Home · Solutions · Finance & accounting
Solution · Finance & accountingDeclarations registered the day they arrive, payments matched before anyone is chased
Local tax accounts that reconcile themselves
Robots register every declaration, compare it with the property and household registers, match payments to accounts and remind before any demand; officials decide and sign.
Executive summary
Forms are typed in twice, payments arrive without a usable reference, and the first letter a resident gets is a demand.
Nothing new needs buying.
A declaration is registered on the working day it arrives, so the charge that follows reflects what was declared rather than last quarter's picture.
the tax and fees system; the document management system; the SharePoint evidence archive
Business problem
Revenue administration
A municipality assesses and collects part of its own revenue, and every account behind it is a relationship with a named person or business: what is owned, how large, how used, how many people live there. Charges and instalments follow from that record, so when the record is wrong the office is the last to know.
Declarations, information forms and corrections arrive from four directions: the counter, the post, the electronic mailbox and the portal. Each is registered, carried to the tax department and keyed in. Nobody compares the typed figure against the property register or the household figures unless something looks unusual, so the registers drift apart quietly and the difference surfaces years later as a backdated correction.
Money behaves no better. Some titles have an account number per taxpayer and match themselves; the rest land in one account with whatever description the payer invented. What a clerk cannot place goes to a suspense list, at once cash the office holds and an arrear the office chases. Arrears are worked when there is capacity, so the order of chasing is set by whoever had time, two taxpayers in the same position are treated differently, and for many the first word about an unpaid instalment is a formal demand.
How it works today
This is the shape we find in most offices, whatever the tax system underneath.
- PersonDeclarations, information forms and corrections are registered from the counter, the post, the electronic mailbox and the portal, then sent to the tax department
- PersonA clerk keys each form into the tax and fees system and files the paper against the case
- Risk of errorThe declared area, use or household size is checked against the property register only on suspicion
- SystemCharges and instalment notices are produced from the tax system and sent to the address it holds
- PersonStatement lines are matched to accounts by hand from the reference, the amount and the payer's name
- WaitingLines nobody can place sit on a suspense list until someone has time, sometimes for months
- PersonArrears are worked after each instalment date from a spreadsheet list, and the first letter many residents receive is a formal demand
Why the current process costs more than it appears
Behind every exception is an hour nobody logged.
- Unmatched money is counted twice: a payment on the suspense list sits in the office's account and on the arrears list at once, so hours go into chasing a debt already paid and the resident into proving it.
- Register drift compounds. A difference nobody looks for grows year on year, and when found it reaches back over several periods, each recalculated, explained and defended to the taxpayer.
- Chasing that begins with a formal demand is the expensive way round: a reminder costs a message, the formal route costs a file, correspondence and the office's standing with the person concerned.
- Two taxpayers in the same position end up treated differently, not by anyone's decision but by capacity, and the knowledge of which accounts are difficult sits with two long-serving clerks rather than in any procedure.
Cost of inaction
Nothing above prices the money that arrives and is not recognised. A payment on the suspense list is cash the office holds and an arrear it chases, and both sides cost hours nobody records. A register difference nobody looks for is free until the year it is found, and then it is a recalculation across several periods, with a decision, a letter and an explanation owed.
The department will keep coping, which is why this survives budget after budget: peaks absorbed by overtime, the arrears list worked when there is room, no month that looks bad. What accumulates is a register agreeing with reality a little less each year, and a chasing order nobody can defend.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
A city office of about 260 staff serving roughly 60,000 residents; a tax and fees department of seven; a domain tax and fees system from a Polish vendor, a document management system and Microsoft 365.
21,000 tax and fee accounts; 2,900 declarations, information forms and corrections a month averaged over a visibly seasonal year; 6,100 incoming payments a month.
Forms are registered in the correspondence system and keyed in by two clerks; the registers are compared only on suspicion. Statement lines are matched by hand and arrears chased from a spreadsheet when there is room.
About six minutes per item across registration and matching, plus unrecorded time on investigations, backdated corrections and calls from residents who have already paid.
Robots collect declarations from every channel, read them with UiPath Document Understanding, compare each against the property register and the household figures, and put differences to a clerk as a decision rather than a correction three years later. Statement files are matched by account number, reference, amount and payer; a reminder precedes any formal step.
Registration moves from days to the same working day in the modelled case, payments matched without a person reach 85 to 92 per cent, and the department's month goes to the cases that need judgement. The figures are a model on stated assumptions.
Proposed solution
Nothing new needs buying. Mientha builds the flow around the tax and fees system the office already runs, on the UiPath Platform and its own Microsoft 365 tenant. Declarations are collected from the electronic mailbox, the scanner library, the portal export and counter scans, then read by UiPath Document Understanding: the forms are fixed layouts, so each gets its own document type and low-confidence fields reach a person first.
The cross-check is the part that pays for itself. For each declaration a robot pulls the account's record, the matching property register entry and, for the waste charge, the office's household figure, then returns agreement, a difference with both values, or missing data. Only agreement is registered without a person; a difference becomes a task in UiPath Action Center delivered in Microsoft Teams, and the robot never decides which register is right.
Payments are handled with the same determinism: each statement line is matched on the individual account number where a title has one, then on reference, amount and payer against open instalments, under tolerances the office sets. Once an instalment passes its date a reminder goes out in the account's channel, on paper where the office holds no electronic address, before the formal route is reached.
UiPath Document Understanding with Validation Station; UiPath Orchestrator queues, triggers, credential store and audit; UiPath Action Center tasks completed inside Microsoft Teams; UiPath Integration Service connectors for Microsoft OneDrive & SharePoint and Microsoft Teams; Power BI refresh and a Teams report tab
The intake per channel, a document type for each form, the comparison rules, the matching engine and its tolerances, the reminder ladder, the reviews in Teams and the dashboards
The tax and fees system through the vendor's interface where one exists and screen automation where it does not; statement file pick-up from the bank; dispatch through the document management system
How the automated process works
- AutomationA new item in the electronic mailbox, the scanner library or the portal export is queued in Orchestrator under its registration number
- AutomationDocument Understanding classifies the form, extracts the fields and sends low-confidence ones to Validation Station
- SystemThe robot reads the account, the property register entry and the household figure, applies the rules and records agreement, a difference or missing data
- AutomationAgreements are registered in the tax and fees system and the evidence filed against the case
- PersonDifferences reach the responsible clerk as an Action Center task in Teams with both values; the clerk decides and the decision is logged
- AutomationStatement files are parsed every working day and matched by account number, reference, amount and payer; the rest becomes a short list with proposals
- AutomationInstalments past their date trigger a reminder in the account's channel, and what is still unpaid reaches the weekly arrears review in Teams
Human-in-the-loop model
Automation handles
- Collection, reading and registration of declarations from every channel, with the evidence filed against the case
- The comparison against the property register and household figures, and the arithmetic behind charges, instalments and due dates
- Matching of statement lines to accounts and instalments, and the short list of what did not match
- Reminders in the recorded channel, escalation timers, and the weekly reviews
People decide
- Which register is right when two disagree, and whether a correction or a request to the taxpayer follows
- Every assessment, correction, allocation of an overpayment, write-off and move to the formal route, each signed by an authorised official who carries the responsibility exactly as before
- The rules themselves: tolerances, reminder intervals, channel preferences and thresholds stay with the department
Before and after
Systems and integrations
Everything below runs on licences and systems you already hold, or would need anyway.
Inputs
- electronic mailbox through the document management system
- counter and post scans on SharePoint
- portal export
- bank statement files
- the property register extract
- the office's household figures
Automation layer
- UiPath Orchestrator
- UiPath Robots
- UiPath Document Understanding
- UiPath Integration Service
- UiPath Action Center
Target systems
- the tax and fees system
- the document management system
- the SharePoint evidence archive
- the Power BI semantic model
Human touchpoints: Action Center tasks in Microsoft Teams; the weekly review in the department's channel; the Power BI report tab in Teams
Technologies used
reads declarations, forms and corrections from scans, PDFs and electronic deliveries; Validation Station for low-confidence fields
Aqueue every form and statement line, run the rules, write to the tax system, retry and audit
Adifference decisions and the arrears review completed without leaving Teams
Ascanner library, evidence archive and the department's channel
Acollection per title, ageing of arrears, open differences and the automatic match rate
Aaccounts, charges, instalments, payments and correspondence
CIllustrative economic model
A model, not a promise.
Registration and matching are the only work priced below; deciding a difference, answering a taxpayer and preparing a correction stay with people. Nine thousand items a month is 2,900 declarations and forms plus 6,100 payments, six minutes is what one costs at a desk when nothing goes wrong, and €19 an hour is a fully loaded cost of an administrative post. Nothing was measured in any office, and the arithmetic shows capacity returned to the department, never a saving and never a post.
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
- A declaration is registered on the working day it arrives, so the charge that follows reflects what was declared rather than last quarter's picture
- Differences between the two registers arrive as a monthly list with both values, instead of surfacing years later in a correction
- Most payments are matched on the day they land, so the arrears list stops naming people who have already paid, and the calls stop with it
- A resident hears from the office before the formal route begins, in the channel they use, which recovers much of the money for the cost of a message
- The order in which accounts are chased follows one written rule, which is what equal treatment requires and what an audit asks to see
The management view
- The department's position is visible without asking anyone: what arrived, what was registered, what waits for a decision, how old it is
- Collection becomes measurable per title and per instalment, so the effect of a reminder before a demand is read rather than assumed
- Every automated step carries its evidence: the document, the values compared, the rule applied, the person who decided and when
- The process stops depending on two long-serving clerks and becomes something a new starter can run
Board-level KPIs
Security and governance
Trust in automation is built on the audit trail, not on a promise.
- Robots hold their own accounts in the tax and fees system, with rights to register and read and none to waive, write off or refund; every write is attributable in the system's log
- Taxpayer data stays where it is: documents in the office's Microsoft 365 tenant, records in the tax system, and the automation in the European Union region of UiPath Automation Cloud or on UiPath Automation Suite the office hosts itself
- The difference list, the suspense list and the arrears review are visible to the tax department only, through Microsoft Entra ID groups; the separation between who prepares and who signs is untouched
- Passwords and keys live in the Orchestrator credential store rather than with a person, and rules, thresholds and reminder intervals are versioned, with changes approved by the department head
Why now
Collection is the part of a municipal budget the office genuinely controls, and every gain there is a decision the council need not take elsewhere; the modelled €17,100 a month of capacity is only the visible part
The electronic channel to residents now exists at scale: the Ministry of Digital Affairs reported over 100 million items delivered through e-Doręczenia and 4.6 million addresses in the electronic address database as at 5 August 2026, which turns a reminder before a demand into a practical option
The building blocks are ordinary and verified: document types trained on fixed forms, Action Center tasks completed inside Teams, statement files parsed deterministically, and no language model in the path
Relevant executive roles
Collection becomes measurable per title and per instalment, and money already in the office's account stops appearing on the arrears list.
Capacity moves from typing and matching to the decisions only an official can take, and the order of chasing becomes a rule that can be shown.
Differences arrive as a decision with both values attached, not as a correction found three years later with a taxpayer on the telephone.
Common questions and objections
No. The robot registers, compares, calculates and drafts. Every assessment, correction, allocation and write-off is decided and signed by an authorised official who carries the responsibility exactly as today; the difference is that it arrives with its evidence, recorded with a name and a time.
Then the robot works the screens with its own service account, doing what a clerk does and logging every step, which takes longer to build and is more sensitive to vendor updates. We ask the vendor first and treat screen automation as a fallback.
Correct, and the design does not assume otherwise. The Supreme Audit Office reported that 15 per cent of people in Poland had never used the internet, on 2022 data, against 9 per cent across the European Union. The channel is set per account and paper stays.
When this is not the right solution
- A small office with a few thousand accounts, where forms are registered on the day they arrive and every statement line is matched before lunch
- A tax register never reconciled with the property register: the first pass produces a difference list larger than the department can work, and that is a project of its own
- No statement file from the bank and no usable reference on payments, so matching has nothing to work with until the banking arrangement changes
A question for the next management meeting
Is the order in which this office chases arrears set by a rule we could show an auditor, or by whichever clerk had time that week?
Implementation approach
The first week looks the same at every client: we look at the data.
We deliver
- A read of one month of your declarations and statement lines: channels, form types, match rate, difference categories, volumes
- Document types trained on your own forms, with thresholds set against your data
- The comparison rules against the property register and household figures, and the matching engine with its tolerances and its short list of what it will not decide
- The reminder ladder and the reviews in Teams, then a pilot on one title and a runbook the department owns
We need from you
- Three months of declarations with the corresponding entries in the tax system
- Three months of statement files and the current suspense list
- A process owner in the tax department and a named owner of the rules and thresholds
- Technical accounts for the tax system, the document system and the statement folder
Stages
Discovery
Channels, form types, difference categories, match rate, volumes
Design
Comparison and matching rules, reminder ladder, thresholds and access model
Build
Document types, robots, integrations, Teams tasks and reporting
Validation
A parallel run on historical months, every difference reconciled with clerks
Go-live and tuning
One title under supervision, the old routine as fallback, then further titles
Departmental. Effort is driven by the number of titles in scope, whether the tax system offers an interface for machines, and how far the two registers already agree.
A payment nobody could match and a demand about the same account, in the same week.
Send us one month of statement lines and one month of registered declarations, with personal data removed. We come back with the match rate we would expect, the difference categories worth automating and the ones we would leave to a clerk.
Match one month of your statement linesThe neighbouring process usually has the same problem
A structured invoice states the amount. It does not state which unit ordered the work.
View solution Customer serviceOne front door for everything a resident reportsA broken street light reported three ways becomes three separate jobs, two of them invisible.
View solution Finance & accountingThe daily cash position without the 7 a.m. spreadsheetStop building the daily cash number by hand from four bank portals; it arrives late and nobody quite trusts it.
View solutionIndustries we deliver this in most oftenPublic sector