There's no shortage of excitement around AI in product analytics right now. 

Conversational interfaces, predictive insights, agentic workflows – the vision is compelling, and product teams are paying attention. But when you look at what's actually shipped versus what's still on a whiteboard, the picture is more measured than the buzz suggests.

Our Embedded Analytics Opportunity 2026 report found that while the vast majority of product teams plan to invest in AI-powered analytics, very few have anything live in front of customers yet. The gap between intention and implementation is wide, and understanding why matters for any product leader trying to figure out where AI fits in their analytics strategy.

Here's what the data tells us about where teams are today, what's working, what's not, and where the real opportunities lie.

Almost everyone is planning to invest, but very few have shipped

The headline numbers tell a clear story of enthusiasm outpacing execution. 

Among teams using or planning embedded analytics, 90.5% say they plan to invest in AI within their product analytics experience in the next 12–18 months. That's an overwhelming majority.

But zoom in on where teams actually are today, and the picture shifts. Only 14.2% of teams have a live or near-live AI-powered analytics implementation. The majority (57.2%) describe their current state as “experimenting” through proofs-of-concept, while 28.6% say they aren't using AI-powered analytics at all.

So we're in a landscape where nearly everyone plans to do this, but most are still figuring out what “this” even looks like in practice. That's not necessarily a bad thing – experimentation is a healthy phase. 

But it does mean product leaders should be cautious about treating AI-powered analytics as a solved problem or a near-term differentiator. For most teams, there's still a meaningful gap between the prototype and production.

Where teams are experimenting – and what's getting traction

Among teams actively exploring AI in their analytics capabilities, a few use cases are emerging as the most common starting points.

Conversational interfaces for personalized analytics top the list, cited by 50% of respondents. This makes intuitive sense – natural language interaction is one of the most visible and accessible ways to make analytics feel different for users. 

Instead of navigating filters and dashboards, users can ask a question and get a direct answer. It's the kind of experience that demos well and addresses a real pain point: 71.4% of product teams say their users want to intuitively find answers to their own questions.

Beyond conversational interfaces, the use cases are fairly evenly distributed. Predictive analytics and forecasting are being explored by 43.3% of teams, while AI-augmented dashboards and reports, and agentic capabilities (recommending next steps or taking actions within workflows) are both at 36.7%.

What's interesting here is the breadth. Teams aren't converging on a single killer use case – they're experimenting across the spectrum, from making existing dashboards smarter to building entirely new interaction models. That suggests the space is still early enough that no one has figured out the definitive best way to do AI in embedded analytics.

Lara Atici, Head of AI Product Management at Talent-Ray, sees the near-term opportunities clustering around accessibility and action: natural language querying, automated anomaly and trend detection, decision recommendations, and personalized analytics based on the user's role and behavior. 

As she put it: “AI shifts embedded analytics from passive dashboards to proactive decision support.”

AI features are live, but they're not dominating usage

Even among teams that have introduced AI features into their analytics experience, usage remains relatively modest. Nearly half of respondents (46.6%) say less than 20% of their overall analytics usage comes from AI-driven features. Another 26.7% report AI accounting for 21–40% of usage.

There is a smaller group seeing deeper adoption – 13.4% say AI drives 41–60% of usage, and 13.3% report over 60%. But these are the exceptions, not the norm.

This is worth keeping in mind when setting expectations internally. AI features in embedded analytics are, for most products right now, a complement to existing experiences – not a replacement. Users are still relying primarily on the traditional dashboards and reports they're familiar with, with AI features playing a supporting role.

This is how new interaction models tend to roll out. But it means product teams should resist the temptation to rip out traditional analytics in favor of AI-driven alternatives. As one survey respondent cautioned: “I don't suggest going straight to a conversational interface and replacing dashboards. We saw that users still want structured visibility.”

The transition will be gradual, and the smartest teams are layering AI on top of existing experiences rather than replacing them wholesale.

The impact so far: Differentiation now, revenue later

When asked about the actual impact of AI on their analytics experience, the most commonly cited outcome is competitive differentiation, reported by 30% of teams. Reduced backlog and increased feature engagement follow at 20% each.

Direct commercial outcomes are less common so far. Only 13.3% report that AI has generated direct revenue, and just 6.7% say it's improved user retention. For most teams, AI isn't yet a major business driver in embedded analytics – its primary value today is helping products stand out in the market and making development teams more efficient.

What's notable is how this compares with what teams expect AI to deliver. When asked about anticipated impact, 31.6% point to increased feature engagement, 23.7% expect direct revenue, and 21% anticipate improved retention. In almost every category, expectations run ahead of current results.

The one exception? Competitive differentiation. Only 18.4% expect this to be the biggest future impact, even though 30% of teams already report it today. 

This suggests teams may be undervaluing one of AI's most immediate and tangible benefits – making your product feel more modern and capable relative to competitors – while overestimating how quickly AI will move the revenue needle.

For product leaders, that's a useful reframe. The near-term case for AI in analytics might be less about commercial returns and more about market positioning. Revenue will likely follow as implementations mature, but differentiation is available right now.

The trust problem is the biggest obstacle

If there's one data point from the AI section of our report that product leaders should circle and underline, it's this: 57.1% of respondents say incorrect or misleading outputs are their top concern when it comes to AI-driven analytics in their product.

That number dwarfs every other concern. Regulatory or compliance risk comes in at 14.3%, engineering complexity at 9.5%, and lack of explainability at 7.1%. Accuracy isn't just the biggest concern – it's the concern.

When you're surfacing AI-generated insights inside a product – insights that users may base real business decisions on – accuracy isn't a nice-to-have. It's the whole ballgame. A wrong recommendation, a hallucinated trend, or a miscalculated forecast doesn't just frustrate users. It erodes trust in the entire analytics experience – and that trust is extremely hard to rebuild.

Shubhojeet Sarkar, Senior Group Product Manager at Meta, put it bluntly: “A miscalibrated insight erodes trust faster than no insight at all, as users who get burned by a false positive start ignoring the surface entirely.”

He also highlighted a measurement problem that makes this worse: “Engagement metrics don't distinguish ‘I looked and acted’ from ‘I looked and ignored.’” In other words, you might not even know trust is eroding until it's too late – because the user still visits the page, they just stop believing what they see.

It's worth noting that concerns shift as organizations grow. Smaller companies tend to focus more on security and data quality challenges, while the largest companies (10,000+ employees) place greater emphasis on explainability and user trust – likely reflecting the higher stakes involved when AI-powered analytics are deployed at enterprise scale.

Where the investment is heading next

Despite the challenges, investment momentum is strong. Among the 90.5% of teams planning AI investments in the next 12–18 months, conversational interfaces for personalized analytics lead the way at 60.5% – up from the 50% currently experimenting with this use case. 

Agents that recommend next steps or take actions within user workflows come in at 55.3%, signaling growing interest in moving analytics from descriptive statistics to tangible, actionable insights.

Predictive analytics and forecasting (42.1%) and AI-augmented dashboards (34.2%) round out the list, suggesting teams want to enhance both the intelligence and the accessibility of their existing analytics experiences.

The investment pattern tells a story about where product teams see the future heading: analytics that's more conversational, more proactive, and more tightly woven into the workflows users are already in. 

Internal tools vs. customer-facing products: different speeds of adoption

One nuance worth flagging is the difference in AI adoption between internal enterprise tools and customer-facing products.

Internal tools appear slightly further along in experimentation – 70% of those teams report experimenting with AI-powered analytics, compared with 53.3% of customer-facing product. Customer-facing products also show a higher share of teams with no current AI usage at all (33.3% vs. 20% for internal tools).

This suggests customer-facing teams are moving more cautiously, which is understandable. When you're putting AI-generated insights in front of paying customers, the bar for accuracy, reliability, and user experience is higher. Getting it wrong doesn't just cause internal frustration – it can damage customer trust and brand perception.

Internal tools, by contrast, offer a lower-risk environment to experiment and learn. The feedback loops are shorter, the tolerance for imperfection is higher, and the users are colleagues who can provide direct input on what's working and what isn't. For teams looking to build confidence with AI in analytics, starting with internal-facing use cases before rolling features out to customers is a pragmatic approach.

Advice from the front lines

One of the most valuable parts of our research was the open-ended advice from respondents who've been through the early stages of AI implementation. Three themes kept coming up.

First, start with the problem. The strongest advice consistently pointed back to product fundamentals: don't start with the AI capability you want to build, start with the user problem you're trying to solve. 

As one respondent put it: “Think about the question you are trying to answer, not the data you are looking for.” AI is a means to an end – and the end should be helping users make better decisions faster.

Second, experiment small before scaling. With 57.2% of teams still in the proof-of-concept phase, there's a temptation to try everything at once. But scattered experimentation rarely leads to clear learnings. 

One respondent captured this well: “Lots of user testing. Just because you can add AI features, it doesn't mean that's what users want.” Pick a specific use case, define what success looks like, and test it properly before broadening your scope.

Third, make trust a design principle. With accuracy as the dominant concern, trust needs to be built into AI-powered analytics from the ground up. That means prioritizing interpretability, giving users control surfaces to adjust or override outputs, and monitoring feedback loops – not just top-line KPIs. One respondent summed it up: “Governance shouldn’t be an afterthought.”

The opportunity is real – the timeline is longer than you think

AI is going to reshape embedded analytics. The data makes that clear. But for most product teams, that reshaping is measured in quarters and years, not weeks and months.

The teams that get this right won't be the ones who rush AI features to market. They'll be the ones who experiment thoughtfully, build for trust, and keep their focus on user outcomes rather than technology for its own sake.

And maybe that's the most important takeaway from this data: in a landscape where nearly everyone is experimenting, the real competitive edge comes not from moving fastest, but from learning fastest.

More embedded analytics insights...

This article is based on findings from the Embedded Analytics Opportunity 2026 report, produced in partnership with ThoughtSpot.

Download the full report for the complete dataset, expert commentary, and analysis across adoption, AI, and build vs. buy decisions. 👇