Home · Solutions · Finance & accounting
Solution · Finance & accountingEvery receipt read and every policy rule applied before the money goes out
Expense reports audited automatically before reimbursement
Each expense report is checked in full against policy, receipts and history before payment; employees answer questions in Teams and finance only sees the exceptions.
Executive summary
Stop paying nine expense claims in ten on trust because nobody has time to check them against policy.
Mientha delivers a policy audit layer between submission and payment: every report is taken from your T&E tool, every receipt is read.
Every report is checked against the same rules, so out-of-policy items are found before payment, not in next year's audit sample.
SAP Concur (status, comments, adjustments, release); SharePoint audit archive; UiPath Insights
Business problem
Travel & expenses
Expense reimbursement has two jobs: pay people back quickly, and pay only what the policy allows, with the VAT the company is entitled to and never the same receipt twice. Because the jobs pull against each other, most expense teams check everything quickly and audit a sample, typically one report in ten, and the sample is the only place where the policy is actually enforced.
The cost lands on several desks: a queue that peaks in the last week of every month, employees waiting two or three weeks for their money, managers approving in bulk on Friday afternoon without opening a receipt, and internal audit looking at a control that operates on 10% of the population.
Scale makes it worse: more first-time travellers, more per-diem regimes and VAT rules, a thinner sample. A taxi receipt claimed in two months is visible only when reports are compared with each other, and nobody compares 3,200 reports by hand.
How it works today
- PersonThe employee photographs receipts in the mobile app and builds the report at month end by hand
- PersonThe line manager approves the whole batch on a Friday afternoon without opening the receipts
- WaitingReports sit at the processor step for one to six working days
- PersonA processor confirms receipts are attached and amounts match, applying the policy from memory; one in ten gets the full audit
- Risk of errorDuplicates, card transactions claimed again in cash, weekend dates with no trip and till receipts unfit for VAT deduction pass unless sampled
- PersonQuestions go to the employee by email; the report is parked and the Excel list of open queries grows
- WaitingParked reports wait a week on average, miss the reimbursement run and are estimated in the accrual
- SystemReleased reports go to payroll or accounts payable; sample findings are filed for the auditors
Why the current process costs more than it appears
The cost grows where nobody is looking.
- Six minutes buy a completeness check, not an audit. The full audit takes 15 to 20 minutes and covers one report in ten; what the tool's own checks miss on the other nine goes to payment.
- The email loop is the slowest part: the employee is on a train, the question reads "please check your hotel line", and the report waits a week.
- VAT leaks quietly. In several countries, Poland among them, a till receipt without the company's details cannot support the input VAT deduction; checked from memory on a sample, that rule is applied inconsistently.
- Duplicates are invisible to a per-report check; finding them means comparing across reports and months, which no team does by hand.
- The policy's unwritten exceptions live in three heads; when one person leaves, the sample shrinks and the questions get vaguer.
Cost of inaction
Because the queue clears every month, this is easy to leave alone, and the labour cost above is modest. The number nobody measures is larger: at an illustrative €400 per report this group reimburses about €1.28 million a month (3,200 × €400); every 0.1% of that spent outside policy is roughly €15,400 a year (€1.28 million × 12 × 0.001) that the sample never sees. We do not know your rate, and with a one-in-ten sample neither do you.
A control covering 10% of the population, with findings nobody follows up, is a management-letter point in waiting, and each new country adds rules to a checklist held in three heads.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
A consulting and engineering group, 1,800 employees in six European countries, SAP Concur for expenses, Microsoft 365 E3; a three-person expense team in a shared-service centre in Poland.
3,200 expense reports a month; roughly 55% corporate-card transactions, 45% cash and per-diem claims; receipts mostly phone photos; one report in ten audited in full.
Managers approve in bulk; the expense team checks completeness, applies the policy from memory and samples one report in ten; questions go by email and are tracked in Excel.
About six minutes per report on average; 600 to 800 reports queued in the last week of the month; a week per query; the ten-working-day reimbursement promise missed every quarter end.
Every report is picked up from Concur, every receipt is read by UiPath Document Understanding and compared with the claim, the policy runs on every line; clean reports are released, employees get precise questions in Microsoft Teams, the reviewer sees only rule hits.
In the modelled case, 100% of reports are audited against the same rules, manual work concentrates on the exception share (modelled at 20 to 25% initially) and clean reports are reimbursed within two working days. Illustrative, not a client result.
Proposed solution
Mientha delivers a policy audit layer between submission and payment: every report is taken from your T&E tool, every receipt is read, your policy runs on every line, and the outcome is release, a question to the employee, or a task for a reviewer. Limits, per-diem tables, city caps, VAT rules per country and duplicate windows sit in a rule table that finance owns and changes without a developer; the robot records the version it applied. The only machine-learning component is the receipt reader; no language model is involved.
Employee questions move from email to Microsoft Teams: a task with the exact question, a link to the receipt and answer options. Policy breaches go to the finance reviewer with the evidence side by side: receipt image, values read and typed, the rule that fired, matching lines from other reports. Items needing a director's sign-off reach the Teams Approvals app with the receipt attached.
SAP Concur offers audit services of its own; ours is complementary: company-specific rules, local VAT logic, receipt compared with claim, checks across reports and months, follow-up in Teams. Your interface to payroll or the ERP is not touched.
UiPath Document Understanding pre-trained Receipts model and Validation Station; UiPath Orchestrator queues, triggers and audit; UiPath Integration Service connectors for SAP Concur, Microsoft Teams, Microsoft Outlook 365 and Microsoft OneDrive & SharePoint; UiPath Action Center actionable notifications in Microsoft Teams; Microsoft Teams Approvals app; UiPath Insights
T&E intake, rule table and rule engine, receipt-versus-claim comparison, query templates with reminders and escalation, reviewer and director routing, write-back, SharePoint audit archive, dashboards, runbook
Release for payment in SAP Concur via the API where the workflow step is exposed, otherwise by a robot in the processor's queue; export or API intake for other T&E tools
How the automated process works
- AutomationA manager-approved report is picked up from SAP Concur through its connector and queued in Orchestrator with lines and receipt images
- AutomationDocument Understanding reads every receipt with the pre-trained Receipts model (merchant, date, total, tax, currency) and compares the values with the claim; unreadable receipts go to Validation Station
- SystemThe rule engine runs the policy on every line: limits, per-diems, VAT support per country, duplicates across reports and card transactions, weekend dates, attendees, mileage; every hit gets a reason code and severity
- AutomationReports with no hits are marked as audited with the rule version and released for payment
- PersonEmployee-fixable hits become an Action Center task in Microsoft Teams with the question and the receipt linked; reminder after two working days, manager copied after five
- PersonPolicy breaches go to the finance reviewer as a task in Teams with the evidence prepared; director items go to the Teams Approvals app with the receipt attached
- AutomationDecisions are written back to the T&E tool, the audit file goes to SharePoint, and Insights shows pass rate, rule hits and query turnaround
Human-in-the-loop model
Automation handles
- Reading every receipt and comparing it with the claim
- Running the whole policy on every line, every time, with the current rule version
- Asking employees routine questions, chasing answers, releasing clean reports, writing the audit record
People decide
- Whether an over-limit item is justified, with the receipt and the employee's history in view
- Whether a suspected duplicate is one, when the evidence is ambiguous
- Director sign-off on exceptions the policy reserves for them
- The rules: limits, per-diem tables, VAT logic and severities stay with finance
Before and after
Systems and integrations
Where a rule suffices we do not use a model. Where judgement is needed, a person decides.
Inputs
- SAP Concur or another T&E tool (reports, lines, card transactions)
- receipt photos and PDFs
- rule tables on SharePoint
Automation layer
- UiPath Orchestrator
- UiPath Robots
- UiPath Document Understanding
- UiPath Integration Service
- UiPath Action Center
Target systems
- SAP Concur (status, comments, adjustments, release)
- SharePoint audit archive
- UiPath Insights
Human touchpoints: Action Center tasks in Teams; Teams Approvals; weekly summary in the team's channel
Technologies used
pre-trained Receipts model reads merchant, date, total, tax and currency; Validation Station for the unreadable
Aqueue every report, run the rules, write back status, retry, log, audit
Apolling events for submitted reports; reads and updates Concur objects
Achannel summary, email fallback, rule tables, audit archive
Aemployee questions and reviewer decisions inside Teams
Adirector sign-off with the receipt attached; audit in Microsoft Purview
Apass rate, rule hits by department, query turnaround
AIllustrative economic model
Start by questioning the assumptions.
Six minutes is the average that falls out of two very different jobs: a four-to-five-minute completeness check on nine reports in ten, and 15 to 20 minutes of full audit on the tenth. The ranges are illustrative and not measured at a client. €32 is a fully loaded hourly cost for an experienced expense processor in a Central European shared-service centre; only the team's time is counted, not the out-of-policy spend recovered.
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
- Every report is checked against the same rules, so out-of-policy items are found before payment, not in next year's audit sample
- Clean reports are released within a day or two of submission, the part employees actually notice
- Checking time for the clean majority drops to zero; the team works on exceptions with the evidence prepared
- Duplicates and receipts unfit for VAT deduction are caught at line level, every time
- Questions reach employees in Teams with the receipt and the exact point at issue, and are answered in minutes
The management view
- A control that demonstrably covers 100% of the population, with rule version and outcome recorded per report
- Policy compliance by department, country and expense type becomes a monthly number instead of an audit anecdote
- Rule changes (a new per-diem rate, a lower hotel cap) apply from the next run, identically for everyone
- The process no longer depends on which processor is on shift; growth is absorbed by robots
Board-level KPIs
Security and governance
Trust in automation is built on the audit trail, not on a promise.
- A dedicated service account, not a person's login, holds rights to read, comment, adjust and release, and none to change bank details or trigger payments; its credentials sit in Azure Key Vault via the Orchestrator credential store
- Receipts carry personal data; images, extracted fields and audit files stay in your Microsoft 365 tenant on SharePoint with retention labels, and processing runs in the UiPath Automation Cloud EU region
- Segregation of duties: the robot never approves a report with a policy hit, reviewers cannot change a rule, rule changes are versioned and signed off by the policy owner; reviewer access via Microsoft Entra ID groups
- Every decision, by robot or person, is logged with rule version, evidence and timestamp; Teams approvals are also audited in Microsoft Purview; the receipt reader is a pre-trained extractor with no generative component
Why now
Each month on the current routine buys 320 hours of checking (about €10,200 in the model) and an audit of one report in ten; the other nine carry a leakage rate nobody has measured
Growing project businesses add headcount and countries, each bringing first-time travellers, a new per-diem regime and another VAT rule: one row in a rule table, a retraining for a memorised checklist
A pre-trained Receipts model, an SAP Concur connector with polling events and Action Center tasks inside Teams are standard, documented platform capabilities today
Relevant executive roles
A control over the whole population, supported VAT deductions, and an expense accrual read from a queue rather than estimated
The expense team stops being the month-end bottleneck and its work becomes measurable by exception
Employees are reimbursed in days instead of weeks, and the questions they get are precise and the same for everyone
Common questions and objections
Concur's checks cover what they are configured to cover; its audit services are a separate offering not every client uses. Our layer reads the receipt itself, applies local rules such as VAT support and project caps, finds duplicates across reports and runs the conversation in Teams; where Concur's audit services are in use, we take the residual.
A consistent check of everyone is fairer than a random sample. Questions arrive within a day, name the exact line and show the receipt; most close with a one-line answer, and employees who do everything right get paid faster.
That is the input the Receipts model is built for. What it cannot read goes to Validation Station or back to the employee for a better photo; the rest of the report continues.
When this is not the right solution
- Volumes in the low hundreds a month, where the sample already covers a meaningful share and the tool's own checks are enough
- A policy without numeric limits, where decisions rest on judgement; writing it down comes first
- Expenses still submitted on paper with stapled receipts and no digital images; digitise the submission first
A question for the next management meeting
Does anyone in finance hold a list of last year's reimbursed expenses that were actually tested against our policy, and how quickly could they produce it?
Implementation approach
We start with one slice of the process and extend only once it is proven.
We deliver
- A sample of your reports, read line by line: hit categories, receipt quality, volumes
- The rule table and rule engine: limits, per-diem regimes, country VAT logic, duplicate windows
- Receipts model configuration and the validation flow for unreadable receipts
- The SAP Concur (or other T&E tool) integration: intake, comments, adjustments, release
- Teams tasks and approvals, the SharePoint audit archive, Insights dashboards and a runbook
We need from you
- The expense policy with numeric limits and per-diem tables per country
- Three months of reports with receipts and your sample audit findings
- A process owner in the expense team and a policy owner in finance
- A technical account for the T&E tool and approval of the UiPath Teams app in your tenant
Stages
Discovery
Rule inventory, hit categories, volumes, receipt quality
Design
Rule table, routing matrix, query templates, security model
Build
Receipts model, rule engine, T&E integration, Teams tasks
Validation
Three months of historical reports compared with your sample findings
Go-live
One country first; the team reviews every automated decision
Optimisation
Rule tuning from false positives, further countries and entities
Quick win. Effort depends on the number of T&E tools and countries, per-diem regimes and receipt quality.
Your expense policy is enforced on the tenth report and trusted on the other nine.
Export one month of expense reports (header data and receipt images, anonymised if you prefer) with your expense policy; we run our rule set and return the hit list with a written assessment of what it implies.
Run our rules on one month of reportsThe neighbouring process usually has the same problem
Stop paying your finance team to move numbers from PDFs into the ERP.
View solution Finance & accountingVendor master changes with bank-account fraud controlOne convincing email is all it takes to send a six-figure payment to a fraudster's account.
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 solutionIndustries we deliver this in most oftenManufacturing & industryServices & IT