Decision Intelligence for Marketing: How to Test Decisions Before Acting

In brief: Decision intelligence is the practice of designing and testing a decision before acting on it. In marketing, that means comparing messaging, creative, positioning, pricing, or audience options before committing budget. At Galaxies, these tests run in Nexus using synthetic personas built from a client’s own real research. They do not replace market research; they bridge the gap between the data a company already has and the approval of its next campaign.

At a recent talk for marketing leaders called “The Consumer Your Data Lost,” I asked the room how many weeks had passed between the brief for their last major campaign and the day it went live. Then I asked them to multiply that number by seven.

The answer is the number of days consumers had to change their minds about the category while the company approved copy, creative, channels, and budget. An eight-week approval process gives them 56 days. That is long enough for a trend to emerge, peak, and fade.

That delay has a name: the Signal-to-Decision Gap, the time between a consumer signal appearing and a company making a material decision based on it. As long as that gap persists, brands make decisions a step behind their audiences.

I bring this up because the conversation about AI in marketing has become stuck on production: more copy, more ad variations, more images. That solves a problem that is no longer the main bottleneck. The harder question is which of the many available options deserves the company’s money. Decision intelligence addresses that question. Gartner takes a broad view, close to BI. Galaxies has focused since 2022 on a narrower challenge: making consumer decisions before spending, for consumer-facing businesses.

What is decision intelligence?

Decision intelligence is the discipline of structuring a decision before making it: define the problem, state a hypothesis, develop alternatives, test how each performs, and record the outcome so the next decision improves. The focus shifts from the data itself to the choice the data must support.

Gartner describes it as a practical discipline that advances decision-making by explicitly understanding and engineering how decisions are made and how outcomes are evaluated, managed, and improved through feedback. In its 2025 AI Hype Cycle, Gartner estimated adoption at 5% to 20% of organizations, with two to five years until maturity [1].

For marketing leaders, the important phrase is “engineering how decisions are made.” Most companies I know design campaigns with great care. Very few design the choice between campaigns. Alternatives are often compared only in an approval meeting, where the most confidently defended option wins, not necessarily the best-tested one.

How does decision intelligence differ from BI, analytics, and generative AI?

These approaches coexist in the same operation. The difference is the question each answers. Business intelligence asks what happened after execution. Historical predictive analytics asks what is likely if past patterns repeat. Generative AI asks what options can be produced. Market research asks what an audience thinks, with sample-based evidence. Decision intelligence asks which alternative to choose, and why, before execution. Its test is only as good as its underlying data and hypotheses.

Generative AI is the most revealing comparison. It multiplied the number of options without improving our ability to choose among them. A team that once debated three campaign concepts now debates thirty, with the same deadline and the same number of people in the room. Without a decision method, more options create more opinions, not better choices.

Why has marketing run out of time to make good decisions?

The first two signals show that consumers now make decisions with help from AI, often in conversations the brand cannot see. The others show how Brazilian marketing leaders view their ability to measure and defend their choices. Clutch reported that 65% of consumers use AI to research products before buying [2]. PYMNTS reported that 53% of Brazilians used AI in their most recent online purchase [3].

In a 2026 Makers and Uncover survey of 113 marketing leaders, 75% said data was a determining factor in their decisions, yet only 9% fully trusted what they measured. Another 41% said their boards judged marketing primarily by revenue growth or market share [4]. The research ran from May to June 2026 and included CMOs, VPs, directors, and managers; 70% worked at companies with revenue above R$500 million. Business results are demanded, but the evidence for defending each decision before spending is still weak.

This is not a skills shortage. These teams have creativity, brand experience, and market knowledge. What many consumer businesses lack is a way to compare options before committing budget and to arrive at an approval committee with defensible evidence.

How does Galaxies’ Testable Decision Cycle work?

At Galaxies, we organize client work into five steps, which we call the Testable Decision Cycle. The method is useful without any particular tool, although Nexus was built to run the testing step in minutes and deliver a structured, human-reviewed verdict within 48 hours.

Figure 1. Galaxies’ Testable Decision Cycle: problem, hypothesis, alternatives, test, and decision. The team brings real-world results into the next decision.

1. Problem: what decision must be made, and by when?

My first question to a client is not “what do you want to research?” It is “what do you need to decide, who signs off, and on what date?” A decision without an owner or deadline becomes a study; a study without a decision becomes a report in a drawer. We also estimate the cost of getting it wrong, in budget, time, or reputation.

2. Hypothesis: what does the team believe would have to be true?

Every campaign proposal contains an implicit bet: “Younger audiences will respond better to convenience,” or “Discounts are the strongest Black Friday trigger.” Writing that belief as a single falsifiable sentence is the step that most changes the quality of the internal conversation.

3. Alternatives: at least two options, plus what already exists

The most common mistake I see is bringing just one option to a test. Testing a single asset validates an opinion; it does not support a choice. We ask for at least two genuine alternatives and, where available, the current version as a baseline. Without a comparison, any result can look good.

4. Test: how does each option perform with each audience?

This is where synthetic personas come in. We show each option to profiles built from the client’s real data and examine three things: which option each audience prefers, why, and what would make them reject it. Barriers are often more useful than preferences because they point to what needs to change before an asset goes live.

5. Decision: choose, record the rationale, and define the next step

The test ends with a documented choice and the criteria behind it. We also decide what still needs formal field validation. After a campaign runs, the team compares the real outcome with the test’s indication and carries that difference into its next decision. Recording that learning turns a series of tests into cumulative knowledge.

Which marketing decisions can be tested before execution?

Not every decision deserves the same effort. Pre-testing is most valuable when the budget is high, the deadline is tight, or the choice is difficult to reverse. Six decisions come up repeatedly in our work: messaging, creative, positioning, launches, pricing, and audience selection.

For messaging, compare two or three ways of expressing the main promise by audience and identify the objections each raises. For creative, rank competing images, videos, or copy before buying media. For positioning, compare alternative brand territories with the existing one to understand fit and the risk of alienating current customers.

For launches, test product concepts, names, and benefits to identify adoption barriers by profile. For pricing, compare how a price, package, or commercial terms are presented and what value each conveys; this does not replace a price-elasticity study. For audience selection, expose different segments to the same strategy and identify where it resonates or needs adjustment.

Each of these decisions is part of this month’s broader series on predictive marketing, predictive analytics, consumer behavior, value propositions, brand positioning, market segmentation, and scenario planning for the 2027 budget. Links to those articles can be added as they are published.

What do the cases show about deciding before spending?

At Mahta Bio, creative validation before the campaign reduced customer acquisition cost by 35%. A decision that used to take six to twelve weeks can now receive a structured verdict in 48 hours.

That result is often treated as a media-efficiency win. I see something else: once testing fits the timeline, it stops being an annual event and becomes part of routine approvals. That change in frequency makes decision intelligence possible in marketing. The simulation did not work alone; it helped select which options went into paid media. Explore our client cases and estimate the impact using our ROI calculator at https://www.galaxies.com.br/pt/roi.

How much can you trust a decision tested with synthetic personas?

I hear this question more than any other, and it is fair. At research-industry events, I describe reactions to the concept in four stages: rejection (“it is just a guess”), understanding, application, and, after a few tests, enthusiasm. Moving from the first stage to the second requires transparency about how reliability is measured.

At Galaxies, we assess reliability by comparing persona responses with those of a control group of real people in the same segment, using questions that were not part of profile construction. Conversations are also evaluated continuously with the open DeepEval framework across five dimensions: relevance, bias, hallucination, fidelity to the profile, and toxicity.

Trust also means explaining the limits. Simulation is not formal market proof. When a board or regulator requires statistical validation, it serves as an initial filter for deciding what to take into the field. It does not predict sales volume; it compares options and explains preferences and barriers.

A persona is only as good as the data behind it. An audience absent from the client’s research should not be simulated from that research. Personas also answer within the domain they know; questions outside it should be redirected. Small percentage differences between runs are expected, so interpretation should focus on direction and reasons rather than tenths of a point.

Nexus or Simile: which simulation approach fits your question?

Anyone researching AI-based behavior simulation will soon encounter Simile, a US company founded by Stanford researchers connected to a 2025 study that created generative agents from two-hour interviews with 1,052 Americans [6]. Both companies work with simulation, but they address different questions. Simile starts with broad populations built from its own large-scale studies. Galaxies starts with each brand’s audience: Nexus personas are built from a client’s own real research, including Brazilian consumers.

If you need to understand how a country’s population might react to a policy change, broad-population simulation is better suited to the question. If you need to know how your own customers might react to a new message, or which of three ads should run for Black Friday in Brazil, a brand-specific audience is more relevant. The same distinction applies to internal employee questions versus testing a value proposition across your customer segments.

Where I see our differentiation is straightforward. The persona belongs to the brand, not an average population: it reflects the client’s research and answers “how would my customers respond?” Brazil is not a translation layered on afterward. Local payment habits, marketplace shopping, regional language, and category-specific skepticism are present when the underlying research was collected in Brazil.

There is also a human-led method around the tool. Nexus Agent is the natural-language entry point, with three study types: Persona Conversation, Questionnaire, and Stimulus Analysis. For enterprise accounts, a forward-deployed engineer works alongside the team, and each decision undergoes structured human review before reaching the committee. Client data is isolated under clear rules, with ISO 27001 certification and compliance with Brazil’s LGPD and the GDPR.

If the goal is to simulate whole populations, public policy, or employees across a multinational company, Simile’s proposition is broader than ours. For a consumer brand selling in Brazil and making decisions about its own audience, we believe our approach is the better fit.

Where does market research fit into this process?

It remains at the beginning and at the end. At the beginning, because Galaxies’ synthetic personas are built from real research conducted by the client or its partners, not generic internet data. At the end, because high-stakes decisions still warrant field validation.

In my talk, I used an image to explain the relationship. Traditional research provides a photograph: clear, costly, and true on the day it was taken. Today’s consumer behaves more like a film, changing scenes between the photo and the campaign launch. Companies can and should keep taking photographs. Without them, the film has nothing to work from. But the next campaign needs a frame from today, and simulation helps produce it.

In practice, research builds and refreshes the foundation. Simulation filters and compares options day to day. Research then returns to validate the few bets that reach the end of the decision funnel.

How can a marketing team start using decision intelligence?

You do not need to reorganize the department. Start with a decision already on the table and run it through the cycle. Choose one with a sign-off date in the next four weeks, preferably one involving media spend. Write a main hypothesis in a sentence the test could disprove. Bring at least two real alternatives, plus the current version when one exists.

Before testing, decide what will determine the choice: preference, barriers, clarity, or another measure. Test with audiences that reflect your actual customer base, not a generic consumer profile. Document the decision and compare it with the campaign’s real outcome 30 to 60 days later.

One recommendation comes from a mistake we made ourselves: do not test an asset that has already been approved just to confirm the choice. Once a decision is made, teams tend to interpret any result as validation, and the exercise loses its purpose.

Choosing has become marketing’s most expensive step

AI made production cheaper. What became expensive is choosing badly: a poor decision now reaches the market through more assets, in more channels, with less time to correct course. Decision intelligence is a practical response: treat the choice as a stage of work, with a hypothesis, alternatives, and a test, not as the final moment of a meeting.

I do not believe simulation will decide on behalf of the CMO. I believe it will change what reaches their desk. Instead of one confidently defended proposal, there will be two or three tested alternatives, with reasons and barriers by audience. The decision remains human. It is simply no longer made in the dark.

Bring a real decision to Nexus

Choose a messaging, creative, positioning, or audience decision you need to make in the coming weeks, and compare the options before spending. Explore how Nexus works.

Frequently asked questions about decision intelligence

What is decision intelligence in marketing?

It is the practice of structuring marketing decisions before execution: define the problem, form hypotheses, compare messaging, creative, pricing, or audience alternatives, and test how each performs. The goal is to choose based on evidence rather than conviction.

How does decision intelligence differ from business intelligence?

BI uses historical data to explain what has already happened. Decision intelligence acts before execution by comparing alternatives that have not yet reached the market. They complement each other: BI reveals the outcome, which informs the next decision.

Do synthetic personas replace market research?

No. At Galaxies, synthetic personas are built from a client’s or its partners’ real research. They help test and filter options before spending. Decisions requiring formal proof still go to the field, with fewer options and clearer hypotheses.

How much does it cost to test a decision with synthetic personas?

It depends on the scope and the number of decisions tested during the year. The useful comparison is between the cost of testing beforehand and the cost of a campaign that gets it wrong. Galaxies’ ROI calculator lets you estimate that with your own numbers.

How long does it take to test a campaign alternative?

A Nexus study runs in minutes. A structured verdict, including human review, is ready within 48 hours.

How do Galaxies and Simile differ?

Simile is a US company that simulates broad populations using its own research. Brazil-based Galaxies builds personas from each brand’s real research with Brazilian consumers. The former is better suited to questions about societies; the latter to a brand’s decisions about its own audience.

About the author

Daniel Victorino is CEO and co-founder of Galaxies, a consumer decision layer for marketing leaders at consumer businesses. Galaxies develops Nexus: Galaxies Lab, a platform for investigating and supporting decisions before spending, using synthetic personas built from real research with Brazilian consumers.

Sources

[1] Gartner, Hype Cycle for Artificial Intelligence 2025, definition and adoption estimate for decision intelligence, as reported by Aera Technology: https://www.aeratechnology.com/blogs/transformational-moment-decision-intelligence-2025-gartner-ai-hype-cycle/

[2] Clutch, “65% of Consumers Use AI to Research Products Before Making a Purchase,” January 22, 2026: https://www.businesswire.com/news/home/20260122526477/en/Clutch-Report-65-of-Consumers-Use-AI-to-Research-Products-Before-Making-a-Purchase

[3] PYMNTS Intelligence, Global Digital Shopping Index: Brazil Playbook, 2,273 consumers: https://www.pymnts.com/study/global-digital-shopping-index-new-brazilian-shopper/

[4] Makers and Uncover, O Futuro da Mensuração em Marketing, survey of 113 marketing leaders, May–June 2026: https://meet.uncover.co/o-futuro-da-mensuração-em-marketing-lkd

[5] Galaxies, client cases: https://www.galaxies.com.br/pt/clientes

[6] Stanford HAI, “Simulating Human Behavior with AI Agents,” May 2025: https://hai.stanford.edu/policy/simulating-human-behavior-with-ai-agents