Conversational AI Platforms For Employee Services

Forrester first evaluated this market in 2019 under the heading chatbots for IT services. The framing tells you everything about the ambition at the time. A chatbot. For IT. A menu tree with a text box in front of it, deflecting password resets.

The analyst who has now run the evaluation twice, Will White, put the distance travelled bluntly in his commentary on the current edition: the market has progressed further than he thought possible. Agentic AI agents are now common, meaning agents that autonomously decide how to resolve a user request, in production.

And then he asks the question that actually matters. What still matters, if we have got these magical AI tools in production?

The answer Forrester arrived at is compressed into the report's own tagline, which is unusually pointed for analyst material: agentic AI is everywhere, so pick vendors who make it useful.

Everywhere and useful are doing different jobs in that sentence. Everywhere is a statement about supply. Useful is a statement about how little of it lands.

The gap between everywhere and useful

An employee services platform sits between a workforce and the systems that serve it. Someone needs a laptop, a policy answer, an expense approval, access to a shared drive, a change to their tax withholding, a replacement badge.

Historically each of those went somewhere different. IT service management for the laptop, an HR portal for the withholding, facilities for the badge, and a shared inbox for whatever did not fit. The employee's problem was navigation as much as resolution.

Conversational platforms collapse the navigation. One place to ask, natural language rather than a form, and the platform works out which backend system holds the answer and whether it can act on it directly.

The agentic version goes further. Rather than routing a request to a queue, the agent reasons about it, decides a resolution path, calls the systems required, and closes the loop. Forrester's own description of the market shift is that it is moving from traditional conversational interfaces toward agentic AI that can reason and act on user requests across a wide range of tasks.

Which is genuinely impressive and genuinely insufficient. An agent that can reason about a request still needs to know your leave policy, have permission to provision the software, understand which of your four expense systems applies to a contractor, and recognise when the answer is that this requires a human. None of that comes from the model. All of it comes from integration, context, and configuration, which is where the useful in Forrester's tagline is hiding.

Inside The Forrester Wave: Conversational AI Platforms For Employee Services, Q3 2026

The current evaluation scored thirteen providers on current offering, strategy, and customer feedback, published in Q3 2026.

Four Leaders: Kore.ai, Salesforce, ServiceNow, and Atlassian.

Kore.ai took the highest possible score in eleven of the evaluation criteria, and is the only provider holding Leader status in both this evaluation and its sibling covering conversational AI for customer service. Its recognised strengths run to legacy system integration, AI model management, and hybrid pricing options, which is a notably unglamorous list for a market this excited about itself.

Atlassian's assessment centred on context rather than conversation. Forrester credited its Teamwork Graph and its ability to embed Rovo into collaborative surfaces including multiuser chat channels, with a strategy prioritising connected user context across tools, proactive support, and embedding the assistant where people already work. The positioning Forrester offered is specific: ideal for existing Atlassian customers and for organisations wanting to automate more service experiences within the platform while experimenting with cross-vendor agents.

Salesforce converges HR, IT, and customer support onto Agentforce Service with a common knowledge and data layer, and draws context from Slack to improve answers to employee queries. Whether an employee services platform should come from a CRM vendor is a fair question, and the counter-argument is that the data layer is the same problem either way.

ServiceNow arrives as the largest provider in employee service management, particularly IT service management, and Forrester's attention there focuses on Moveworks, now included in certain ServiceNow packages following acquisition.

Forrester also flagged a limitation worth carrying into vendor conversations: Microsoft and IBM lack the depth in preconfigured use cases that their expansive ecosystems might lead a buyer to expect. Ecosystem breadth is not the same as employee services depth, and the gap between those two is where implementation timelines expand.

EasyVista appeared for the first time, one of few IT service management vendors in the evaluation, assessed on its conversational and agentic system integrated with its service manager.

What happened to the specialists

The previous edition tells a different story, and the comparison is the most instructive thing available in this category.

The Forrester Wave: Conversational AI Platforms For Employee Services, Q3 2024 evaluated fifteen providers against twenty eight criteria. Moveworks placed as a Leader, tying for the top rank in strategy with the maximum score in five of six strategy criteria including vision, innovation, roadmap, and pricing flexibility, plus top marks in fourteen of twenty current offering criteria. Forrester described it as ideal for customers wanting an easy-to-adopt solution with access to a capable and complex language model chain.

Two years later Moveworks is a ServiceNow product.

That single fact reframes the current Leader list. Three of the four Leaders are large platform vendors that own the systems employees are asking about. The fourth, Kore.ai, is the specialist that stayed independent. The specialist that led on strategy in 2024 is now a component of one of them.

The logic is not hard to follow. A conversational platform is only as good as its access to the systems behind it, and the vendors that own those systems start with an advantage no amount of integration work fully closes. Fifteen providers in 2024, thirteen in 2026, and the direction of travel is visible.

For a buyer this changes the shape of the decision. The realistic question is less which conversational platform is best and more whether your existing service management, HR, and collaboration estate already contains one that is good enough. Buying the specialist means better conversational capability and more integration work. Buying the incumbent means less work and accepting whatever depth it happens to have.

Forrester's own positioning language for Atlassian, ideal for existing customers, is an honest statement of exactly that trade-off.

Governance has come a long way and has a long way to go

The most useful line in Forrester's commentary is also the least satisfying, and it is that governance in this market has improved substantially and remains incomplete.

Both halves matter. Two years ago the governance conversation was about whether a bot might say something embarrassing. The controls that emerged, guardrails on responses, grounding in approved knowledge, and escalation rules, largely addressed that problem.

Agentic changes the exposure entirely. An agent that autonomously decides how to resolve a request is an agent that takes actions against production systems on behalf of an employee. It provisions access. It approves expenses within a threshold. It changes records.

The governance questions that follow are not about tone. What is this agent permitted to do without human approval? Under whose authority does it act, its own identity or the requesting employee's? What happens when it takes a wrong action, how quickly is that detected, and can it be reversed? What record exists afterwards?

Those are the same questions Forrester's separate agentic control plane category exists to answer, which is itself a signal. When a firm establishes a distinct market category for governing agents independently of the platforms that run them, it is saying that platform-native governance is not sufficient on its own.

The practical implication for anyone evaluating here is to treat governance depth as a first-order criterion rather than a compliance checkbox, and to be specific about it. Ask what the agent can do unsupervised, ask to see the audit record of an agent-initiated action, and ask what reversal looks like. The answers vary considerably more than the marketing does.

The correction nobody expected

Forrester's commentary opens with a sentence that reads almost like relief: the AI market has finally come back around to the correct conclusion that people are important.

That is a notable thing for an analyst to write about a market whose entire commercial pitch has been automation, and the evidence for it shows up in the vendor assessments.

EasyVista's recognised strength was its ability to operate alongside human agents, and specifically that experts can guide AI agents in real time rather than simply receiving a transferred request. That is a meaningful inversion of the standard escalation model. The conventional pattern is that the agent tries, fails, and hands the problem to a human who starts over. The pattern being credited here is a human steering the agent mid-resolution, keeping the automation's speed while adding judgement.

Atlassian's strength in multiuser chat channels points the same direction. An agent in a shared channel is participating in human collaboration rather than replacing it, and the design assumption is that people are still in the room.

This matters commercially because the deflection metric that has governed employee services investment is a poor fit for what these platforms now do. Deflection counts requests that never reached a human, which made sense when the alternative was a ticket in a queue. It counts nothing about whether the employee got a good outcome, and it actively penalises the human-in-the-loop patterns Forrester is now crediting, because a request an expert guided to resolution counts as a failure of deflection.

An organisation measuring this category on deflection will systematically undervalue exactly the capability that separates the current Leaders.

Where this leaves the category

The 2019 version of this market solved a navigation problem. The 2026 version solves a resolution problem, and the difference in ambition is the difference between telling someone where the form is and completing the process for them.

What has not changed is where the difficulty sits. It was never the conversation. It was always the systems behind it, the permissions governing them, the quality of the knowledge they contain, and the willingness of the organisation to let software act on its behalf.

Which is why the Leader list looks the way it does. Three platform vendors that own the systems, one specialist that got very good at connecting to everyone else's, and a consolidating field around them. And why Kore.ai's cited strengths are legacy integration, model management, and pricing structure rather than anything about conversation quality.

Forrester's own thirty-vendor Landscape of this market, published in Q1 2026, describes vendors varying by size, offering type, geography, and use case, which is the analyst way of saying that thirteen scored providers do not describe the whole field. A buyer whose requirement is narrow, regional, or specific to one function may find the right answer outside the Wave entirely.

The tagline is the thing to hold onto. Agentic AI is everywhere. Useful is the scarce part, and useful is mostly determined by decisions your organisation makes about access, knowledge, and authority rather than by which platform you license.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.