Four things to demand from AI-powered consumer insights

Sensia Gen 8 · For brand and insights teams

A brand team does not need another summary of what consumers say. It needs to decide which promise to lead with, which barrier to fix first, where to innovate and how to stand apart from the products next to it on the shelf. AI now produces fluent consumer insight in minutes; what is harder to see is whether that insight can carry a decision, and what it rests on. Here are four things worth demanding, and how Sensia Gen 8, the new generation of our engine, is built to meet them.

What a decision-ready analysis looks like

Take a question many skincare teams face: where is there still room to grow in face serums? A Sensia analysis of opportunity territories answers with a priority, a reading of who holds which position, and the consumer voices behind it.

Illustration of a Sensia analysis of opportunity territories, face serums, Europe. The case shows the format of the output; it is not a measure of the engine's performance.

This is the kind of output Gen 8 is built to produce at the scale of a whole category. Its specialised models read every review in the category's own language and know what each one says, about which product and with what tone. For each question, the engine selects the material that matters and passes it to a general-purpose language model for synthesis. The result can then be explored in conversation: a team can ask why, narrow the question to a range, or look at a competitor more closely.

Each of the four criteria asks something of an analysis like this one.

1. Priorities consistent enough to use

A team needs a dependable reading of what matters: the needs to address, the barriers to investigate and the opportunities worth pursuing. These findings can shape a claims architecture, an innovation route or a creative brief. A useful test is whether that strategic reading holds when the analysis is given substantially more evidence or uses a different language model.

We tested this on five production analyses, regenerated at six levels of depth, from a few hundred reviews per section to the writing model's maximum context window. On two sections, we also used a more powerful reasoning model. Across the conditions tested, the main findings remained consistent: the core opportunity territories, the leading choice driver and the central trade-offs held in the analyses where they were assessed.

For a brand team, the value is a stable basis for discussion and action, with the option to explore the supporting evidence more deeply when the brief requires it. Additional depth can provide more detail without requiring the team to reconsider the main strategic reading.

2. A market scope that is explicit and relevant

A market-level finding and a product-specific question require an explicit scope. The evidence should reach the products carrying relevant feedback on the subject, while also allowing the team to check the particular ranges, launches or competitors its brief concerns. Product breadth and the weight of the relevant feedback are complementary ways of examining that scope.

What to look for is a stated reach. We rebuilt, for four questions, all the relevant reviews available in each portfolio, and measured the share of them associated with the products present in the analysed sample. This measures how far the sample reaches across what consumers wrote. It is not a share of consumers, and it is not a percentage of insights. With around 1,500 selected reviews per question, the products in the evidence accounted for 98.1% to 99.5% of the relevant reviews.

These results show the broad reach of the selected evidence across the relevant review volume. For a specific brief, the next step is to confirm the products and topics to investigate, then choose the depth appropriate to that question. Gen 8 supports that adjustment, from a focused selection to a much larger body of evidence.

3. Evidence you can open

A finding has to travel: to a leadership meeting, an agency brief, an innovation committee. It travels better with the words of real consumers, and when anyone can check which product they are talking about.

What to look for is traceability: the ability to move from a finding to the consumer voices supporting it, and from each voice to its source. The analyses examined contained dozens of consumer extracts tied to products or brands. Greater depth could bring additional supporting detail or longer quotations; the essential requirement remains the ability to inspect the evidence behind the finding. The quoted consumer voices are real reviews, not text generated or completed by AI.

Traceability in a Sensia analysis: from a finding back to the review it comes from.

Measurement and synthesis have distinct roles: Sensia computes counts and weights on qualified review data, while the language model develops the narrative analysis from the prepared evidence.

For a brand team, it means a quote can go in front of leadership with the product it refers to, and a claim can be checked when someone asks.

4. A reading framed for a decision

An insight earns its place when it opens a decision: a positioning to test, a formula to adjust, a message to change. A list of themes describes a market; a reading framed for a decision helps a team move in it.

What to look for is a finding expressed in the team's own terms, with the products involved and a clear next question. In the skincare example above, the analysis does not stop at "tolerance matters". It shows which positions are held and asks whether there is room for brightening designed for sensitive skin: a question that can become a concept test, an innovation route or a brief. Because the analysis can be explored in conversation, the next step is often another question: which products own that space today, what do their critics say, does the pattern hold in another market?

Insight without the usual trade-offs

For years, consumer insight has meant choosing. Panels measure very well what happened. Studies answer the questions there is budget and time to ask. Between the two, some decisive questions stay open, or get answered on a slice of the market. When an analysis can be fast, reach the part of the market that matters for your question and rest on real voices at the same time, which questions would your team finally put on the table?

About the tests: real Sensia analyses on three large reference portfolios in skincare and pet food, regenerated with six amounts of evidence and, on two sections, with a more powerful reasoning model, October 2026. Reach measured on four questions. The conditions, measures and full results are in the Gen 8 benchmark report.