Home · Solutions · Other solutions
Solution · Other solutionsOne sourced answer in Teams, filtered for country and contract, with a person one tap away
The same right answer to every HR policy question
Employees ask in Microsoft Teams, the agent answers only from the approved policy library and cites the paragraph and its version date, and anything policy cannot settle becomes an HR case.
Executive summary
Two advisors answer the same handbook questions all year, from different versions of the handbook.
We build an HR answers agent and publish it into Microsoft Teams, where employees already ask each other.
Advisor capacity is released because most lookups never enter the queue, and the queue is left for the cases that need judgement.
the HR case queue; the policy library and its document owners; the interaction log
Business problem
HR self-service
Policy and benefits questions are the largest category of HR mailbox traffic, and they are treated as work rather than as a knowledge problem. The answer already exists in a document. The cost is in finding it, interpreting it for the asker's country and contract type, and writing it out one more time.
Advisors handle these between payroll deadlines and case work, so response times drift from hours to days at month end. Consistency suffers because policy documents are updated on different cycles and old copies survive on desktops and in Teams channels. Nobody owns answer quality as a metric, so nobody notices when two entities are told two different things.
What makes it durable is that every individual question is small. No single lookup justifies a project, the HR portal stores policies where employees do not ask, and the advisor who knows the handbook becomes the bottleneck the week she is on holiday. Each acquisition and each new country multiplies policy variants without adding a single advisor.
How it works today
Below is the route a simple question takes when nothing has been built for it.
- PersonAn employee looks for the rule on the intranet, gives up, and asks a colleague or their manager instead
- PersonThe question reaches the HR shared mailbox and joins a first-come queue behind case work
- WaitingA lookup waits days behind a complex case, especially in the week before payroll closes
- PersonThe advisor hunts for the current policy version on SharePoint and interprets it for the entity and contract type
- PersonA reply is written from scratch and sent; nothing is recorded for reuse
- Risk of errorThe follow-up lands with a second advisor, who answers from a different document
- Risk of errorThe employee acts on a half-remembered rule, and the correction surfaces months later in payroll or a grievance
Why the current process costs more than it appears
Time that disappears before anyone measures it.
- Advisor minutes are the visible cost. The employee's own waiting time is roughly as large, and it never appears in an HR budget because it is spent in someone else's cost centre.
- Inconsistent answers between entities are a compliance exposure under collective agreements and works-council rules, and the exposure is invisible until somebody compares two replies.
- When a policy changes, HR cannot show who was told what and when, because the answers live in individual mailboxes rather than in a record.
- Employees who wait twice stop asking. They act on a colleague's version of the rule, which is how a benefit-eligibility assumption becomes a payroll correction or a grievance months later.
- One advisor who knows the handbook is a single point of failure and a training problem, and every new country adds variants that only she can reconcile.
Cost of inaction
Question volume follows headcount and policy complexity, and both usually rise. Add one country or one benefits provider and the policy variants multiply faster than advisors do, so the second row moves before anyone has planned for it. Over twelve months the two advisors answering lookups stay occupied by lookups, and the case backlog gets whatever is left at month end.
The quieter risk is that employees stop asking. That is not relief. Decisions about leave, expenses and benefits then get made on hearsay, and they surface months later as a payroll correction or a grievance. A wrong assumption about eligibility costs far more than the twelve minutes it would have taken to check, and nothing in the current process records it against this queue.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
An illustrative food-processing group, 3,400 employees across four European countries, with a nine-person HR shared-service team serving the plants and offices. Policies live in SharePoint in three languages; benefits are administered in SAP SuccessFactors.
Around 1,100 policy and benefits questions a month arrive in the HR mailbox and a Teams channel, and two advisors spend most of their day on them.
Each question is answered by hand from whichever handbook version the advisor opens, with no record of what was said and no way to compare answers between countries.
Version hunting and interpretation, repeated for every question, inside a queue where a two-minute lookup sits behind a two-week case.
An HR answers agent published in Microsoft Teams, grounded only on the approved policy library, filtering by country, entity and contract type, citing the paragraph it used, and turning anything it cannot settle into an HR case with one tap.
In the modelled case most lookups never enter the queue, answers carry a source and a version date, and policy owners receive a monthly list of the questions their documents do not answer. Every figure is a model; none of it is a client measurement.
Proposed solution
We build an HR answers agent and publish it into Microsoft Teams, where employees already ask each other. Its knowledge is an approved policy library on SharePoint, the handbook, benefits guides, leave rules and the expense policy, indexed with UiPath Context Grounding so that answers come from current, permission-controlled documents and nothing else. Each answer shows the paragraph it used and the version date of the document it came from, which is what makes it checkable by the person who receives it.
Before the agent searches anything, it works out who is asking. Country, entity and contract type come from Microsoft Entra ID and the HRIS, so a plant worker in one country does not receive the office rule from another. Where a question needs a personal fact, remaining leave, an entitlement, a seniority date, a robot reads that employee's own record in SAP SuccessFactors through UiPath Integration Service. Where policy does not settle the matter, an "Ask HR" button creates a UiPath Action Center task with the whole conversation attached, so the advisor starts with context instead of a forwarded thread.
The AI does one job and is fenced in on both sides. It retrieves and phrases; the citation proves where the answer came from, and the guardrails route health, disciplinary and grievance topics to a person without attempting an answer. When the library is silent, the agent says so rather than improvising. HR owns the library: publish a new version and every answer changes the same day, which is the part that makes this governance rather than a chatbot. Where the client already holds Microsoft 365 Copilot licences, the same design runs in Microsoft Copilot Studio, with agent usage billed through Copilot Credits.
UiPath conversational agent published to Microsoft Teams; UiPath Context Grounding over the SharePoint library with inherited permissions; agent guardrails and escalations; UiPath Action Center tasks completed inside Teams; SharePoint versioning, content types and document owners
The agent instructions and guardrails, the country, entity and contract filters, the "Ask HR" escalation and its routing, the question and answer log, and the monthly policy-gap list for document owners
UiPath Integration Service connectors for SAP SuccessFactors and Microsoft OneDrive & SharePoint; Microsoft Entra ID attributes for the asker's entity and contract type
How the automated process works
- PersonAn employee asks in Microsoft Teams, in their own language, the same way they would ask a colleague
- AutomationThe agent resolves country, entity and contract type from Microsoft Entra ID and the HRIS before it searches
- AutomationIt searches only the approved policy index and answers with the paragraph, the document and its version date
- SystemWhere a personal fact is needed, a robot reads that employee's own record in SAP SuccessFactors
- PersonIf policy does not settle it, "Ask HR" creates an Action Center task with the whole conversation attached
- AutomationEvery question, answer, citation and escalation reason is logged, and low-confidence exchanges are flagged
- PersonHR reviews the weekly gap list, publishes a corrected document, and the answers change the same day
Human-in-the-loop model
Automation handles
- Retrieving and citing the current policy paragraph together with its version date
- Filtering the answer by country, entity and contract type before it is written
- Fetching personal facts from the HRIS, for the asker's own record only
- Creating the HR case with full context and logging every exchange
People decide
- Advisors take edge cases, disputes and individual circumstances
- Policy owners approve every document that enters the library, and set its review date
- HR reviews the weekly list of questions the library could not settle
- Health, disciplinary and grievance topics go to a person by design, never to the agent
Before and after
Systems and integrations
We do not add technology to make an architecture look serious. Every element below has a specific job in this process.
Inputs
- questions asked in Microsoft Teams
- the approved policy library on SharePoint
- country, entity and contract attributes from Microsoft Entra ID
- employee records in SAP SuccessFactors
Automation layer
- UiPath conversational agent
- UiPath Context Grounding
- UiPath Action Center
- UiPath Integration Service
Target systems
- the HR case queue
- the policy library and its document owners
- the interaction log
Human touchpoints: the "Ask HR" button in Microsoft Teams; the advisor's Action Center task; the weekly policy-gap review
Technologies used
takes the question, applies the filters and returns the answer with its citation
Aindexes the approved SharePoint policy library and inherits its permissions
Athe single approved source, with an owner, a review date and a version per document
Awhere employees already ask, and where the "Ask HR" button sits
Athe escalation becomes an advisor task with the conversation attached
Amodel allow-list, PII masking and a full interaction audit
Acountry, entity and contract-type attributes that filter every answer
Apersonal facts from the HRIS and access to the library
Athe same design where the client prefers the Microsoft side; agent usage is billed through Copilot Credits
AIllustrative economic model
What it is worth, with the arithmetic shown.
Advisor minutes are only half of what a policy question costs; the other half is the employee's, and it sits outside the calculator on purpose. The 0.65 share the library can settle without an advisor is folded into the volume, so the calculator runs on 715 questions a month rather than 1,100, at 12 minutes of average handling and €34 a fully loaded HR advisor hour. The employee side, 9 minutes per question at €41 an hour, is about €52,644 a year and appears only in the next section. Nothing here was measured at a client, and licensing, manager time and the cost of a wrong answer are all excluded.
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
- Advisor capacity is released because most lookups never enter the queue, and the queue is left for the cases that need judgement
- Answers are consistent across sites and countries, because each one comes from the same approved version and shows which paragraph it used
- Employees get an answer in the channel where they asked, at any hour, which ends the habit of asking a colleague instead
- Policy owners see which questions are asked most and where documents are unclear, which improves the policies themselves
- HR can show what employees were told and when, because every answer carries the version date of the document behind it
The management view
- Question volume by topic, country and hour becomes visible for the first time, which is a planning input as well as a policy one
- The share answered without an advisor is a number that can be tracked, argued about and improved
- Policy owners receive the most honest backlog they will ever get: the questions their own documents cannot answer
- Every document has a named owner and a review date, enforced by the library rather than by a reminder
Board-level KPIs
Security and governance
Where the data sits and who can see it.
- The agent reads one SharePoint library with named owners, and Context Grounding inherits the permissions of that library, so nobody sees a document they could not open themselves
- Model choice, PII masking and a full interaction audit are enforced by the UiPath AI Trust Layer rather than by convention, and the agent runs in the EU region of UiPath Automation Cloud
- HRIS lookups use a service account limited to the asker's own record; credentials sit in the Orchestrator credential store and are never held in the agent
- Guardrails send health, grievance and disciplinary topics to a person without answering, and that routing is tested before go-live rather than tuned afterwards
- Every answer is logged with its policy version and timestamp under a Microsoft Purview retention label, which is what lets HR show what employees were told before and after a change
Why now
Policy libraries already sit in SharePoint, so the source material exists and the first work is governance rather than writing; the clean-up is a deliverable, not a prerequisite you have to fund separately
Retrieval-grounded agents answer from approved documents with a citation and decline when the library is silent, and the UiPath AI Trust Layer keeps model allow-lists, PII masking and the interaction audit inside your own EU tenant
Employees have already moved their questions into Teams, so the real choice is whether HR answers there with a controlled agent or by email while managers improvise; the modelled €4,862 a month of advisor capacity is the smaller half of that decision
Relevant executive roles
Consistent, defensible policy answers across countries, and the first measurable view of what employees actually ask
Advisor capacity moves from lookups to cases without adding headcount, and holiday cover stops being a risk
A lookup that never enters the queue is the cheapest ticket there is, and the queue is where the service level is lost
Common questions and objections
The agent answers only from documents HR approved, shows the paragraph it used, and says when the library is silent. Wrong answers today come from memory and from an old handbook version, and neither of those leaves a trace.
That is the first deliverable rather than an obstacle. The pilot forces the clean-up one country at a time, and the weekly gap list keeps naming what is missing after go-live.
They already ask in Teams, and the answer arrives where the question was asked. The "Ask HR" button means nobody is trapped with a machine, which is usually what the objection is really about.
When this is not the right solution
- Question volume is low and a maintained FAQ page with a named owner would do the same job for less
- Policies are not written down, or differ by manager rather than by entity; agree the rules first, automate the answers second
- An HRIS replacement will change the self-service portal within months, or works-council agreement on an AI-assisted HR channel cannot be obtained in the timeframe
A question for the next management meeting
Nothing in our reporting says whether two employees in two countries received the same answer to the same policy question last month; what would it take to know?
Implementation approach
What we deliver, and what we need from you to start.
We deliver
- A categorised month of HR mailbox questions and a review of which policy documents are actually current
- The policy library clean-up plan: one approved version per topic, with an owner and a review date
- Agent design, guardrails, the country, entity and contract filters, and the "Ask HR" escalation flow
- HRIS integration for personal facts, scoped to the asker's own record
- Testing against an HR-approved question bank before any employee sees it, then deployment, training and monitoring
We need from you
- Named policy owners and an HR sponsor who can approve what enters the library
- Tenant access, a test group, and the entity and contract attributes that drive the filtering
- Where required, the works-council consultation planned into the timeline rather than discovered in it
Stages
Discovery
One month of questions categorised; current policy versions identified and the gaps listed
Design
Agent behaviour, guardrails, filters, escalation rules and the topics that never reach the agent
Build
Library index, agent, HRIS lookups, Teams deployment and the interaction log
Validation
Accuracy, citation correctness and escalation rate measured on an HR-approved question bank
Go-live
One country and the top twenty question types, then further countries, languages and lookups
Quick win. Effort is driven by the state of the policy library rather than by technology: how many documents are current, how many countries differ, and how much of the answer depends on personal data.
If two advisors answer the same question today, nothing tells you whether they agreed.
Send us your ten most-asked policy questions and the documents that answer them. We come back with the agent answering them on your own content, the questions your library cannot settle, and the clean-up those gaps imply.
Test ten of your policy questionsThe 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 Other solutionsOne place to ask, an answer with its source, gaps reportedThe same questions are asked in channels and tickets because nobody can find the current answer.
View solution Other solutionsEvery HR and payroll question becomes a case with an ownerNobody can say how many payroll questions are older than five days, because a mailbox does not know what one is.
View solution Other solutionsLeave requested, checked and posted without an e‑mailLeave should not be approved from memory, rekeyed twice and confirmed by asking.
View solution Case studyThe HR agent on TeamsAn employee asks on Teams about leave, a certificate or a benefit.
View case study Case studyAutomated new-hire onboardingNew hires ready to work sooner — less paperwork for HR.
View case studyIndustries we deliver this in most oftenManufacturing & industryRetail & e‑commercePublic sectorServices & ITShared services