Home · Solutions · Other solutions
Solution · Other solutionsThe Thursday question, answered before it is asked: what is open, and who owns it
Every HR and payroll question becomes a case with an owner
Every message to the HR mailbox and the Teams channel becomes a case with a category, an owner and a deadline; routine requests are answered by a robot and the employee sees the status.
Executive summary
Nobody can say how many payroll questions are older than five days, because a mailbox does not know what one is.
The intake channels stay where they are.
Triage disappears as a job.
the case store (ServiceNow, Microsoft Dynamics 365 Customer Service or Microsoft Lists); SAP SuccessFactors; the payroll provider's portal
Business problem
HR operations
HR and payroll questions arrive through three channels and are worked by whoever opens them first. There is no case number, so an employee who wants to know what is happening has to restate the whole question to a person who has not read it yet. There is no deadline either, so a query about a missing allowance can sit until the payroll cut-off has passed, at which point it stops being a question and becomes next month's correction.
Advisors spend the first part of every shift reading and sorting rather than answering. Because the queue is a mailbox, the loudest thread wins over the oldest, and urgency is judged from tone instead of from the payroll calendar. Roughly a fifth of what arrives is somebody asking again about something already open, so the same message is read twice and the same lookup repeated by a second person.
The arrangement survives because it never fails outright. Nothing is lost forever, a case tool sounds like an IT project with a budget behind it, and the cost is paid in employee patience, which no report measures. The service level the centre promises its business units can be neither proved nor disproved, so it is argued about instead, and the knowledge of who handles what sits in three heads that take holidays.
How it works today
What follows is the pattern in shared-service centres that never bought a case tool, whatever the HR system underneath.
- PersonAn advisor opens the shared mailbox at the start of the shift and reads down the subject lines, picking whatever looks urgent
- PersonEach message is read in full to work out what is being asked, then the employee is looked up in the HR system
- SystemAnything touching pay is forwarded to the outsourced payroll provider, and the thread waits for a reply that returns to HR, not to the employee
- WaitingThe employee hears nothing for four days and writes again; the new message enters the queue as a new question
- PersonThe Excel tracker is rebuilt from memory on Thursday morning, before the operations call
- Risk of errorA query about a missing allowance passes the payroll cut-off unnoticed and becomes an off-cycle payment and a correction
- PersonThe thread goes quiet and the matter counts as closed, with no record of the answer or how long it took
Why the current process costs more than it appears
The budget shows headcount, not what it is spent on.
- Follow-ups are rework in its purest form: the same question read again, the same record opened again, usually by a second person starting from nothing.
- A payroll query that misses the cut-off does not simply arrive late. It becomes an off-cycle payment for the provider to run, a correction on the next payslip, and for the employee a month with less money than expected.
- Service levels promised to business units cannot be measured, so they are negotiated on impressions and defended with anecdotes, which is an expensive way to run an internal customer relationship.
- When the advisor who knows the German pension rules is on leave, that part of the queue stops moving, because routing lives in habit rather than in a rule anyone else can apply.
- Demand has no visible shape. A benefits change or a bonus run produces a spike felt as pressure and understood, if at all, a quarter later, so staffing is set against last year's guess.
Cost of inaction
Volume follows headcount, but follow-ups follow speed: the slower the centre answers, the more often people ask again, so the chasing grows faster than the business does. Twelve months on, the same fourteen advisors absorb more by reading faster and thinking less, which comes back as corrections, escalations and complaints from business units that were promised a service level nobody can measure.
What ends the arrangement is rarely a decision. A business unit that cannot get an answer hires its own HR coordinator, a second country does the same, and the shared-service model dissolves quietly into local administration that costs more and reports less.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
A business-services group with 2,700 employees across five countries; an HR shared-service centre of fourteen people, employee data in SAP SuccessFactors, payroll outsourced per country, Microsoft 365 E3.
3,400 HR and payroll inquiries a month through one shared mailbox, one Teams channel and phone notes; about 18% are follow-ups on items already open, and about a quarter are routine requests needing no judgement.
Advisors read down subject lines, decide by tone, look the employee up in the HR system, forward pay matters to the provider and answer from the thread. Outlook categories are the only case tool, and an Excel tracker is rebuilt before the Thursday call.
Four minutes of reading and triage on every message before anyone answers anything, seven minutes on each follow-up, and nine on each routine request a rule could have handled.
UiPath Communications Mining labels intent, urgency and the fields behind every message; a case with a category, an owner and an SLA is opened in the case store; routine intents are resolved by robots; the rest reach an advisor queue with the facts attached, prioritised against the payroll cut-off calendar.
In the modelled case the first hour of each shift returns to answering, follow-ups fall because status exists without anyone asking for it, and pay questions are ordered by the cut-off rather than by tone. Illustrative arithmetic, not a client result.
Proposed solution
The intake channels stay where they are. Employees keep writing to the HR mailbox, keep posting in the Teams channel and keep phoning; what changes is the thirty seconds after a message lands. UiPath Communications Mining reads it, decides what it is about (payslip query, contract copy, leave balance, benefits change, a grievance), how urgent it is, and pulls out the employee number, country and pay period the answer will need.
From there the message is a case rather than an email. A record is opened in whichever case store the group already owns: ServiceNow, Microsoft Dynamics 365 Customer Service, or Microsoft Lists where the volume justifies neither. The case carries a category, an owner, a due date and a link to the original thread. Follow-ups are attached to the case they belong to instead of opening a second one, and pay-related cases are checked against the country's payroll cut-off calendar and prioritised accordingly.
Intents that need no judgement are answered by a robot: a payslip copy, a leave balance, a certificate of employment, the address held on file. Everything else reaches an advisor's queue with the employee record and the extracted facts attached, so the advisor starts at the decision rather than at the reading. Sensitive categories, and anything the model is unsure about, never take that path; a named person triages them first. Employees see case number, owner and state in Teams, and Power BI turns the queue into numbers a shared-services leader can put in front of a business unit.
UiPath Communications Mining intent labels, confidence scores and validation; UiPath Orchestrator queues, triggers and audit; UiPath Action Center tasks completed inside Microsoft Teams; the case store's own queues and SLA timers; Microsoft Teams cards for status and escalation
The intent taxonomy and extraction fields, the routing and SLA rules, the payroll cut-off logic, resolution robots for the routine intents that carry the most volume, the employee status experience and the management reporting
UiPath Integration Service connectors for Microsoft Outlook 365 and Microsoft Teams; the case-store API through the ServiceNow or Microsoft Dynamics 365 CRM connector; read access to SAP SuccessFactors for employee, absence and pay-period data
How the automated process works
- AutomationEvery message landing in the HR mailbox or the Teams channel is read by Communications Mining, which labels intent and urgency and extracts employee number, country and pay period
- AutomationFollow-ups are matched to their thread and attached to the existing case, and the employee receives the current status instead of a new place in the queue
- SystemA genuinely new inquiry becomes a case with a category, an owner, a due date and a cut-off flag where the category touches pay
- AutomationRoutine intents are answered end to end by a robot from the HR system, confirmed to the employee and closed with the answer on the case
- PersonEverything needing judgement reaches an advisor's queue with the employee record and the extracted facts attached
- PersonMessages below the confidence threshold, and every sensitive category, go to a named advisor for triage before anything is routed
- AutomationThe employee sees case number, owner and status in Teams; Power BI shows queue age, SLA compliance and intent mix by country and business unit
Human-in-the-loop model
Automation handles
- Reading, labelling and prioritising every inbound message, whichever channel it arrived on
- Extracting the employee number, country and pay period behind the question
- Opening cases, linking follow-ups to the case they belong to and starting the SLA clock
- Resolving routine intents end to end and confirming the answer to the employee in Teams
People decide
- Every case that needs interpretation, judgement or an exception to a rule
- Anything below the confidence threshold and every message in a sensitive category
- Corrections to categories and priorities, which is how the model gets better
- The taxonomy itself, the service levels behind it and the escalation rules
Before and after
Systems and integrations
Where a rule suffices we do not use a model. Where judgement is needed, a person decides.
Inputs
- Outlook shared mailbox
- the HR channel in Microsoft Teams
- phone notes logged by advisors
Automation layer
- UiPath Communications Mining
- UiPath Orchestrator
- UiPath Robots
- UiPath Action Center
- UiPath Integration Service
Target systems
- the case store (ServiceNow, Microsoft Dynamics 365 Customer Service or Microsoft Lists)
- SAP SuccessFactors
- the payroll provider's portal
- Power BI
Human touchpoints: advisor queues in the case store; Action Center triage and escalation tasks in Microsoft Teams; the employee's status card in Teams
Technologies used
reads the HR mailbox and Teams channel, labels intent and urgency, extracts employee number, country and pay period
Aresolve routine intents, apply routing and SLA rules, hold the queue, retries and audit trail
Atriage and escalation tasks completed inside Microsoft Teams
Aintake from the shared mailbox and the channel, status and escalation cards back to people
Aholds the case, its category, owner, SLA timer and history
Aemployee record, absence balances and pay period behind each answer
Aqueue age, SLA compliance and intent mix by country and business unit
AIllustrative economic model
Start by questioning the assumptions.
Reading, chasing and answering the routine are priced together, because the same fourteen advisors do all three at the same €33 an hour: 3,060 messages triaged (0.9 of 3,400) at four minutes, 612 follow-ups at seven and 850 routine requests at nine come to 24,174 minutes a month, or 7.1 minutes averaged across every inquiry that arrives. The volume below is therefore the full 3,400 with blended minutes, so one calculator covers all three jobs. Employee waiting time, off-cycle payment fees and the worth of being able to report at all sit outside the table, and nothing here was measured at a client.
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
- Triage disappears as a job. Every message is categorised, linked and prioritised before an advisor opens it, so the shift starts with answering
- Follow-up volume falls because the employee has a case number and a visible status, and asking again stops being the only way to find anything out
- Pay questions are ordered by the cut-off calendar instead of by tone, which keeps them out of next month's off-cycle payment run
- The centre reports a measured service level per category, country and business unit, and defends its cost with the same numbers
- Holidays and resignations stop stalling the queue, because routing follows a rule rather than one person's memory
The management view
- Open cases by category, age, country and owner replace a weekly guess assembled from a mailbox and a spreadsheet
- Demand acquires a shape: a benefits change or a bonus run shows as a spike the same week, not a quarter later
- Staffing follows the measured arrival pattern per country and intent, so the centre plans instead of reacting
- Handovers and escalations cost nothing to prepare, because each case carries its own history
Board-level KPIs
Security and governance
The automation holds exactly the rights it needs, and not one more.
- Classification runs in the EU region of UiPath Automation Cloud under the AI Trust Layer, with PII masking and an allow-list of models; correspondence reaches no model outside the boundary you approve
- Mailbox access uses Microsoft Graph application permissions scoped to the HR mailbox alone through RBAC for Applications, not a tenant-wide grant that would read every mailbox in the company
- Sensitive categories, grievances, health, disciplinary cases and anything touching a works-council process, are routed by rule to a restricted queue that named advisors see and robots never do
- Robot accounts hold least-privilege read rights in the HR system, secrets sit in a credential store integrated with Orchestrator, and every classification, score and correction is logged, so the model can be explained to a works council or an auditor
Why now
Business units now write response times into their internal agreements with shared services, and a centre that cannot measure its own by category negotiates from anecdote against a number
Capacity is the second argument: €13,277 a month is what triage, chasing and routine answers cost this illustrative centre, and it recurs whether or not anyone decides anything
Communications Mining exposes confidence scores that advisors tune with their own corrections, which turns automated reading of an HR mailbox into a controlled operation with a threshold and an audit trail
Relevant executive roles
An HR service that can be measured, defended in front of the business and improved on evidence rather than on complaints
Advisors work cases instead of inboxes, and the backlog has a shape: category, age, owner
Pay questions stop missing the cut-off, off-cycle payments fall, and the centre's cost has a service level behind it
Staffing follows measured arrival patterns per country and intent, not the pressure of the week
Common questions and objections
They can carry on. The mailbox stays an intake channel and the case is created from the email, so nobody learns a portal to ask a question. What changes is that the email has a case number, an owner and a clock.
Sensitive intents go to human triage by rule, not by model confidence, and anything the model is unsure about goes the same way. The model decides where routine questions go; a person decides everything else.
Then do not buy it. Microsoft Lists holds cases, owners and due dates well enough at this volume, and the intake, the robots and the Teams experience are identical either way.
When this is not the right solution
- A case tool with categories, owners and SLAs is already in place and genuinely used; the work is then routing and resolution, not intake
- Volumes of a few hundred messages a month, where a disciplined mailbox routine costs less than a model to maintain
- No works-council agreement for automated reading of employee correspondence, or a consolidation that will replace the channels within the year
A question for the next management meeting
Nobody in this company can name today's oldest unanswered payroll question, so what exactly is the service level we promise our business units, and on what evidence?
Implementation approach
We start with one slice of the process and extend only once it is proven.
We deliver
- Analysis of three months of anonymised mailbox and channel history, and the intent taxonomy built from it with HR
- Communications Mining configuration, shadow-mode validation and confidence thresholds agreed with the advisors
- Case-store integration: categories, owners, SLA timers, escalation paths and the cut-off rules per country
- Resolution robots for the routine intents that carry the most volume
- The employee status experience in Microsoft Teams and the Action Center triage and escalation tasks
- Power BI reporting on queue age, SLA compliance and intent mix, plus advisor training and a taxonomy runbook
We need from you
- Three months of mailbox history, and a channel export where the Teams channel is in real use
- Access to the case store and read access to the HR system for the resolution robots
- A process owner in HR operations and two advisors available for labelling
- The works-council position on automated reading of employee correspondence, settled before the pilot
Stages
Discovery
Mailbox history, intent taxonomy, sensitive categories, volumes and today's real response times
Design
Case model, routing and SLA rules, cut-off calendar, confidence thresholds and the security model
Build
Model training, case-store integration, resolution robots and the Teams experience
Shadow run
Classification runs without routing while advisors correct labels until precision is accepted
Go-live
One country first, with two resolution robots and daily supervision
Optimisation
Countries, intents and robots added; drift monitored while HR owns the taxonomy
Departmental. Effort follows the number of countries and payroll providers, the spread of intents in the mailbox, and how much of the correspondence a works council allows to be read automatically.
1,900 unread messages in the HR mailbox, and no name against the oldest payroll query.
Export one month of HR mailbox metadata, subject lines and timestamps only, and we come back with the intent categories we find and your real response-time distribution.
Categorise one month of HR subject linesThe neighbouring process usually has the same problem
Your best agents spend their morning sorting mail instead of answering it.
View solution HR & peoplePayroll inputs consolidated and checked before the runPayroll errors are found by the employee on payday, not by the team that built the file.
View solution Other solutionsTickets triaged and routed before an analyst reads themPriority follows the requester's capital letters, and a P1 waits in the queue until a dispatcher has time to read it.
View solution Other solutionsThe same right answer to every HR policy questionTwo advisors answer the same handbook questions all year, from different versions of the handbook.
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