What Moveworks AI actually does
Most enterprise software starts as a tool you have to learn. Moveworks AI goes the other direction: it answers employee questions and completes routine requests inside the tools people already use, chiefly Slack and Microsoft Teams.
Think about the last time you needed a Figma license, a password reset, or a straight answer about parental leave policy. Someone filed a ticket, waited two days, and a human on the service desk did work that didn’t need a human. Moveworks exists to intercept that request, understand it, and either resolve it outright or hand it to the right system with context attached.
Founded in Mountain View in 2016 by Bhavin Shah, Vaibhav Nivargi, Jiang Chen, and Varun Singh — all ex-Google engineers — the company built its name in IT service management before pushing into HR, finance, and facilities. Customers include Broadcom, Palo Alto Networks, Unilever, Nutanix, and Databricks. Those are organizations where a few thousand employees generating a few thousand questions a month adds up to serious overhead.
Under the hood: how it understands a question
The marketing shorthand is “AI assistant.” The mechanics are more interesting than that.
Language models tuned for one domain
Moveworks doesn’t lean on a single off-the-shelf model. The platform blends large language models with models trained on enterprise-specific data: the vocabulary of your ticketing system, your application names, your org chart. A general-purpose chatbot may not know that “the VPN thing keeps dropping” maps to a specific client on macOS Sonoma. A tuned system does.
Retrieval and the reasoning layer
Answering well means finding the right source first. Moveworks crawls knowledge bases, closed tickets, SharePoint, Confluence, and policy documents, then uses retrieval to ground its response in something real rather than a plausible guess. Sitting on top of that is an orchestration layer — the company calls it a reasoning engine — which decides the order of operations: check entitlement, confirm the user’s location, call the HRIS, open the ticket, ping the approver.
That sequencing is the part IT leaders actually care about. Anyone can generate a paragraph. Far fewer systems can submit a correctly formatted access request to Okta and confirm back to the employee that it’s done.
Where it shows up during the workday
Deployments vary, but the patterns repeat across industries:
- Access and identity issues. Account unlocks, MFA device resets, application access granted according to role.
- Software provisioning. License requests checked against seat budgets, then routed to a manager for approval.
- HR policy questions. Leave balances, benefits eligibility, and onboarding timelines pulled from live documents instead of a stale FAQ.
- Onboarding and offboarding. Provisioning sequences that touch a dozen systems in a fixed order, where missing one step creates a security gap.
- Ticket triage. Classifying, prioritizing, and routing the requests that genuinely need a person.
The measurable win is deflection: a substantial share of incoming requests resolved without a human ever touching them, which frees service desk staff for work that requires judgment.
From chatbots to agents
The industry vocabulary shifted around 2023 and 2024, and Moveworks moved with it. Early virtual agents answered questions and handed off. The newer framing, agentic automation, means the system takes multi-step actions across systems on its own, with guardrails and an audit trail behind every step.
The practical difference for an IT leader: you aren’t buying an FAQ bot, you’re buying something that performs work. That changes the evaluation criteria entirely. Failure modes matter more. If a chatbot gives a wrong answer, someone shrugs. If an agent provisions the wrong access level, that’s a security incident. Permissions, logging, and rollback paths deserve as much attention in a demo as the natural-language quality.
What it costs and how teams evaluate it
Moveworks doesn’t publish list pricing. Deals are typically quoted per employee per year, varying with headcount, modules, and how many systems you want connected. Mid-size enterprises often land in the low-to-mid six figures annually; global rollouts go well north of that. Expect a discovery process, not a checkout page.
Questions worth asking in a pilot
A 60- to 90-day pilot against one defined slice of traffic beats a broad, vague rollout. Get answers on these before signing anything:
- What’s the realistic deflection rate for our ticket mix, not a reference customer’s?
- How does the system behave when it can’t answer — escalate cleanly, or improvise?
- Which connectors exist today, and which require custom engineering?
- How are permissions enforced, and can an agent be constrained to least-privilege actions?
- What does the audit log look like when compliance asks for it?
Run the pilot against a live queue rather than a curated demo set. The gap between benchmark performance and Tuesday-afternoon performance is where most AI projects get judged.
ServiceNow’s $2.85 billion bet
In March 2025, ServiceNow announced an agreement to acquire Moveworks for $2.85 billion, the largest purchase in the company’s history. The logic is easy to follow. ServiceNow dominates enterprise IT workflows. Moveworks brought a conversational front door and a reasoning layer built specifically for employee support. Combine them and you get an assistant sitting on top of the platform of record instead of beside it.
For existing customers, the live questions are about roadmap and integration. Multi-platform support, meaning the ability to run in Slack and Teams regardless of which ITSM tool sits underneath, was always part of the pitch. Consolidation tends to reshape those promises over time, so it’s worth watching how connector priorities shift and whether the platform stays as open as it has been.
Limits worth knowing before you buy
This isn’t a fit for every organization. Below roughly a thousand employees, ticket volume may not justify the contract — a well-run service desk and a clean knowledge base cover a lot of ground. Companies running heavily customized legacy systems will spend real effort on integration. And any vendor claiming near-total automation is describing its best customers, not the median one.
Accuracy also depends on what you feed it. If your knowledge base hasn’t been updated since 2021, no language model will rescue it. Teams that get the most from these platforms treat content hygiene as part of the project rather than an afterthought, and they assign someone to own it permanently.
Making the rollout stick
Technology is rarely why these programs fail. Adoption is. Employees keep emailing the service desk out of habit unless someone actively redirects them, and that redirect has to be visible: a pinned Slack channel in every workspace, a Teams bot added by default, a line in the onboarding checklist.
Set a baseline before launch. Track tickets per employee, average resolution time, and first-contact resolution for at least a quarter. After go-live, watch three numbers: deflection rate, escalation accuracy, and employee satisfaction with the answers they receive. Together those tell you whether the system is genuinely working or simply moving work around.
Start with one high-volume, low-risk use case — password resets are the classic for a reason. Get it to 80% automation. Write down what broke and why. Then expand, module by module, system by system. The organizations seeing the results in the case studies treat Moveworks as a multi-year operating change, not a software install with a go-live date.