A decision intelligence platform decides what to do for each customer.

For a decade the industry has tried to connect ad tech and martech with data: customer data platforms, clean rooms, marketing clouds. Forrester reports that over 90% of enterprises are planning or executing strategies to align the two. The gap persists, because the two sides optimize for different things.

Ad platforms optimize for the cheapest next conversion inside a short attribution window. They find fast converters, because those register in time. And they get more expensive every year: Triple Whale measured Meta CPMs up 20% in 2025 across every industry, and WordStream saw Google Shopping cost per click rise 33.7%.

Engagement tools optimize for opens, clicks and sends. Neither side knows the value of the relationship, so neither can optimize for it. Sharing data between them does not change what each one optimizes. A decision layer with one goal does.

Gartner named the category

The January 2026 Magic Quadrant populates it with vendors built for single, governed decisions in regulated industries: should this loan be approved, how should this supply chain route. A B2C relationship is a different problem: a continuous stream of small decisions, across every channel, for millions of people.

Brinker mapped the architecture

Brinker's canvas has five rings: a data core, a semantic layer, context services, decisions, and apps and agents. He warns that when decision logic lives inside individual channels it fragments, and suggests decisions work better as an independent service that consumes context. The report spends two paragraphs on that ring, so the implementation is left to vendors.

Attribution 2.0 asked the right question

Brinker and Riemersma's report moves from counting conversions to tracking revenue, from campaigns to customer journeys, from reporting to decision support. It stops at support: a human still decides. A decision platform goes one step further and acts, within the goal and the limits the human sets.

1. One goal for every tool

The brand chooses what to optimize: conversions, basket size or long-term customer value. Every decision on the site, in messages and in ad audiences works toward it. Not opens, not clicks, not a seven-day return.

2. Bridge ads and owned channels

Send predicted customer value to Meta and Google next to each purchase, exclude customers you already have from acquisition, and run the owned channels from the same profile. One goal on both sides (ads audiences).

3. A separate layer, not logic inside each tool

Exactly Brinker's warning: decisions embedded in each app fragment into contradictions. The layer reads context from everything and gives every tool its instruction.

4. See every visitor

A decision made on incomplete data is incomplete. Adjust puts the average iOS tracking opt-in at 35%, and browser pixels lose events to ad blockers and cookie limits. Server-side tracking and a profile that starts at the first interaction, not the first login, are the foundation (guide).

5. Autonomous, with human governance

The human sets the goal and the limits. The machine decides and learns continuously. Most brands start with the system proposing and a person approving, and move to full autonomy as the holdout results come in. One store has not sent a campaign by hand in three months and routes 47.5% of its revenue through the agent (case study).

6. Scale and transfer

Buying cycles, category expansion, price sensitivity and channel responsiveness look alike across markets and categories, so what the platform learns in one market speeds up the next. One marketplace client runs 22 million shopper profiles and 131 million events a month on one agent, across three countries (case study).

A customer data platform unifies data into profiles. It answers "who is this customer?" A decision intelligence platform takes those profiles and decides what to do next, for whom, where, toward which outcome. One is a data layer, the other a decision layer.

Marketing automation fires rules: if the customer did X, do Y. Rules look backward and treat everyone the same. A decision platform predicts: this customer is likely to buy in ten days, answers push in the evening, and should see a new category rather than a discount. The difference is autonomous optimization against manual orchestration.

The how it decides page shows what this looks like in the agent, placement by placement: you set the goal, it decides per person, and a holdout group without the agent shows what it added.