Around this time last year, we posted a blog asking what the future would hold for Conversational AI in 2025. Here’s where we look back and see what happened versus what we predicted!

Where our predictions hit the mark (or pointed in the right direction)

Widespread piloting of conversational AI in customer-facing applications continues to grow

Our first prediction (that 2025 would see conversational AI widely piloted in customer-facing contexts) felt like a fairly safe one. Market reports from 2025 (such as this from MarketsandMarkets) indicate strong adoption of Conversational AI (CAI) solutions by enterprises seeking to improve customer service, support, and engagement.

Although it might sometimes be that there’s more talk than action when it comes to AI, McKinsey & Company’s global workplace AI report found that nearly all companies are investing in AI, though only a small fraction consider themselves “mature”, emphasising that many are still in pilot or early deployment phases. This aligns with the “piloting and incremental rollout” phase envisioned in our original blog.

Growing focus on tangible value, ROI and meaningful business impact

Our second prediction (that firms would shift from implementing AI because it’s new and shiny to more purpose-driven, ROI-oriented projects) seems to be unfolding, albeit sometimes more slowly than perhaps expected. That said, in 2025, conversations around AI in enterprises increasingly targeted measurable gains (efficiency, cost reduction, improved workflows) rather than novelty for its own sake.

Notably, recent analyses of conversational AI deployments report efficiency gains (e.g., in customer support, HR or sales workflows), helping underline that initial experiments are indeed being converted into purposeful, value-oriented use. (Wonderchat.io, Fullview.io)

Emergence of new LLM-based bot platforms and diversification of models

Our original predictions foresaw a wave of new bot-platforms based on large language models (LLMs). That wave has definitely arrived, with, for example, the release of Mistral AI’s Le Chat Enterprise, Botpress and Intercom. All the flagship model providers’ chatbots have continued to evolve this year, including Chat GPT now at 5.2, evolutions in Google’s Gemini and Anthropic’s Claude 3, and in fact it’s interactions with these products which many people now equate with AI. Meanwhile, consolidation and variety in LLM providers have expanded choices, both open and closed source, giving organisations more flexibility to pick models that match their use-case, cost, privacy or performance constraints.

Rise of “agentic AI” – but more gradual evolution than wholesale replacement

Our blog predicted we’d hear a lot more about “agentic AI” in 2025, and that there’d be evolution rather than revolution. That nuance seems accurate. The narrative within the industry is indeed shifting toward AI that can autonomously carry out tasks, yet enterprise deployments and real-world uses remain mostly semi-autonomous or assisted. (IBM)

Academic research published mid-2025 reflects this: a survey of LLM-driven conversational agents outlines the current capabilities including reasoning, tool use, and multi-turn dialogue, but also flags major gaps, such as self-monitoring, long-term consistency, and safe autonomy. (arXiv)

In practice, many real-world bots are hybrid, combining LLM-driven conversation with retrieval-augmented generation (RAG) and reliance on canned/intent-based responses to balance performance, reliability and safety when it’s most appropriate. (arXiv)

What’s unfolding more slowly than we hoped

Voice remains important, but it hasn’t “taken over” conversational AI

Our original prediction that 2025 might be “the year of voice” for conversational AI has only partially materialised. While voice assistants and voice-enabled bots continue to evolve, much of the momentum in 2025 remains in text-based LLM chatbots or hybrid systems.

Moreover, many enterprises deploying CAI are doing so in customer support, helpdesks and knowledge management, contexts where a text interface remains dominant because of clarity, traceability, and data management requirements.

For this reason, voice is growing, but it hasn’t (yet) replaced more traditional text-based assistant interactions in the way that we speculated.

Knowledge-library practices and content hygiene remain a stubborn bottleneck

Last year’s article argued (or more accurately, hoped!) that 2025 would force companies to take content strategy and knowledge-library maintenance seriously. The “garbage in = garbage out” adage remains true, but in many organisations, the challenge has proven deeper and more persistent than they might have anticipated.

Recent literature highlights that for large enterprises, building and maintaining reliable knowledge bases (for RAG, QA bots, etc.) remains one of the most significant barriers to scaling CAI systems effectively. (arXiv)

In short, while there may be engineering buy-in, organisational politics and practice can make content-engineering, editorial discipline, and taxonomy/ontology design as challenging as ever.

Analytics tooling is improving, but the landscape remains fragmented

Our 2024 piece predicted an increased demand for specialist bot and CX analytics that can handle the unstructured and semi-structured data generated by bots. There has been progress in tooling but there’s not yet a universal “analytics stack” for hybrid / agent-based CAI solutions. For many organisations, combining structured metrics with qualitative measures remains manual, ad-hoc and difficult to scale. Emerging industry tools increasingly focus on checking response quality, grounding accuracy and user feedback for RAG- and LLM-based chatbots.

Many organisations continue to rely on a patchwork of logs, dashboards and manual analysis, but it’s painful. Combining traditional performance metrics (such as usage, containment and resolution rates) with richer qualitative insight needs deep conversational AI understanding and can be difficult even for experienced enterprise tech teams. This challenge has driven the emergence of purpose-built platforms, such as Chatpulse, designed specifically to provide end-to-end visibility into conversational performance, quality and business impact across modern CAI architectures.

How did we do?

It feels fair to say that our predictions got most of the broad strokes right. 2025 has indeed been a pivotal year, not because conversational AI has “arrived” everywhere just yet, but because the industry is just beginning to shift from hype to pragmatism. The path forward for CAI is likely to be incremental and structural, and success will depend not just on having a “big model under the hood,” but on good content governance, modular architecture, observability, and a clear sense of what value the bot is meant to deliver.

We’d love to hear what conversational AI delights you have coming up for 2026, and as ever, you can ask us your burning CAI questions!