Pressemitteilung

AI removes friction. But not all friction is bad. 

A point of view on moving AI-enabled Insights from experimentation to repeatable decision-making 

I’ve spent quite some time over the past months thinking about a question that keeps surfacing in conversations around AI and Insights: 

When is AI-generated or synthetic evidence strong enough to act on? 

I wanted an answer that would work in two places at once. Externally, it should help organizations move from experimenting with AI to using it with confidence in real decision-making. Internally, it should give us a consistent way to position AI, synthetic and human evidence without turning every discussion into a debate about methodologies. 

The thought I keep coming back to is simple: 

AI removes friction. But not all friction is bad. 

Speed is one of AI’s greatest advantages. We can generate, test, analyze and learn faster than ever before. The opportunity now is to combine that speed with a decision standard that preserves judgement rather than bypassing it. 

From experimentation to adoption 

Most Insights teams are no longer starting from zero. They have tested generative AI, explored synthetic data, compared outputs with traditional research and built an initial understanding of where these approaches perform well. 

That experimentation has created valuable learning. The next stage is about turning it into repeatable use. 

This matters because the opportunity is larger than replacing an existing study with a faster or cheaper version of the same study. 

AI and synthetic data can bring evidence into decisions that historically received little or no consumer input at all. Early creative routes, pricing hypotheses, product ideas, positioning territories or communication options were often decided primarily through experience, intuition and existing knowledge because traditional research was simply not proportionate to every decision. 

AI changes that equation. It expands the space in which Insights can contribute and makes it possible for more decisions to become evidence-led. 

That is where the real opportunity begins. 

Start with the decision, not the methodology 

The most useful starting point is also a familiar one: the cost of being wrong. 

A relatively reversible decision with limited commercial impact carries a different evidential burden from a major launch, a significant investment or a fundamental change in brand direction. As the business risk rises, the strength of the evidence supporting the decision should rise with it. 

Evidence strength, however, comes from more than the label attached to the source. 

Synthetic evidence may be entirely appropriate for an important decision when the approach has demonstrated reliability for that use case. Human research can add depth where context, emotion or behavior matter particularly strongly. Sales data, behavioral signals, trackers and previous studies may provide additional context. 

The source can vary. The decision standard should not. 

This shifts the discussion toward a more useful assessment: whether the available evidence is strong enough for the decision at hand. 

Three dimensions of decision-ready evidence 

In practice, I find three dimensions particularly useful. 

1. The cost of being wrong 

The first is the consequence of the decision. 

Financial value is one part of that, but so are reversibility, strategic importance and the potential impact on customers or the brand. Together, these factors determine how strong the evidence should be before acting. 

A decision that is easy to reverse allows more room to learn through action. A decision with lasting consequences deserves a higher evidence threshold. 

2. How proven the evidence is for the use case 

Trust should accumulate. 

When an AI or synthetic approach has demonstrated strong and repeatable performance for a particular type of question, that learning should carry forward. Organizations can build on what has already been established rather than reopening the same validation question with every new project. 

This is one of the most important shifts from experimentation to maturity. Validation becomes an organizational asset. 

Over time, teams build a clearer understanding of which approaches are decision-ready for which types of business questions. As models develop and the evidence base grows, that playbook develops with them. 

3. How the result fits into the wider evidence context 

No decision exists in a vacuum. 

Organizations already have historical research, behavioral data, sales results, market knowledge, trackers and accumulated experience. New evidence becomes more useful when it is interpreted in that wider context. 

Consistency can strengthen trust. A contradiction can be equally valuable if it reveals something genuinely new, but it deserves closer attention before action. 

The objective is therefore not to make every source agree. It is to understand whether the overall evidence creates a sufficiently coherent basis for the decision. 

This is where triangulation adds value: as context and sense-check, rather than as a routine parallel test. 

AI removes friction — deliberately put some back 

This is where I think the AI discussion becomes particularly interesting. 

Removing friction is usually positive. AI allows us to move from question to answer with extraordinary speed, and in many cases that is exactly what organizations need. 

But speed also changes behavior. When answers become easier to generate, the distance between question, answer and action becomes shorter too. That creates enormous efficiency, while also increasing the value of knowing when a decision deserves a second look. 

Sometimes the most useful thing we can add back into the process is a moment of deliberate reflection. 

Not a barrier, but a checkpoint. 

Stop. Reflect. Proceed. 

At that point, the organization pauses just long enough to consider three things: 

Are we solving the right problem? 

Do we trust the evidence enough for this decision? 

Does it make sense alongside what else we know? 

When the evidence clears that threshold, the organization can move forward. 

When something material remains unresolved, the next step is to strengthen the evidence in the way that adds the most decision value. That might mean reframing the original question, looking at an existing data source, benchmarking the result or adding further research. 

The purpose of the pause is simple: to make sure that speed continues to serve judgement. 

From individual decisions to an organizational playbook 

This is where the next stage of AI maturity becomes tangible. 

Organizations can codify what they learn: where synthetic evidence has consistently performed well, which types of decisions can already be supported directly, which contextual signals are most useful and where a deliberate reflection point creates additional value. 

That knowledge becomes a living decision playbook. 

And the playbook should evolve. As models improve, validation accumulates and teams gain experience, the range of decisions that can be supported confidently by AI and synthetic evidence should expand. 

The result is a fundamentally different operating model for Insights: less effort spent repeatedly proving that an approach works, and more focus on applying accumulated learning to the next decision. 

The bigger opportunity 

For me, the most exciting part is not simply doing existing research faster. 

It is expanding the number of decisions that can become insight-led. 

AI and synthetic data make evidence available for questions that previously sat outside the practical reach of research. At the same time, organizations can direct additional research effort toward the decisions where it genuinely changes the level of trust and improves the quality of the outcome. 

At Toluna, we have already conducted more than 3,000 synthetic persona validations across 20+ markets, measuring how closely synthetic-persona responses align with results from matched human samples. 

That accumulated evidence gives us a growing foundation for understanding where synthetic approaches are already ready to support decisions and where additional evidence still adds meaningful value. 

The opportunity now is to turn that learning into repeatable organizational practice. 

Because the next stage of AI maturity in Insights is not simply about moving faster. 

It is about knowing when the evidence is strong enough to move — and when a short moment of reflection makes the decision better.