AI Service Desk Software for Internal IT: 9 Tools Compared for 2026

Table of Contents

• Why Internal IT Ticket Queues Break Before Customer Queues Do

• What an AI Service Desk Is, and How It Differs From a Help Desk

• How an AI Service Desk Actually Works, Step by Step

• The IT Requests AI Resolves Today Versus the Ones It Should Not

• Deflection Is Not Resolution

• How We Evaluated AI Service Desk Software for Internal IT

• The 9 Best AI Service Desk Tools in 2026

• Platform Summary Table

• What an AI Service Desk Costs in 2026

• Consolidation Watch: Who Owns Whom

• Compliance Duties for Employee-Facing AI Agents

• Why Rollouts Stall After Password Resets

• How to Choose the Right Platform

• Implementation Checklist

• Final Verdict

Why Internal IT Ticket Queues Break Before Customer Queues Do

An AI service desk is an internal IT support system where AI agents handle the front line: answering employee questions, resolving routine requests such as password resets and access grants, triaging and routing tickets, and escalating everything else to a human with full context attached. It sits on top of your ITSM platform rather than replacing it. The end user is an employee, not a customer, which changes the integration surface entirely.

Internal IT queues fail quietly. Nobody churns over a slow VPN ticket, so the pain shows up as lost engineering hours and a service desk manager who cannot staff to peak. Onboarding weeks, laptop refresh cycles, and quarterly access reviews all spike the same queue at once.

The market has noticed. SysAid, one of the top-five ranking properties for this term, promotes July 2026 research reporting that AI in ITSM has reached 61% adoption (SysAid). Adoption is no longer the differentiator. What separates a working deployment from a stalled pilot is which request types the agent can actually complete inside your identity and endpoint systems.

This guide is written for the IT side of the house. If you are buying an agent to answer customers rather than employees, our companion comparison of customer-facing AI support platforms covers that buying decision instead, with a CX vendor set and CSAT-shaped metrics.

What an AI Service Desk Is, and How It Differs From a Help Desk

An AI service desk resolves and acts; a help desk answers; a traditional service desk logs and routes. The distinction matters at procurement because the three categories are priced differently and integrate at different depths. A help desk chatbot needs a knowledge base. An AI service desk needs write access to Okta or Entra, your MDM, and your ITSM ticket object.

SysAid's glossary frames it as AI-driven virtual assistants and automation that manage service requests, incidents, and FAQs, using natural language processing to understand queries and route complex issues to humans (SysAid). That definition is accurate but stops short of the 2026 reality: current agents execute changes, not just classify them.

Read the table as a layering diagram, not a replacement chart. Most 2026 buyers keep ServiceNow, Jira Service Management, or Freshservice as the system of record and add a resolution layer above it.

How an AI Service Desk Actually Works, Step by Step

An AI service desk runs a six-stage loop on every inbound request: intake, intent classification, knowledge retrieval, action in connected systems, escalation with context, and a learning pass. The stages that break in production are almost always four and five, because those require write permissions and a clean handoff object, not better language modelling.

1. Intake. The request arrives through Slack, Teams, an employee portal, email, or a phone line. Voice intake needs telephony integration and, for a natural handoff, control over pacing and prosody.

2. Intent classification. The agent maps free text to an ITIL object: incident, service request, problem, or change candidate. Misclassification here is what produces the misrouted-ticket complaints in year one.

3. Knowledge retrieval. Grounding against approved runbooks, KB articles, past resolved tickets, and configuration data. Ungrounded generation in an IT context invents policy, which is worse than saying "I don't know."

4. Action. Reset a password in Entra, unlock an account, add a user to a security group with approval, push a software install through Intune or Jamf, provision a license in the SaaS admin console.

5. Escalation. The human fulfiller receives the transcript, the classification, the systems already checked, and the actions already attempted. Escalation without that payload is just a slower ticket.

6. Learning. Resolved tickets and fulfiller corrections feed back into retrieval so the same request resolves autonomously next month.

Stage four is where vendor claims diverge most. Ask every shortlisted vendor to demo an action that writes to a system of record, not a read-only lookup.

The IT Requests AI Resolves Today Versus the Ones It Should Not

AI service desks reliably close identity, access, and software requests where the action is an API call against a system with a clear audit trail. They should not close anything requiring physical presence, security judgement, or an irreversible change without human sign-off. Drawing that line before the pilot is the single best predictor of whether the deployment survives quarter two.

Resolve autonomously today:

• Password resets and MFA re-enrollment, with step-up identity verification

• Account unlocks after failed login thresholds

• Application access requests routed through an existing approval policy

• SaaS license provisioning, reassignment, and reclaim

• VPN and Wi-Fi connectivity triage using guided diagnostics

• Software installs and updates via MDM policy

• Onboarding and offboarding checklist execution

• Status lookups: ticket state, order state for hardware, approval position

Keep human, or gate behind explicit approval:

• Hardware failure diagnosis and replacement dispatch

• Security incidents, suspected phishing, and anything touching an active investigation

• Privileged access elevation to admin roles

• Change requests that touch production infrastructure

• Anything requiring physical action: badge issuance, desk-side visits, device collection

• HR-adjacent requests involving employment status or compensation data

Regulated environments narrow the list further. Financial services teams running kyc automation alongside internal IT, or health systems where the same agent touches prior authorization automation workflows, need action-level permissions rather than a blanket allow list.

Deflection Is Not Resolution

Deflection means the contact never reached a human queue. Resolution means the employee's problem is fixed and does not resurface. Vendors routinely report the first number and call it the second, which is why two platforms quoting identical percentages can deliver wildly different fulfiller workload reductions.

2026 benchmark write-ups stress this conflation directly, noting that a large share of contacts counted as deflected never reach genuine self-service resolution (Service Desk Agents). The reopened-ticket rate is the honest counter-metric: if 20% of "resolved" tickets reopen within seven days, your true resolution rate is 20 points lower than the dashboard says.

Two major vendors now bill on this definition, so it has moved from a marketing question to a contract question. Intercom publishes an explicit outcome definition: a customer confirms the issue is resolved, or does not ask for more help after Fin responds, or Fin completes a workflow, billed at $0.99 per outcome (Intercom). Zendesk bills AI agents per automated resolution with reported dual verification of billed resolutions, and does not publish a per-resolution rate (Richpanel).

Audit checklist for any vendor-reported resolution rate:

• Does the vendor's definition require positive user confirmation, or only the absence of a follow-up?

• Are abandoned sessions counted as resolutions?

• Is a reopened ticket within 7 days deducted from the count?

• Does a handoff to a human count as a resolution because the agent "completed a workflow"?

• What is the reported rate at 30 days versus 180 days post-launch?

• Is the figure measured on all inbound, or only on the subset routed to the agent?

How We Evaluated AI Service Desk Software for Internal IT

Nine platforms were scored against seven criteria weighted for the internal IT buyer, not the CX buyer. Pricing was taken from live vendor pages where published and labelled as third-party reported where the vendor blocks or omits it. Vendor-claimed performance percentages without a dated primary source were excluded rather than repeated.

ITSM stack compatibility. We checked whether the agent runs on top of ServiceNow, Jira Service Management, Freshservice, or a comparable system of record without forcing a migration. This matters because the ITSM platform holds your CMDB, SLA clocks, and approval workflows, and no IT org rips that out to buy a chatbot. Platforms that require becoming the system of record scored lower.

Action depth in identity and endpoint systems. Read-only answers do not reduce fulfiller load. We looked for native write actions against Okta or Microsoft Entra, MDM tools such as Intune and Jamf, and HRIS systems for joiner-mover-leaver flows. A vendor that can only draft a response for a human to send is an assist tool, not a service desk agent.

Resolution definition and measurement transparency. We asked whether the vendor publishes what counts as a resolution, whether reopened tickets are deducted, and whether the reported rate covers all inbound or a curated subset. Vendors billing on resolutions have a direct financial interest in a loose definition, so published definitions carry more weight than case study percentages.

Compliance posture for employee-facing agents. Minimum bar was soc 2 type ii and GDPR handling, with iso 27001 and documented data residency options weighted for EU and UK buyers. Since 2 August 2026, EU AI Act Article 50 disclosure duties also apply to employee-facing agents, so we checked whether the platform supports a persistent AI disclosure.

Pricing model transparency and TCO at volume. Six pricing units are now in play: per agent, per fulfiller, per employee, per session, per task, and per resolution. We modelled each at 5,000, 20,000, and 100,000 monthly tickets. Vendors that publish nothing at all were marked "does not publicly state" rather than filled in with third-party estimates.

Deployment timeline to first autonomous resolution. Time-to-value is measured from contract signature to the first ticket closed without human touch, not to the first demo response. Platforms requiring engineering resources to change a workflow scored lower than those where a service desk manager can edit behaviour.

Escalation quality and audit logging. Every autonomous action needs an immutable log entry naming the agent, the request, the policy applied, and the approver. We tested whether escalations carry full context to the fulfiller and whether the audit trail satisfies a SOX or ISO evidence request without a custom export.

The 9 Best AI Service Desk Tools in 2026

Nine platforms make the 2026 list, ranked by fit for internal IT teams that want autonomous resolution over an existing ITSM system of record. Forethought has been removed as a standalone entry: Zendesk acquired it in March 2026 and folded its products into the Zendesk Resolution Platform. Pricing is drawn from live vendor pages where public, and explicitly labelled where it is not.

1. Fini

Fini is a resolution layer that installs above your existing service desk rather than replacing it. For internal IT, that means Jira Service Management, Freshservice, ServiceNow, and Slack or Teams stay exactly where they are while Fini handles intake, classification, retrieval, and action. The agent resolves 90% of incoming requests with 99% accuracy, grounded strictly in content your team has approved.

The architectural choice that matters for IT buyers is grounding. Fini generates answers only from approved runbooks, KB articles, and resolved-ticket history, which removes the failure mode where an agent invents an access policy that does not exist. Actions in connected systems run through explicit permission scopes, so a password reset can be autonomous while a security group addition requires an approver. Every action writes an audit log entry naming the request, the policy applied, and the outcome.

Deployment is live in 30 days, including implementation, which is bundled into every plan rather than sold as a separate professional services line. Voice is included on Scale and Enterprise, which matters for the IT teams still running a phone line for lockouts and after-hours incidents. Compliance covers SOC 2 Type II, ISO 27001, HIPAA-compliant deployments with BAA eligibility, GDPR, and CCPA, with data residency options for EU and UK buyers.

Pricing is resolution-based with no per-seat fees. Growth is $3,600/mo, or $3,000/mo billed yearly at $36,000/yr, with 2,000 resolutions included and $0.89 per resolution beyond that. Scale is $9,000/mo, or $7,500/mo billed yearly at $90,000/yr, with 8,000 resolutions and 500 answered voice calls included, and $0.69 per resolution beyond that. Enterprise is custom pricing, contact Fini. Annual billing gives two months free, and unused allowance rolls forward one month.

Key strengths:

• 90% resolution at 99% accuracy, measured on completed requests rather than deflected sessions.

• Layers over your ITSM, so the CMDB, SLA clocks, and approval workflows you already built stay intact.

• Live in 30 days with implementation bundled into the plan price, not billed separately.

• Action-level permissions so autonomous resets and gated access grants run under different policies.

• Full compliance strip: SOC 2 Type II, ISO 27001, HIPAA-compliant, BAA-eligible, GDPR, CCPA.

• Voice included on Scale and above, priced per answered call from $0.89 down to $0.35 at volume.

Best for: IT organisations running an established ITSM platform that want autonomous resolution of identity, access, and software requests without migrating the system of record.

2. ServiceNow (Now Assist)

ServiceNow remains the default system of record for large enterprise IT, and Now Assist is its generative AI layer for incident summarisation, virtual agent conversations, and workflow generation inside the same platform. The advantage is proximity to data: the CMDB, service catalogue, and change calendar all live in the same schema the agent reasons over. For organisations already licensed on the platform, there is no integration project.

The cost structure is the friction point. ServiceNow prices per fulfiller with a platform commitment underneath, and AI capabilities attach to higher SKUs, so the marginal cost of adding autonomous resolution is entangled with a broader licence negotiation. ServiceNow does not publicly state list pricing for Now Assist. The competing #3-ranking analysis for this keyword anchors its ITSM ROI case partly on a 75% reduction in service ticket volume attributed to a ServiceNow and Ernst & Young engagement (eesel AI), which is a single-engagement figure rather than a platform benchmark.

Implementation is the other consideration. Now Assist configuration typically runs through the same governance process as any ServiceNow change, which means change advisory board review, sandbox testing, and a partner implementation team for anything non-trivial.

Pros:

• Native access to CMDB, service catalogue, and change records with no integration layer

• Deep ITIL process coverage including problem and change management

• Enterprise governance, audit logging, and role-based access already established

• Single vendor for system of record and AI layer simplifies security review

Cons:

• Does not publicly state pricing; AI capability is bundled into higher-tier negotiations

• Per-fulfiller licensing scales with headcount rather than with resolutions delivered

• Configuration changes usually require partner or platform-team involvement

• Implementation timelines measured in quarters rather than weeks for most enterprises

Best for: Large enterprises already standardised on ServiceNow that want AI inside the platform of record and can absorb a longer implementation cycle.

3. Zendesk (Resolution Platform, including Forethought)

Zendesk repositioned from ticketing to what it now calls the Resolution Platform, and its 2026 story is defined by acquisition. On 11 March 2026, Zendesk announced a definitive agreement to acquire agentic customer service startup Forethought, reported as its largest acquisition in two decades, to add self-improving AI agents to the platform (TechCrunch). Zendesk announced completion shortly after, with Forethought's Solve, Triage, Assist, Discover, and Agent QA products folding into the platform (Zendesk).

The commercial model changed too. Zendesk moved AI agents from a per-seat add-on to outcome-based billing per automated resolution, expanded at its Relate conference in May 2026, after introducing automatic overage billing above committed volume in January 2026. Third-party 2026 analyses report Suite plans spanning $19 to $115 per agent per month with Suite Team at $55 as of July 2026, a Copilot add-on at $50 per agent per month, a Workforce Engagement bundle at $50, and Contact Center voice at $83 (Richpanel). Zendesk publishes no per-resolution rate; treat all of these as third-party reported and verify on the official US pricing page.

For internal IT specifically, Zendesk is a competent employee-service deployment but was built for customer support. Teams running ITIL change management or a populated CMDB will find the object model thinner than a dedicated ITSM platform.

Pros:

• Outcome-based AI billing means cost tracks resolutions rather than fulfiller headcount

• Forethought's triage and agent QA capabilities now available inside one platform

• Large integration marketplace covering identity, commerce, and analytics tools

• Dual verification of billed resolutions gives buyers a check on the invoice

Cons:

• No published per-resolution rate, so budgeting requires a sales conversation

• Seat costs stack under the AI meter: Suite plus Copilot plus voice adds up per fulfiller

• ITIL depth is limited compared with dedicated ITSM platforms

• Post-acquisition product consolidation creates near-term roadmap uncertainty

Best for: Organisations that already run Zendesk for external support and want one vendor covering both employee and customer queues.

4. Freshservice (Freddy AI)

Freshservice is Freshworks' ITSM product, positioned as the mid-market alternative to ServiceNow with faster setup and a lighter administrative burden. Freddy AI provides the conversational layer: employee-facing virtual agent in Slack and Teams, ticket summarisation, and suggested resolutions drawn from the knowledge base. The service catalogue and asset management modules are genuinely usable out of the box, which is the main reason mid-market IT teams pick it.

Freshworks blocks automated verification of its pages, so pricing and performance figures here should be confirmed directly with the vendor. The competing #3-ranking page attributes a 65.7% ticket deflection figure to Freshworks (eesel AI); note that this is deflection, not verified resolution, and the two are not interchangeable per the audit checklist above.

Freshservice licenses per agent with AI features attached to higher tiers, which puts it in the same structural bucket as ServiceNow: cost scales with staff rather than with automated outcomes. For teams under 50 fulfillers that is usually cheaper than a resolution-metered contract.

Pros:

• Fast ITSM setup relative to enterprise platforms, often weeks rather than quarters

• Native asset management and service catalogue included rather than bolted on

• Slack and Teams virtual agent covers employees where they already work

• Mid-market pricing accessible to IT teams without enterprise procurement

Cons:

• Pricing and performance figures could not be verified directly; vendor blocks automated fetching

• Per-agent licensing does not reward automation with lower cost

• Advanced AI capability sits on higher tiers, raising the effective entry price

• Deflection-framed metrics require independent validation against reopened-ticket rates

Best for: Mid-market IT teams that want a complete ITSM platform and a competent virtual agent from a single vendor without an enterprise implementation cycle.

5. Jira Service Management (Atlassian Intelligence)

Jira Service Management is the ITSM choice for organisations already running Jira and Confluence for engineering work. Atlassian Intelligence supplies the AI layer: a virtual service agent in Slack and Teams, request classification, and answer generation grounded in Confluence spaces. The gravitational pull is knowledge: if your runbooks already live in Confluence, retrieval quality starts higher than a vendor ingesting from scratch.

Its strongest fit is DevOps-adjacent IT, where incident tickets and engineering work items need to live in the same system. Change management, on-call routing through Opsgenie lineage, and post-incident reviews connect naturally. Atlassian does not publicly state a separate price for Atlassian Intelligence in the context relevant to this comparison; it is packaged into plan tiers, so verify with Atlassian directly.

The limitation for classic IT service desks is action depth. The virtual agent excels at answering from Confluence and creating well-formed tickets, but executing writes into identity and endpoint systems generally requires automation rules or a third-party resolution layer sitting above it.

Pros:

• Retrieval grounded in Confluence, where most engineering-led IT teams already document runbooks

• Tight coupling between service tickets and engineering work items in one system

• Strong incident and change workflows for DevOps-adjacent IT organisations

• Slack and Teams virtual agent deployment is straightforward for existing Atlassian tenants

Cons:

• Does not publicly state standalone AI pricing; capability is bundled into plan tiers

• Action execution in identity and MDM systems typically needs automation rules or a layer above

• Knowledge quality is only as good as Confluence hygiene, which varies widely

• Less compelling for IT organisations with no existing Atlassian footprint

Best for: Engineering-heavy organisations already on Jira and Confluence that want service requests and engineering work managed in one system.

6. Moveworks

Moveworks was built specifically for the employee-facing use case, which distinguishes it from every CX platform retrofitted for internal IT. It sits in Slack, Teams, or a web portal and resolves IT, HR, and facilities requests by executing against connected systems: identity providers, MDM, HRIS, and the ITSM ticket object. Access requests, software provisioning, and password flows are the core competency rather than an add-on.

The integration model is the product. Moveworks maintains connectors into the enterprise systems where the actions live, and the value proposition rests on how many of those connectors are live in your specific stack. Buyers should map their top 20 request types against available connectors during evaluation, because coverage gaps are what determine the real resolution ceiling.

Moveworks does not publicly state pricing. Reporting on the category consistently describes employee-headcount-based contracting for this class of vendor, which means cost scales with company size rather than ticket volume. For a 5,000-employee company with a low ticket rate per employee, that model can be less favourable than resolution-metered pricing.

Pros:

• Purpose-built for employee support rather than adapted from a customer support product

• Deep connector library into identity, MDM, HRIS, and ITSM systems for real action execution

• Covers IT, HR, and facilities from a single conversational surface in Slack or Teams

• Strong intent classification tuned specifically to internal request language

Cons:

• Does not publicly state pricing, making budget modelling impossible before a sales cycle

• Headcount-based contracting decouples cost from actual ticket volume

• Connector coverage varies by stack; gaps directly cap the resolution rate

• Enterprise-shaped deployment is heavy for IT teams under a few thousand employees

Best for: Enterprises with several thousand employees seeking a single conversational front door across IT, HR, and facilities.

7. Aisera

Aisera positions as an agentic AI platform spanning IT service management, HR, customer service, and operations, with an emphasis on domain-specific language models per function. For internal IT it offers conversational intake, ticket auto-resolution, workflow execution, and analytics on request patterns. Its breadth is genuine: the same platform is sold into IT and non-IT service functions.

The evaluation question for Aisera is depth versus breadth. A platform covering four service domains spreads engineering effort across all of them, so buyers should test the specific IT request types they care about rather than accepting a general resolution percentage. Ask for a demo that resets a password, then grants an application entitlement through an approval workflow, using your own identity provider.

Aisera does not publicly state pricing. Contracting is enterprise-shaped and negotiated, typically with an annual platform commitment. Compliance posture and audit logging depth should be verified in the security review rather than assumed from the marketing site.

Pros:

• Single agentic platform spanning IT, HR, and other internal service functions

• Domain-tuned models rather than one generic assistant across all use cases

• Workflow execution and analytics included rather than sold as separate modules

• Enterprise integration breadth across common ITSM and identity systems

Cons:

• Does not publicly state pricing; requires a full enterprise sales cycle to budget

• Breadth across four domains raises the question of depth in any single one

• Vendor-reported resolution percentages need independent validation against your ticket mix

• Implementation is consultative rather than self-serve

Best for: Enterprises consolidating IT, HR, and other internal service functions onto one AI platform with a single vendor relationship.

8. Tidio (Lyro AI)

Tidio is the outlier on this list: a small-business conversational platform whose Lyro AI agent handles repetitive requests autonomously. Its natural home is customer-facing ecommerce support, but small IT teams and internal ops functions do deploy it as a lightweight front door when the alternative is nothing. Setup is measured in hours, not weeks.

Tidio publishes full public pricing, which corrects a common misconception that it is quote-only. As of 6 August 2026, Starter is $24.17/mo for 100 billable conversations, Growth starts at $49.17/mo from 250 billable conversations, Plus starts at $300/mo plus monthly usage, and Premium is contact-for-pricing. Lyro AI starts at $32.50/mo from 50 Lyro AI conversations, with 50 free lifetime Lyro conversations and a 7-day trial (Tidio).

The limitation for internal IT is action depth. Lyro answers from a knowledge base and completes simple flows, but it does not execute privileged writes into Entra, Intune, or an ITSM change object. Treat it as a knowledge front end, not a resolution layer.

Pros:

• Fully published pricing with a genuine low entry point at $24.17/mo

• Free tier of 50 lifetime Lyro conversations plus a 7-day trial with no card required

• Deployment in hours with minimal configuration overhead

• Conversation-metered billing keeps costs proportional to actual usage

Cons:

• Built for customer conversations, not employee IT requests or ITIL objects

• No meaningful write actions into identity, MDM, or ITSM systems

• Conversation counting can escalate quickly at volume beyond the entry tiers

• Limited compliance and audit tooling for regulated internal deployments

Best for: Small teams wanting a low-cost conversational knowledge front end, with no expectation of autonomous action in enterprise systems.

9. eesel AI

eesel AI connects to Zendesk, Slack, Google Drive, Freshdesk, Confluence, and Notion, acting as an AI layer across both support queues and internal knowledge. Its ITSM-adjacent positioning is credible: the company publishes its own service desk research and comparison content, and its knowledge ingestion across scattered internal apps is genuinely useful for IT teams whose documentation lives in five places.

Pricing is fully published and purely usage-based, with no seats. As of 6 August 2026, light tasks such as dashboard questions and lookups are free, regular tasks such as support tickets and chat sessions are $0.40 each, and heavy tasks are $4.00 each. There is no platform or seat fee below Enterprise, a default $250 monthly usage cap, a $50 free trial credit, 25% off for annual prepay, and an Enterprise tier at $1,000/month plus usage (eesel AI).

For internal IT, the strength is retrieval across fragmented knowledge and the weakness is the same as Tidio's: limited privileged action execution. It answers well and drafts well. It does not reset MFA enrollment in your identity provider under an approval policy.

Pros:

• Fully transparent per-task pricing with no seat or platform fee below Enterprise

• $0.40 per regular task is among the lowest published unit costs in the category

• Ingests knowledge from Slack, Drive, Confluence, Notion, and multiple help desks at once

• Default $250 monthly usage cap prevents runaway spend during a pilot

Cons:

• Task-based counting makes TCO harder to forecast than a fixed resolution allowance

• Limited privileged action execution in identity, MDM, and ITSM systems

• Value depends heavily on the structure and quality of connected source systems

• Enterprise tier adds a $1,000/month platform fee on top of usage

Best for: IT and support teams with knowledge scattered across many apps that want cheap, transparent per-task pricing and mainly need answers rather than actions.

Platform Summary Table

The table below compares published facts only. Where a vendor does not publish pricing or certifications, the cell says so rather than repeating a third-party estimate as fact. Accuracy figures are included only where the vendor states them.

Forethought no longer appears as an independent vendor. Zendesk acquired it in March 2026 and its products are being merged into the Zendesk Resolution Platform.

What an AI Service Desk Costs in 2026

Six pricing units are in circulation, and comparing across them is the hardest part of the evaluation. The unit determines whether automation lowers your bill or leaves it flat. Per-agent and per-fulfiller models charge for staff you are trying not to hire; per-resolution and per-task models charge only when work gets done.

Worked comparison at three volumes, using published rates only. Intercom's Fin bills $0.99 per outcome on top of seat tiers of $29, $85, or $132 per seat per month (Intercom). eesel bills $0.40 per regular task with no seat fee (eesel AI). Fini's Growth plan includes 2,000 resolutions at $3,600/mo with $0.89 beyond; Scale includes 8,000 at $9,000/mo with $0.69 beyond.

The table is not apples-to-apples, and that is the point. eesel's task is a chat session or ticket touch, not a verified resolution. Intercom's outcome can be triggered by the absence of a follow-up. Fini's allowance covers verified resolutions with implementation and platform bundled, and unused allowance rolls forward one month. Normalise on your own definition before comparing any two numbers.

Zendesk cannot be placed in this table honestly. It bills AI agents per automated resolution but publishes no rate, and third-party analyses report mid-market deals landing roughly in the $1 to $2 per resolution range (Richpanel). Do not budget against a number the vendor has not published.

Consolidation Watch: Who Owns Whom

Category consolidation is now a procurement risk, not just industry news. A vendor acquired mid-contract can reprice, deprecate, or merge the product you bought, and a vendor sitting on a very high valuation carries pressure to raise prices toward that mark. Three 2026 events should shape how you write your contract.

Zendesk acquired Forethought in March 2026, its largest acquisition in two decades, folding Solve, Triage, Assist, Discover, and Agent QA into the Resolution Platform (TechCrunch). Anyone who signed a Forethought contract in 2025 is now a Zendesk customer with a different roadmap.

Decagon was valued at $4.5 billion in January 2026 after a $250M round led by Coatue and Index Ventures, completing an employee tender at the same valuation in March (Bloomberg). Sierra raised $950M at a $15.8 billion post-money valuation on 4 May 2026, roughly tripling its mark from 18 months earlier and signalling expansion beyond customer support into broader enterprise agents (TechCrunch). Neither publishes pricing.

Three clauses to add to any 2026 contract: price protection for the full term including after a change of control, a data export guarantee in a documented format, and a product-continuity commitment naming the specific modules you are buying.

Compliance Duties for Employee-Facing AI Agents

EU AI Act Article 50 transparency obligations became enforceable on 2 August 2026 and apply to any AI system that interacts directly with people, including employee-facing service desk agents. You must ensure the person knows they are dealing with an AI unless it is obvious from context, and synthetic audio, image, video, or text must be marked in a machine-readable, detectable format (Goodwin). SOC 2 and GDPR alone no longer constitute a complete compliance criterion for this category.

The high-risk timeline moved, but the transparency rules did not. The Digital Omnibus on AI, signed 8 July 2026, defers stand-alone Annex III high-risk obligations to 2 December 2027 and Annex I embedded systems to 2 August 2028, while leaving the August 2026 transparency rules on schedule (Gibson Dunn). Employee-facing agents that touch HR-adjacent decisions should still plan against the 2027 date.

RFP questions worth asking every shortlisted vendor:

• Does the agent display a persistent AI disclosure in Slack, Teams, and the portal, and can it be enforced by policy?

• Are synthetic voice outputs marked in a machine-readable format?

• Where is data processed and stored, and what data residency options exist for EU and UK employees?

• Which certifications are current: SOC 2 Type II, ISO 27001, and where relevant iso 42001 for AI management systems?

• For health, finance, or public sector deployments, what does the ai compliance evidence pack contain?

• Has the model been through ai red teaming, and can you share the summary findings?

• For EU financial entities, how does the vendor support dora compliance obligations on ICT third-party risk?

• Does every autonomous action produce an immutable audit log entry naming the policy applied?

Why Rollouts Stall After Password Resets

The most common failure pattern is a pilot that resolves password resets beautifully and then plateaus at 15 to 25% of ticket volume. Password resets are the easy case: a single system, a clear verification step, a reversible action. Everything after that requires integration work, permission design, and someone owning the agent's behaviour.

Five reasons deployments stop growing, and what fixes each:

The honest expectation curve: 15 to 25% autonomous resolution in the first quarter, concentrated in identity and lookups. 40 to 60% by the end of year one if connectors and approvals are built out. Higher only where the agent has genuine write access across identity, MDM, and the ITSM object, and where knowledge is maintained rather than archived.

Budget for the integration work, not just the licence. The vendor demo resolves a password because your identity provider makes that easy. Your actual queue is 40% access requests with approval chains that live in a spreadsheet.

How to Choose the Right Platform

Choosing an AI service desk in 2026 comes down to matching the pricing unit to your automation goal and the action depth to your top request types. Everything else, including model quality, converges quickly across serious vendors. Work through these five steps in order.

1. Pull your last 90 days of tickets and rank the top 20 request types by volume. This list, not a vendor feature matrix, defines what "good" means for your deployment. If access requests are 30% of your volume, a platform that only answers questions cannot solve your problem regardless of its accuracy claim.

2. Map each request type to the system the action lives in. Password reset means Entra or Okta. Software install means Intune or Jamf. License provisioning means the SaaS admin API. A vendor's connector coverage against that specific list is the ceiling on your resolution rate.

3. Force every vendor to state its resolution definition in writing. Run it through the audit checklist above. Where the vendor bills on resolutions, this definition is a pricing term, so it belongs in the contract rather than in a slide.

4. Model TCO on your own volume across all six pricing units. Convert per-agent, per-fulfiller, per-employee, per-session, per-task, and per-resolution quotes into cost per verified resolution at your actual monthly ticket count. Include implementation, which some vendors bundle and others bill separately.

5. Run a bounded pilot on three request types, not twenty. Pick one identity request, one access request with an approval chain, and one troubleshooting flow. Measure autonomous resolution net of 7-day reopens, fulfiller time saved, and employee satisfaction. Expand only after all three clear your bar.

6. Check the compliance and continuity clauses before signing. Article 50 disclosure support, data residency, audit logging, price protection through change of control, and a data export guarantee. Given the acquisition activity in this category, the continuity clauses matter more in 2026 than they did two years ago.

Implementation Checklist

A structured rollout across four phases keeps the deployment from plateauing at password resets. Budget roughly two weeks for pre-purchase analysis, four weeks for evaluation, thirty days for deployment, and continuous review after launch.

Pre-purchase

• Export 90 days of ticket data and rank the top 20 request types by volume and handle time

• Map each request type to the target system and required permission scope

• Document current cost per ticket including fulfiller time, tooling, and escalation overhead

• Confirm which ITSM platform remains the system of record post-deployment

Evaluation

• Obtain each vendor's written resolution definition and run the audit checklist against it

• Demo one write action into your own identity provider, not a vendor sandbox

• Normalise all quotes to cost per verified resolution at your actual monthly volume

• Collect SOC 2 Type II and ISO 27001 reports, data residency terms, and Article 50 disclosure capability

• Add price protection, change-of-control, and data export clauses to the contract redlines

Deployment

• Ground the agent on approved runbooks and KB articles; archive anything more than two reorgs old

• Configure action gates: autonomous for reversible actions, approval-gated for entitlement changes

• Define escalation payload contents and confirm fulfillers receive full context

• Enable AI disclosure in every channel: Slack, Teams, portal, and voice

• Verify audit log entries capture agent, request, policy applied, and outcome

Post-launch

• Report autonomous resolution net of 7-day reopened tickets from week one

• Review misclassified intents weekly for the first month, monthly thereafter

• Expand scope by one request type per sprint against the ranked top 20 list

• Name a permanent owner for agent scope and knowledge freshness

Final Verdict

For internal IT teams that already run a system of record and want autonomous resolution rather than a better chatbot, Fini leads this comparison: 90% resolution at 99% accuracy, live in 30 days with implementation bundled, and resolution-based pricing at $3,600/mo for Growth and $9,000/mo for Scale with no per-seat fees. ServiceNow and Jira Service Management remain the right answer when the AI must live inside the platform of record. Moveworks and Aisera are credible enterprise alternatives if you can absorb an undisclosed-price sales cycle.

The decision usually turns on two things. Whether your pricing unit rewards automation or penalises headcount, and whether the agent can write into the identity and endpoint systems where your top 20 request types actually resolve.

If your queue is dominated by password resets, MFA lockouts, access requests, and license provisioning, and you want those resolved on top of Jira Service Management, Freshservice, or ServiceNow without migrating anything, book a walkthrough with the Fini team and bring your ranked top 20 request types to the call.