.png)
In the 2026 State of Conversational AI, we surveyed 30 enterprise leaders actively running conversational AI programs across financial services, healthcare, retail, government, and telecom.
Most of them (67%) are expanding or scaling their programs, and they reported an average confidence score of just 4.37 out of 7 in conversational AI's ability to handle complex customer conversations. Of the 20% who described themselves as skeptical of AI's capabilities, nearly all of them are scaling their programs anyway.
This is the defining tension of enterprise conversational AI in 2026, when the opportunities of embracing new technology outweigh the risks of standing still.
This post draws on findings from Rasa's 2026 State of Conversational AI report, which surveyed 30 enterprise decision-makers across finance, healthcare, retail, government, and telecom. Download the full report →
Waiting has its costs
The data poses an important question: If enterprise leaders aren’t confident that conversational AI is driving results, why are they investing in it?
The answer might be found in the demand from customers for high-quality self-service experiences. To meet those expectations, competitors are racing to deploy technology that can best meet customers’ needs so they can stand apart in the marketplace. As contact center costs continue to rise, AI-handled interactions also cost a fraction of their human-only counterparts while promising improved first-contact resolution, revenue outcomes, and retention rates. Against that backdrop, even moderate conviction can justify moving forward.
What our data captured, in other words, is that the risk of falling behind has started to outweigh the risk of building with less than full confidence.
What low confidence actually reflects
It's worth specifying where enterprise leaders lack confidence. Our survey asked about conversational AI's ability to handle “complex customer conversations” reliably. That's a more demanding bar than "does AI work at all" — and it points to something real about where the market currently stands.
Most enterprises aren't questioning whether conversational AI belongs in their customer operations. That debate is largely over. What they're less sure about is whether their systems can handle more complicated conversations, logical escalation paths, compliance-sensitive flows, and the nuanced judgment calls that many human conversations demand. That tested, production-grade reliability is where many leaders want to see proof before they’re all in. And when you’re innovating with new technology, you often need to deploy it before you can really know how it performs.
Confidence levels vary at the top of the org chart
One exception stands out. Within the C-suite, confidence scores ranged from 2 to 7, the widest variance of any group in our survey. The average lands near the overall mean, but the spread signals something important: At many organizations, leadership isn't aligned on how well their AI is actually performing.
That disagreement matters. Budget allocation, governance frameworks, vendor relationships, and build-versus-buy decisions all get made in rooms where leadership consensus is assumed. When the most and least confident voices in the organization are both in those rooms, it can lead to productive discussions and critical forks in strategy. Each company is dealing with its own goals and internal perceptions of AI, and that’s a good thing.
Shaky confidence underscores the importance of performance measurement
In our survey, achieving performance metrics is the most commonly cited pain point along the AI journey, outpacing deployment difficulty by nearly 3:1. Most organizations find themselves without the infrastructure they need to measure what their AI is actually doing (often by way of evaluation frameworks, testing protocols, and human review processes) because they treated measurement as something to build later. By the time those teams try to retrofit a metrics framework, they're already under pressure to justify investments and improve results, with limited visibility into what's failing or why.
That dynamic compounds the confidence problem. Low confidence without a measurement framework means leaders don't have a reliable way to know whether their confidence should, in fact, be higher. They're operating on instinct instead of data.
The enterprises best positioned to close the confidence gap aren't necessarily the ones with the most sophisticated AI. They're the ones that built the infrastructure to see what their AI is doing clearly enough to improve it. That means planning for response accuracy tracking, compliance adherence monitoring, containment rate analysis, and escalation frequency before they scaled, not after.
Rasa's platform is built around the principle that visibility and control are design decisions, not features to layer on later. For organizations scaling through uncertainty, that architecture difference matters.
The reality of the current enterprise AI market
Much of the current conversation around enterprise AI adoption doesn't describe the experiences enterprise leaders are actually having. Vendor narratives skew starry eyed, while case studies only highlight the companies that got it right.
What's less visible is the majority: Leaders who are building because they’re optimistic but skeptical, at organizations where alignment at the top isn’t guaranteed and performance measurement infrastructure is still catching up to organizational goals and innovation.
Our survey reflects a market doing difficult work under difficult conditions, making carefully calculated bets with imperfect information. These enterprise leaders are looking, more than anything else, for honest guidance on what "good" actually looks like from here.
If that describes where your organization is, you're in good company. The question isn't whether to keep building. It's how to build in a way that closes the gap between where confidence is and where it needs to be.
Go deeper on confidence, control, and the state of enterprise AI
Our 2026 State of Conversational AI report breaks down how challenges, expectations, and design preferences vary across industries, roles, and titles.
See where confidence is highest, which teams are actually driving infrastructure decisions, and what the organizations with the clearest path forward are doing differently.
