The build vs. buy question has always been part of the embedded analytics conversation. But for most of the last decade, it was a relatively straightforward decision. 

If you needed analytics in your product, you either built a custom solution in-house or integrated a third-party tool. The trade-offs were well understood: more control vs. faster time-to-market, more flexibility vs. less engineering overhead.

AI has scrambled that calculation.

Our Embedded Analytics Opportunity 2026 report found that 78.9% of product leaders say AI has changed their approach to building vs. buying embedded analytics. That's a fundamental reassessment of how teams think about their analytics stack. And the direction of that shift might not be what you'd expect.

Here's what the data tells us about how AI is reshaping the build vs. buy decision, where teams are heading, and how to think through the choice for your own product.

AI is pulling teams toward building in-house

The most striking finding in our data is the extent to which AI is pushing product teams toward internal development.

Among respondents who say AI has changed their approach, 33.3% report that they previously had no embedded analytics at all, but AI capabilities convinced them to build it. Another 26.7% say they used to buy analytics from a partner but now build it themselves. Combined, that's 60% of the teams whose approach has shifted, moving in the direction of building.

This is a significant trend. AI isn't just making embedded analytics more appealing – it's making teams feel they need to own it. And the reasoning isn't hard to follow. AI-powered analytics features like conversational interfaces, predictive models, and agentic workflows often need deep integration with a product's specific data, user context, and workflows. That level of customization can be difficult to achieve with an off-the-shelf solution.

There's also the differentiation argument. When 90.5% of product teams plan to invest in AI-powered analytics in the next 12–18 months, having a generic analytics layer that looks the same as your competitors' starts to feel like a liability. Teams want to build something distinctive – and that usually means building it themselves.

But the pull toward building comes with real risks, and the data elsewhere in our report makes those risks hard to ignore.

The complexity trap

Here's the tension: the same teams being drawn toward building are also the ones most likely to struggle with the technical realities of doing so.

Across our report, technical complexity and engineering resources are consistently the number one barrier to embedded analytics, cited by 57.1% of teams who've already implemented the capability and 42.9% of those still considering it. Data security (42.9%) and tech stack integration (35.7%) are the next biggest challenges – all foundational, engineering-heavy problems.

Building embedded analytics in-house doesn't just mean building it once. It means committing to ongoing maintenance, security, performance optimization, and feature development as data volumes grow and user expectations evolve. 

When you layer AI on top of that – with its own infrastructure requirements, model management, and accuracy challenges – the engineering commitment multiplies.

Shubhojeet Sarkar, Senior Group Product Manager at Meta, was direct about this trade-off: 

“Building a general-purpose analytics layer is an expensive distraction from the product problems only you can solve. Most teams that build here underestimate the ongoing cost of maintaining it as data volumes and schema complexity grow.”

So the question isn't just, “Can you build it?” It's whether the ongoing cost of building and maintaining it is the best use of your engineering capacity – especially when that capacity is already the most constrained resource on your team.

The hybrid model is gaining ground

While building dominates the shift, a meaningful number of teams are landing on a middle path. Among those whose approach has changed, 23.4% say they used to build analytics internally but now use a mix of both in-house development and external solutions. 

Smaller shares report adopting hybrid approaches from other starting points – 6.7% shifting from buying to a blend, and 3.3% entering the space for the first time with a combined model.

The hybrid approach tries to resolve the tension between customization and complexity. The idea is to use third-party solutions for the heavy infrastructure – query engines, visualization layers, natural language interfaces – while building the product-specific intelligence in-house: the data models, the insight ranking, the action integrations that are unique to your users' workflows.

Shubhojeet articulated this boundary clearly: “Hybridize when the workflow is proprietary, but the infrastructure isn't. Buy the query engine, the visualization layer, and the NL interface. Build the schema logic, the insight ranking, the action integration, and the parts that require understanding your users’ decision patterns.”

That framing is useful because it gives product leaders a concrete way to draw the line. Making the more relevant question: “Which parts of the analytics experience are commodity infrastructure, and which parts are where our product's differentiation lives?”

Very few teams are moving toward buying

It's worth noting what's not happening. Only 6.6% of teams whose approach has changed are moving toward buying from a partner, split between those who previously built internally (3.3%) and those entering the space for the first time (3.3%).

That's a remarkably small number, and it signals a meaningful shift in sentiment. 

As recently as a few years ago, buying was the default recommendation for most product teams – especially those without deep analytics expertise. The logic was straightforward: analytics infrastructure is complex, specialized vendors do it better, and your engineers should focus on your core product.

AI seems to be changing that logic. When analytics becomes a potential differentiator (and 30% of teams already say AI has helped differentiate their product), handing it off to a third party feels like ceding strategic ground. The desire for tighter integration, proprietary data loops, and distinctive user experiences is outweighing the convenience of an off-the-shelf solution.

But this shift also raises a question that not enough teams are asking: just because you want to build doesn't mean you should. The 57.1% of teams citing technical complexity as their biggest challenge didn't arrive there by accident. 

The ambition to build needs to be tempered by a realistic assessment of engineering capacity, data readiness, and long-term maintenance costs.

How AI is changing what teams actually build

Beyond the build vs. buy decision itself, our report captured how AI is changing the nature of the analytics experience teams are creating. When we asked respondents how AI has shifted their approach, some key themes emerged – and they paint a picture of a capability in transition.

The most common shift is from static dashboards to contextual insights. Teams describe moving toward interactive, conversational experiences where users can ask questions and receive answers dynamically, rather than navigating pre-built reports. 

As one respondent put it: “We are transitioning from static dashboards to contextual insights. Some of our internal tools already have a conversational interface, and now analytics are accessible to non-technical users without (or minimal) training.”

Greater personalization is another recurring theme. AI is enabling analytics experiences that adapt to the individual user – surfacing the most relevant data based on their role, behavior, and context, rather than serving the same generic dashboards to everyone.

And for some teams, AI is specifically enabling in-house development that wasn't feasible before. Multiple respondents noted that AI has “opened up opportunities to build new features in-house” and created “new possibilities for a completely bespoke user experience.”

That said, a significant portion of respondents say AI hasn't meaningfully changed their approach yet. Some are constrained by tech stack dependencies, others are taking a wait-and-see approach, and some are simply still figuring it out. One respondent was candid: “We're reviewing it, but we don't want to risk getting caught out if the bubble bursts.”

How to think through the decision for your product

There's no universal right answer to the build vs. buy question – it depends on your product, your team, your resources, and where analytics sits in your value proposition. But the data and expert insights from our report point toward a few principles that can help sharpen the decision.

Ask where your differentiation actually lives

This is the most important question, and it cuts through a lot of the noise. 

If your analytics experience is a core part of why customers choose your product – if the quality of insights, the personalization, or the connection between data and action is a genuine differentiator – then owning that experience makes strategic sense.

But if analytics is something your users expect, but it's not the reason they chose you, the case for building weakens. As Shubhojeet put it: “If analytics is table stakes – users expect it, but it's not why they chose you – buy.”

Be honest with yourself here. It's easy to convince yourself that analytics is core to your product's value when in reality it's a supporting feature that would be served just as well by a partner solution.

Consider the feedback loop

One of the more nuanced points from our research is about behavioral signals. AI-powered analytics improves over time when it can learn from how users interact with insights – what they act on, what they ignore, what they reformulate. That feedback loop is valuable, and it's harder to close when the analytics layer is owned by a third party.

Shubhojeet framed this as a key decision filter: “Do you own the behavioral signal? If your users' interaction patterns are the training data for better insights, that signal has to stay inside your stack.”

If you're building AI features where user interaction data feeds back into improving the experience, that's a strong argument for ownership. If your analytics are more static, the feedback loop argument weakens, and buying becomes more defensible.

Be realistic about what you can sustain

The report consistently shows that engineering resources are the biggest constraint on embedded analytics. If your team is already stretched, committing to building an analytics stack from scratch – plus the AI infrastructure on top of it – could crowd out higher-priority work.

This is where the hybrid model has real appeal. By buying the foundational infrastructure and focusing your engineering effort on the differentiated layer, you can move faster without shouldering the full burden of maintaining the entire stack.

As Lara Atici, Head of AI Product Management at Talent-Ray, put it: 

“Product leaders should choose the option that best enables their teams to own and evolve the analytics capability in the long term. In the end, the right solution is the one that fits user needs and drives adoption, because tools only create value if people actively use them.”

The real question isn't build or buy – it's what to own

If there's one takeaway from the data, it's that the build vs. buy framing is increasingly too binary for the reality of embedded analytics in 2026. The more useful question is: what do you need to own, and what can you leverage from others?

For most teams, the answer will land somewhere in the middle. Own the parts of the analytics experience that are unique to your product and your users' workflows. Leverage external solutions for the infrastructure and tooling that's common across the industry.

The teams that get this balance right won't just ship embedded analytics faster – they'll build experiences that are genuinely differentiated, deeply integrated with their product, and able to evolve as AI continues to reshape what users expect from the tools they use every day.

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. 👇