb adjacent AI functions at lower marginal cost than standalone vendors can introduce them as products.
The AI Susceptibility Index scores each strategic element across five dimensions: digitizability of value, data intensity and availability, automation potential, network-effects sensitivity, and capital reallocation friction. Element scores aggregate into a strategic-formula exposure profile.
Industry category taxonomies label function at a point in time and do not update when underlying functions migrate between firms or between layers of the stack.
When AI agents consume structured event data directly, the rendering function performed for human consumption becomes non-essential to the workflow.
Definitions
Learning-Loop Economics: Learning-Loop Economics is a theory of compounding competitive advantage in the AI era, requiring three preconditions: exclusive data capture, tightly integrated architecture, and rapid feedback tempo.
AI Susceptibility Index: The AI Susceptibility Index scores each element of the Periodic Table of Business Strategy across five fixed dimensions to identify where AI-driven structural pressure concentrates: digitizability of value, data intensity and availability, automation potential, network-effects sensitivity, and capital reallocation friction.
Prefer to listen? Here’s the audio edition of this post:
When I saw this year’s chiefmartec landscape report a couple of weeks ago, the AI fallout was already obvious in the surface numbers. The report shows 0.79% growth: 15,384 products at the start of the year, 15,505 by year-end. That’s the headline, but it’s the wrong number.
The fallout is in full swing, and I don’t think many people are surprised by it. What isn’t obvious is why specific platforms got hit hardest — or which ones get hit next. That’s where my own tools do their work. Learning-Loop Economics tells me which architectures could compound under AI competition and which could not. The AI Susceptibility Index tells me where the pressure is structurally concentrated. Run the 2026 report through both, and the pattern that produced this year’s contraction also names the next wave of exits.
The Number That Matters Is Underneath
1,488 tools were added in 2026. 1,367 were removed. Inflow dropped roughly 40% year over year. Outflow climbed 13%. The market lost its replacement engine before it lost its incumbents. A flat top line was the surface effect of a category that has already been priced.
Of the 1,367 removed tools, 51.7% were SaaS startups founded between 2010 and 2019. Roughly 80% had fewer than 50 employees. That is not a story about execution failure or capital cycles. It’s a story about architecture.
The 2010–2019 Cohort Failed an Architecture Test
I want to be precise about which architecture. The 2010–2019 cohort built the modern point-solution SaaS playbook: narrow function, fast time to value, integration-first distribution, and a bet that switching cost would emerge from how deeply the tool got embedded in someone else’s workflow. That playbook was correct for a decade. It produced category leaders. It produced exits.
It’s also the playbook that disqualifies you from compounding under AI competition.
Learning-Loop Economics describes the conditions under which a business converts use into structural advantage. The theory includes three preconditions: the company must capture interactions competitors cannot easily replicate; data, models, and product surfaces must be tightly integrated; and the cycle from interaction-to-improvement-to-user must be fast enough to compound before competitors can respond.
The dead cohort had none of the three. The interactions they captured were available, usually identical, to anyone running a similar pipe through the same foundation model API. The architecture was thin: a UI on top of a model on top of a connector library, with every layer rented. The feedback tempo was bounded by how quickly an external model provider released improvements, which is to say, not their tempo at all. They were data projects, not learning loops. The data they generated fed someone else’s loop.
That someone else was Adobe, HubSpot, and Salesforce. Those companies didn’t win the AI content fight because their AI was better. Their AI is the same AI. They won because they already owned the customer’s data, the customer’s workflow, and the customer’s contract. Absorbing an AI feature into that position cost them less than building it as a standalone product cost a challenger.
The pattern repeats across the contracting categories. Content Marketing lost 37 net tools in 2026. Sales Automation Enablement & Intelligence lost 23. Email Marketing lost 22. Live Chat & Chatbots lost 23. The vendors removed were not bad products. They worked. They did the thing on the box. They just could not compound, and the customer’s contract was held by a suite that could.
Capability Is Rented. Loop Position Cannot Be.
Those two sentences are the most expensive lesson of 2026. Buyers, founders, and investors spent two years evaluating AI features as capability decisions — does this product do AI well, does it have the right model, does it generate competitive output — when the structural question was always loop position. Loop position is not a feature you can buy or build at a vendor’s pace. It is a function of what the company already has: the exclusive interactions, the integrated stack, the tempo. If those weren’t there before AI arrived, they were not going to be created in time to matter.
The casualties looked like normal point-solution wins right up until they didn’t. They were running the playbook that was correct for a decade. Distribution velocity, feature parity, integration depth. Those were the levers, and they moved them well. But the optimum changed underneath them. The playbook was designed to produce switching cost from workflow embedding. That turned out to be the thing AI most easily compresses, because AI absorbs the workflow rather than working around it.
The Same Logic, Applied Forward: Mobile & Web Analytics
I spend a meaningful share of my own working week using Power BI dashboards and several other analytics tools. The category in the chiefmartec dataset that contains this work — Mobile & Web Analytics — grew 11.3% net in 2026. Read against the contracting categories, that looks like good news. Read forward, it isn’t.
The function those tools perform is rendering event data for human consumption. That is the actual product, even though the vendors describe it as analytics. Someone instruments a system, the system emits events, the events get aggregated, and a dashboard renders the aggregated state in a form a human can read. The product is the rendering layer. The events sit underneath it, mostly invisible to the people the tool exists to serve.
The AI Susceptibility Index profile of that function is ugly. Digitizability of value is at the top of the scale; the entire output is already digital, with nothing physical in the chain. Data intensity is high, and the data is broadly accessible to any model authorized into the customer’s stack. Automation potential is high; every step from query to summary to recommendation can be performed by a model with access to the same data. Capital reallocation friction is low. The customer is not locked into a pipeline that takes years to migrate. The data lives upstream, in databases and event stores the customer already controls.
What makes the score binding is what changes when agents start consuming events directly. Today’s analytics stack assumes a human bottleneck: a person who needs the data summarized, visualized, and explained before they can act on it. An agent does not need any of that. The agent reads the events, decides, and acts. The rendering layer is not a feature it requires. It is a residue from the human era.
The precursor is already in the dataset. iPaaS and Data Integration, the layer where events live before they are rendered, grew 8.0% net in 2026. The category closest to where agents will actually consume data is growing while the rendering layer above it grows on borrowed time. When that shift gathers pace, the function the analytics category has been organized around stops generating value. The vendors will adapt. Some will absorb agent-readable surfaces, some will reposition as data infrastructure, some will be acquired into incumbent suites. Most will be removed from a future chart.
What the Taxonomy Hides
The chiefmartec dataset is doing useful work, but its taxonomy is a lagging instrument. The category labels persist while the functions migrate. Content Marketing lost 37 net tools in 2026 because the function moved into Adobe and HubSpot. Mobile & Web Analytics gained 11.3% in 2026 because the function still sits with the human reader. Both numbers are warnings about the next cycle, not conclusions about this one.
The decision that matters for anyone operating inside one of the currently growing categories is not whether to add AI features. It is whether the architecture you are sitting on top of qualifies for a learning loop or describes a function AI agents will route around. The answer is structural. It was set when the company was founded, or it wasn’t.
If it wasn’t, the question is what comes next: a pivot toward exclusive data and integrated tempo, or a slot in the 2027 removal report.
About the author: Eric D. Noren is the creator of the Periodic Table of Business Strategy and author of The Strategic Formula: A New System for Business Strategy in the AI Age (August 2026 — order on Amazon). Learn more at ericdnoren.com.
AI Disclosure: This essay is fully conceived, argued, and structured by me. I use AI tools as a research and drafting partner, but the strategic ideas and final decisions are my own