Home · Solutions · Other solutions
Solution · Other solutionsThe fix is already written down. An agent in Teams runs it; analysts see only what needs them.
The service desk request that never becomes a ticket
Employees describe the problem in Microsoft Teams; a UiPath conversational agent answers from your knowledge base or runs the approved fix by robot, and opens a ticket only when a person is needed.
Executive summary
Four IT tickets in ten follow a known procedure, and every one of them still queues for an analyst.
The front door changes first.
Desk capacity returns to engineering work, because the repetitive four tickets in ten are resolved in the chat rather than in the queue.
ServiceNow or Jira Service Management; VPN, print and file services; the software catalogue and device state in Microsoft Intune
Business problem
IT service desk
The desk is sized for the peaks, staffed for the average and judged on the exceptions. The repetitive requests have a documented answer somewhere, usually a SharePoint page nobody finds, but the path from "I have a problem" to "someone applied the known fix" still runs through a human queue. Employees post in the IT channel in Teams or email the analyst who helped them last time; the desk creates the ticket afterwards, by hand.
Analysts switch between a certificate renewal and a mapped-drive question forty times a day. Knowledge articles age because nobody owns them, so the procedure for clearing a print queue lives in the habits of the one person who "knows the plant printers". First response is measured in hours, and the employee has usually found a workaround before the first question comes back.
The pattern persists for a reason. Earlier chatbots could only point to articles, not perform the fix, and the deterministic automations that could perform it were never wired to a conversation. So desk cost grows with headcount, and the best analysts are spent on work a written procedure could execute.
How it works today
One repetitive request takes this route today, whichever ticketing tool sits underneath.
- PersonAn employee notices the VPN profile has stopped working, searches the intranet, gives up, and posts in the IT channel in Teams
- PersonAn analyst creates the ticket in ServiceNow by hand, pasting the message into the description
- WaitingThe ticket sits in the unassigned queue for hours; the analyst who picks it up first asks for the device name and the error text
- WaitingThe employee has moved on to a workaround and replies the next morning
- SystemThe analyst applies the fix from the runbook on SharePoint: a re-issued profile, a cleared print queue, a mapped drive
- Risk of errorThe ticket closes without a word to the employee, who re-opens it or asks a colleague; nobody counts the second contact
- Risk of errorMeanwhile the controller's broken SAP GUI and an expiring certificate wait their turn in the same queue
Why the current process costs more than it appears
The cost grows where nobody is looking.
- Analyst minutes per ticket are the visible cost; the waiting around them is not. A sales engineer without VPN dials in from a personal hotspot, a controller emails a spreadsheet to a private address because the shared drive will not map, and the desk never sees either exposure.
- Repeat contacts consume a second and a third analyst touch that never appear in the ticket count, because the re-opened ticket keeps its number.
- Knowledge concentrates in people. The analyst who knows the plant printers becomes the only route to a working printer, and his week of leave becomes a backlog.
- Risk waits behind trivia. The odd login pattern, the expiring certificate and the failing disk are exactly the tickets deferred behind eleven password questions.
Cost of inaction
Headcount and system count both grow, and ticket volume follows them: two depots, a new warehouse release, another round of MFA changes. With the same nine analysts the queue lengthens, first response stretches, and the desk quietly relaxes its SLA targets to match. Employees learn that the desk is slow and route around it, which hides demand and breeds shadow fixes, while the 480 hours a month they spend waiting stay off every report.
The quieter loss is knowledge. The analyst who knows the plant printers leaves for an engineering role, and the runbook that existed only in his habits leaves with him. No single day makes this a crisis; the desk simply costs more each quarter for the same service, and the tickets that carry real risk keep waiting behind the trivial ones.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
A logistics group with 3,200 employees in four countries; a central IT desk of nine analysts on ServiceNow; Microsoft 365 and a warehouse management system as the main platforms; two depots opening next year with no desk hires planned.
Around 3,600 tickets a month, mostly by email and a Teams channel; the desk's own categorisation puts four in ten in categories with a documented procedure: VPN and Wi-Fi profiles, Outlook and Teams client issues, printer queues, drive mappings, approved-software installs and "where do I find" questions.
Every request becomes a ticket typed in by an analyst, waits in the unassigned queue, and is resolved from a runbook on SharePoint after a round of clarifying questions. First response takes hours.
Twelve minutes of analyst handling per repetitive ticket, across 1,440 such tickets a month; twenty minutes of waiting and re-contact for each employee, which no report records.
A UiPath conversational agent in Microsoft Teams answers "how do I" questions from the SharePoint knowledge base and runs approved procedures through UiPath Robots; anything else becomes a ServiceNow ticket that already contains the summary, the diagnostics and the steps tried.
In the modelled case the four repetitive tickets in ten are closed in the chat within minutes, analysts open tickets that are halfway done, and the two new depots are served by the same nine people. Illustrative figures, not a client result.
Proposed solution
The front door changes first. Employees describe the problem in a chat in Microsoft Teams: the agent, a UiPath conversational agent published as a Teams app, answers "how do I" questions from a Context Grounding index over the client's own SharePoint knowledge base, with a link to the source article, and recognises requests that match a procedure the desk has approved. For those it calls the matching UiPath Robot through Orchestrator: a re-issued VPN profile, a cleared print queue, a mapped drive, an application from the catalogue.
Everything else becomes a ticket the agent opens itself, in ServiceNow or Jira Service Management through UiPath Integration Service, carrying the conversation summary, the diagnostics gathered and the steps already tried; the analyst's first touch is a resolution rather than a question. Anything that needs a judgement, a privilege or a security decision goes to a person as an Action Center task, completed without leaving Teams.
The fix library grows from the client's own runbooks, one approved procedure at a time. The language model chooses among released tools; it does not write commands. Every question the agent could not answer is logged as a knowledge gap for the content owner, which is how the knowledge base stops ageing.
UiPath conversational agent as a Microsoft Teams app; Context Grounding index over SharePoint with inherited permissions; agent guardrails and the AI Trust Layer; Orchestrator queues, credential store and audit; Action Center tasks completed in Teams; Microsoft Entra ID for the agent identity
The procedure catalogue as robot workflows with exception handling; the agent's tools, prompts and guardrails; the ticket template with summary and diagnostics; the knowledge-gap report
ServiceNow or Jira Service Management through UiPath Integration Service connectors; user lookups through the Integration Service connector still named 'Microsoft Azure Active Directory'; device state read from Microsoft Intune where a fix depends on it
How the automated process works
- PersonThe employee describes the problem to the agent in Microsoft Teams, in their own words, from the window already open
- AutomationThe agent classifies the request, asks the one or two questions the runbook needs, and checks the Context Grounding index
- AutomationA "how do I" question gets an answer with a link to the source article; a request matching an approved procedure is queued for a robot in Orchestrator
- SystemThe robot runs the fix, a re-issued VPN profile, a cleared print queue or a mapped drive, and the agent confirms the result in the chat
- SystemAnything outside the catalogue becomes a ServiceNow ticket with the summary, the diagnostics and the steps tried
- PersonRequests touching privileges, data access or a security signal reach an analyst as an Action Center task in Teams; the analyst decides
- AutomationUnanswered questions are logged for the knowledge owner; a weekly report lists which procedures ran and where the agent hesitated
Human-in-the-loop model
Automation handles
- Understanding the request and asking the clarifying questions the runbook needs
- Answering from the approved knowledge base, with a link to the source article
- Running approved fixes through robots: profiles, queues, drives, catalogue installs
- Creating the ticket with summary, diagnostics and attempted steps, and reporting the questions it could not answer
People decide
- Every request touching privileges, data access or security signals
- Diagnosis of anything outside the approved catalogue
- Approval of each new procedure before the agent may execute it
- A weekly review of a sample of conversations for quality and tone
Before and after
Systems and integrations
The stack is deliberately short: one engine, one execution layer, one place where a person decides.
Inputs
- the agent chat in Microsoft Teams
- the SharePoint knowledge base
- the ServiceNow ticket history used to select the first procedures
Automation layer
- UiPath Agents
- UiPath Context Grounding
- UiPath Orchestrator
- UiPath Robots
- UiPath Integration Service
- UiPath AI Trust Layer
Target systems
- ServiceNow or Jira Service Management
- VPN, print and file services
- the software catalogue and device state in Microsoft Intune
- Microsoft Entra ID
Human touchpoints: the agent in Microsoft Teams; Action Center tasks in Teams; the weekly conversation review; the knowledge-gap list for content owners
Technologies used
the front door: understands the request, answers from the index, chooses among approved tools, escalates
Apermission-inherited index over the SharePoint knowledge base; answers carry a source link
Arun the approved procedures; queues, service-account credentials, retries, audit trail
Acreates and updates the ticket in ServiceNow or Jira Service Management; posts to Teams; user lookups in Microsoft Entra ID
Aescalation tasks for analysts, completed inside Microsoft Teams
Amodel allow-list, PII masking in prompts, GenAI audit logs, EU region routing
Awhere employees ask and get the fix, and where analysts receive escalations
Aidentity of the agent app, group membership, single sign-on
AIllustrative economic model
Numbers you can check against your own data.
Four in ten is the figure to challenge before anything else: 3,600 tickets a month × 0.4 gives the 1,440 the calculator starts from, and it comes from this illustrative desk's own categorisation, not from a client. Twelve minutes is analyst handling per such ticket; €47 an hour is a fully loaded IT specialist cost in Central Europe. The employee's twenty minutes of waiting and re-contact per ticket, 480 hours a month, is deliberately left unpriced, and licensing, implementation and catalogue curation sit outside the model.
Run the numbers on your data
An illustrative estimate from your own inputs. It models released capacity; it is not a promise of savings.
Business benefits
- Desk capacity returns to engineering work, because the repetitive four tickets in ten are resolved in the chat rather than in the queue
- Employees get the fix in minutes from the window already open, which removes the hotspot and private-email workarounds
- Tickets that reach analysts arrive with a summary and diagnostics, so the first touch is a resolution, not a question
- Knowledge improves on its own rhythm: every unanswered question becomes a work item for a named content owner
- New sites are absorbed without proportional hiring, since the agent runs the same procedures in every depot
The management view
- Demand becomes measurable: which procedures run, how often, where the agent hesitates and which articles are missing, per site and per month
- Staffing for the two new depots is decided from the measured deterministic share rather than from an estimate
- Every automated action is traceable in Orchestrator and every conversation in the agent logs, an audit trail nobody on the desk has to write
- Ownership becomes explicit: someone owns the procedure catalogue and someone owns the knowledge
Board-level KPIs
Security and governance
An auditor should be able to reconstruct every decision.
- The agent has its own identity in Microsoft Entra ID with the minimum permissions its tools need; robots run under dedicated service accounts per target system, with secrets held in Orchestrator's credential store or Azure Key Vault, never inside a workflow
- Each tool is an approved, versioned automation: the language model chooses among actions IT has released and cannot improvise a command, and a procedure enters the catalogue only after the desk lead signs it off
- The AI Trust Layer enforces the model allow-list, masks personal data in prompts and keeps GenAI audit logs; every robot action is logged in Orchestrator against the conversation that triggered it, and privileged requests never run autonomously
- Data stays in the EU: UiPath Automation Cloud in the EU region, Microsoft 365 within the EU Data Boundary; the knowledge index inherits SharePoint permissions, so the agent cannot show an article the employee could not open
Why now
The trigger is rarely technology: a hiring freeze, a merger, or here two depots opening without desk hires. The desk absorbs the new users either with a procedure or with a longer queue, and the modelled €13,536 a month of analyst time on repetitive tickets is the price of the second option
Conversational agents can be published directly into Microsoft Teams as governed apps, generally available since April 2026, with escalation into Action Center, so conversation and execution finally sit in one place
Retrieval over the client's own SharePoint content answers with sources rather than guesses; the deterministic fixes have been buildable for years, and what was missing was a front door employees would use
Relevant executive roles
Desk cost and response time are the most visible IT metrics in the business, and this changes both without a headcount request
The desk gets its engineers back, and a governed procedure catalogue replaces tribal knowledge
The same chat-to-fix pattern extends to HR and finance requests once IT has proved it
Operations staff lose fewer hours to IT friction, and new sites open without a support bottleneck
Common questions and objections
They trust it when it fixes the drive in two minutes and names a person when it cannot. The agent always offers a human route, and abandonment is measured in the pilot rather than assumed.
That is the usual starting point. The agent's list of unanswered questions is the best clean-up list a content owner ever receives, and the first catalogue can start with procedures that need no article at all.
It chooses among approved tools and does not write commands. Anything privileged goes to a person as an Action Center task, and every action is logged against the conversation that triggered it.
When this is not the right solution
- Ticket volume below a few hundred a month; a Teams form plus robots is enough, and cheaper
- No documented procedures at all; we start with process analysis, not with an agent
- The ticketing tool is due for replacement within months, or security policy forbids any LLM use; in the second case deterministic self-service still applies
A question for the next management meeting
Two depots open next year with no desk hires planned: which of their VPN, printer and drive requests will a written procedure close, and which will simply lengthen the queue?
Implementation approach
A scope without ambiguity, before anything is signed.
We deliver
- Analysis of one quarter of ticket data, clustered by procedure, to agree the first ten to fifteen with volume and low risk
- Design of the chat-to-fix flow and the escalation path, agreed with the desk lead
- The agent: tools, guardrails, the Context Grounding index, publication into Teams for the pilot group
- Robot workflows for the approved procedures, with exception handling and a desk review of every action in the first weeks
- Ticketing integration, the ticket template, analyst training, the knowledge-gap report, then support through our AMS service
We need from you
- A ticket export for the last quarter, with categories and resolution notes
- The runbooks the desk actually uses, and a named content owner for the knowledge base
- A product owner from the desk who approves each procedure before release
- A security review of the tool list and a service account per target system
Stages
Discovery
Cluster the last quarter of tickets by procedure; agree the first catalogue and the exclusions
Design
Chat-to-fix flow, escalation rules, guardrails, the ticket template, the knowledge scope
Build
Agent, index, robot workflows, ticketing integration, Teams publication for one department
Pilot
One department; every robot action reviewed by the desk; abandonment and repeat contacts measured
Scale
Procedure by procedure, each built, tested and approved by the desk lead before release
Optimisation
Weekly conversation review, knowledge-gap closure, catalogue growth from measured demand
Departmental. Effort follows the size of the catalogue, the state of the runbooks and the number of target systems a fix has to touch; the ticketing tool does not change it.
Six categories, four tickets in ten, and not one of them needs an engineer.
Send us one month of ticket categories from ServiceNow or Jira. We come back with the procedures an agent could run tomorrow, the ones that should stay with your analysts, and the reason for each.
Cluster one month of your ticketsThe neighbouring process usually has the same problem
The same policy questions reach the same three busy people, and the answer is already written somewhere.
View solution Customer serviceEvery customer email classified and routed on arrivalYour best agents spend their morning sorting mail instead of answering it.
View solution Other solutionsPassword reset in two minutes, without a ticketThe most frequent ticket on the desk still needs a phone call, three consoles and an identity check nobody records.
View solutionIndustries we deliver this in most oftenManufacturing & industryTransport & logisticsServices & ITShared services