Home · Solutions · Legal & compliance
Solution · Legal & complianceEvery indicator has an owner, a source, a formula and the document it came from
Sustainability figures collected once, evidence attached
Energy, workforce, waste, transport and supplier figures are collected to one definition each period, validated against the prior period, and filed with the document behind every number.
Executive summary
Last year's sustainability numbers cannot be reproduced without the people who sent them.
The product is the indicator register, built before any robot runs.
Any figure in the pack traces back to its document, its formula and its approver without anyone being asked.
the register and evidence library on SharePoint under Microsoft Purview retention labels; the Power BI semantic model; the reporting pack
Business problem
Sustainability reporting
Almost none of the data in a sustainability report is produced by the people who report it. Electricity and gas consumption sit on utility invoices in accounts payable and on meters read by facilities. Workforce figures live in the HR system, on a definition renegotiated every year. Waste is on contractor reports, freight is in the transport system, and part of what must be disclosed exists only at suppliers. The reporting team owns the definitions and none of the sources.
Collection therefore becomes correspondence. A request goes to seven site contacts; answers return over weeks as spreadsheets and as numbers in the body of an email. They are normalised by hand: units converted, billing periods aligned to the reporting period, the leased warehouse ruled in or out.
Reproducibility is what fails under examination. A figure is worth the trail behind it: source document, factor version, formula, approver, date. When that trail lives in a mailbox and a workbook, last year's number cannot be rebuilt, and the pack is defended from memory by whoever still holds the role.
At group scale the arithmetic works against everyone. Seven sites, 120 indicators, most of them per site and per period, plus supplier responses that arrive incomplete. The cycle becomes a project with a fixed cost and no residue.
How it works today
This is the shape of the cycle in most groups reporting for the first or second time.
- PersonThe sustainability manager copies last year's indicator list into a fresh workbook and emails seven site contacts
- WaitingAnswers come back over three to six weeks: spreadsheets, numbers in email bodies, photographs of a meter panel
- PersonUtility invoices are opened one by one in the accounts payable folder and the consumption is retyped into the energy tab
- SystemWorkforce figures are exported from the HR system by a payroll colleague, on a definition agreed again each year
- PersonSupplier questionnaires go out by email; a third come back incomplete or in the wrong unit and are chased individually
- Risk of errorValues are compared with last year by eye, so a unit error, a double-counted site or a missing month is found late
- WaitingSign-off happens weeks after collection, when nobody remembers which invoice a number came from
Why the current process costs more than it appears
Nobody planned this work; it accumulated.
- Collection, not judgement, consumes the year. The technical decisions a specialist is paid for take a fraction of the time; the rest goes on finding the source, converting it and writing it down.
- Definitions are renegotiated annually because they were never written down: whether the leased warehouse counts, whether contractors are in headcount, which meter belongs to which site.
- Chasing is invisible work with a real price. About a third of requests need a reminder, a call and a correction, and none of that appears in a plan or a budget.
- A figure without a trail is a liability that surfaces only when someone samples it. Until then it looks identical to one that can be evidenced in a minute.
- When the person holding the method leaves, the method leaves. What remains is a workbook of formulas nobody can explain.
Cost of inaction
Sixteen thousand euro a year is an affordable number, and that is what keeps this process alive. The cycle finishes, the pack is signed, no line in the accounts carries the cost, and next year two people start again from an empty workbook. The rows price the collecting, not the fact that nothing built in one cycle survives into the next.
What grows is a different exposure. A figure that cannot be traced to a document is indistinguishable from one that can, until somebody samples it. A sample that lands on an untraceable number turns a reporting exercise into a remediation project on someone else's timetable.
A plausible organisation with realistic proportions. The figures are there to be recalculated on your data; they are not a client result.
A European manufacturing group: seven production sites in four countries, around 3,400 employees, SAP S/4HANA for finance and logistics, workforce data in the HR system, Microsoft 365 E3. Reporting sits with two people in group finance and one contact per site.
120 indicators in scope, most per site and per period: energy and fuel across about 40 utility accounts, waste from six contractors, freight from the transport system, workforce from the HR system, plus roughly 90 supplier questionnaires. The cycle produces around 340 collection and validation tasks, about 28 a month when spread across the year.
Requests by email, answers in spreadsheets, consumption retyped from invoices, workforce data pasted from an export, supplier responses chased individually, all assembled in one master workbook.
Roughly 75 minutes per collection and validation task: finding the source, converting the unit, aligning the period and noting what was done. Nothing links the value in the workbook to the document behind it.
Each indicator gets a record: owner, unit, source, formula, factor version, evidence rule, validation rule. Robots collect what a system holds, questionnaires collect what only a person holds, every value is tested against its prior period, and the pack carries its evidence.
In the modelled case the cycle becomes a period that opens and closes rather than an annual project, outliers are examined while the source is current, and any figure traces to its document in a few clicks. That is the design's intended output, not an observation at a client.
Proposed solution
The product is the indicator register, built before any robot runs. Each of the 120 indicators gets one record: owner, unit, source, formula, factor version, evidence, period boundary and a plausible range. It lives as a versioned SharePoint list, surfaced as a tab in Microsoft Teams, so the definition used for a given period is knowable years later.
Collection is then deterministic. A time trigger in UiPath Orchestrator opens the period and turns the register into a queue: one item per indicator, site and source. Robots pull utility and fuel invoices from the accounts payable library and the energy mailbox, read by the pre-trained UiPath Document Understanding Utility Bills model: supplier, meter point, period covered, quantity, unit, reading type. Meter exports arrive as files, workforce data comes from the HR system, waste and freight from operations, and contractor reports with no fixed layout go through Generative Extraction. What only a person can supply is requested through a Microsoft Forms questionnaire, which returns typed fields with units.
Validation happens on arrival: unit, period boundary, prior period, plausibility band, site figures reconciled to the group. Outliers and low-confidence readings become tasks in Microsoft Teams for the named owner; what is outstanding is listed per owner in a Teams channel and escalated to the site manager. Every figure carries its chain from formula and factor version to source document, approver and timestamp. The pack is filed in SharePoint under Microsoft Purview retention labels, the same values feed a Power BI report, and sign-off is recorded in the Microsoft Teams Approvals app.
UiPath Document Understanding pre-trained Utility Bills model and Generative Extraction; UiPath Orchestrator queues, time triggers and audit log; UiPath Action Center tasks in Microsoft Teams; UiPath Integration Service connectors for Microsoft Outlook 365, Microsoft OneDrive & SharePoint and Microsoft Teams; SharePoint lists and libraries with metadata and versioning; Microsoft Forms; Microsoft Teams Approvals app; Microsoft Purview retention labels; Power BI
The indicator register and its versioning; the collection plan per period; extraction and apportionment of consumption to reporting periods; the validation and reconciliation rules; the chase and escalation cycle; the evidence chain; the sign-off flow, the pack and the Power BI model
Read access to SAP S/4HANA and the HR system through UiPath SAP activities (BAPI, OData or a reporting view); meter exports where no interface exists; loading Microsoft Forms responses
How the automated process works
- AutomationA time trigger opens the period and turns the register into a work item for every indicator, site and source
- AutomationRobots collect what a system holds: utility and fuel invoices, meter exports, workforce data, waste and freight
- SystemDocument Understanding reads each consumption invoice, and the consumption is apportioned to the reporting period
- AutomationWhat no system holds is requested through a Microsoft Forms questionnaire, and the response lands as fields with units
- SystemEvery value is tested against its unit, its period boundary, the prior period and a plausibility band, and site figures are reconciled to the group
- PersonOutliers and low-confidence readings become tasks in Microsoft Teams for the indicator owner, with the prior value and the document attached
- AutomationWhat is still missing is listed per owner in a Teams channel, reminded and escalated to the site manager
- PersonEach site and topic is signed off in Teams Approvals; the pack is assembled with its evidence chain and filed under retention labels
Human-in-the-loop model
Automation handles
- Collecting every value a system already holds, and reading consumption from invoices, meter exports and contractor reports
- Requesting and structuring what only a person can supply, then chasing what is late
- Testing units, period boundaries, prior-period variance and site-to-group reconciliation
- Assembling the pack, attaching the evidence and recording who approved which version
People decide
- What each indicator means: boundary, definition, inclusions, leased and joint operations
- Whether an outlier is an error or a real change, and what explanation is recorded
- Which factor set applies for the period, and when a version is superseded
- Sign-off per site and per topic, and any restatement of a published figure
Before and after
Systems and integrations
Every entry can be checked in vendor documentation. The evidence class is stated next to each one.
Inputs
- utility and fuel invoices in the accounts payable library
- the energy mailbox in Outlook
- meter and building management exports
- SAP S/4HANA and the HR system
- waste contractor reports
- transport data
- Microsoft Forms questionnaires
Automation layer
- UiPath Orchestrator
- UiPath Robots
- UiPath Document Understanding
- UiPath Integration Service
- UiPath Action Center
Target systems
- the register and evidence library on SharePoint under Microsoft Purview retention labels
- the Power BI semantic model
- the reporting pack
Human touchpoints: validation tasks in Microsoft Teams; the gap list in a Teams channel; Teams Approvals for sign-off; the register as a Teams tab
Technologies used
pre-trained Utility Bills model for consumption invoices; Generative Extraction for contractor reports
Aopen each period, queue one item per indicator and site, run collection, validation and chasing
Aowners confirm outliers and low-confidence readings in Teams
Areads the energy mailbox and the invoice library, writes values back
Asite and supplier questionnaires returning typed fields with units
Athe register, the values and the evidence behind each figure
Asite and topic sign-off against the version approved
Acoverage, gaps and variance on the values that go into the pack
Aheadcount, turnover and training data, and the invoice records
AIllustrative economic model
The arithmetic is open, so it can be argued with.
The labour below is the smaller half of the case and the only half that prices cleanly, so read it as a floor rather than as the benefit. An annual cycle of roughly 340 tasks spread evenly gives 28 a month; 75 minutes is what one takes when it includes finding the source, converting the unit, aligning the period and recording what was done; €38 an hour is a fully loaded cost for a group finance role in Central Europe. 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
- Any figure in the pack traces back to its document, its formula and its approver without anyone being asked
- Collection stops being an annual project: the period opens on a date and closes on one
- Errors a human eye misses are caught on arrival, when the source is current and the correction is cheap
- Site contacts answer fewer questions and answer them once, because the definition travels with the request
- Adding a site, an indicator or a supplier is a change to the register, not a new column and a new conversation
The management view
- Progress through the cycle is visible while it runs: which indicators are in, which are outstanding, and who owes them
- The reporting boundary and its definitions become documented assets that survive turnover
- Preparation for an external review becomes a filter rather than a search, because the evidence sits with the figure
- The same numbers feed management reporting and the published pack, so internal targets and external disclosure stop diverging
Board-level KPIs
Security and governance
The automation holds exactly the rights it needs, and not one more.
- Every account the robots use can read and nothing more: no posting in SAP, no change to HR records
- Secrets are held in a vault the platform reads at run time; no password, token or connection string sits in a workflow
- Collected values are never overwritten. A correction creates a new version with a reason and an author, and the superseded value stays readable
- Whoever collects a figure does not sign it off, and the approval is recorded against the version approved
- Values, evidence, approvals and the audit trail stay in your Microsoft 365 tenant; the automation layer runs in the EU region of UiPath Automation Cloud
Why now
The European framework has moved repeatedly while companies prepared for it. The European Commission's sustainability reporting page records the simplification omnibus proposed on 26 February 2025, the directive of 14 April 2025 postponing first reporting for part of the population, a quick-fix delegated act adopted on 11 July 2025, and political agreement on 9 December 2025. Which wave a group lands in is a question for its auditors; the data foundation underneath is the same either way.
Assurance turns producing a number into producing a number somebody else can test, and that trail costs a fraction to build while the sources are current. Alongside it sits the modelled €1,330 a month of collection.
The components are ordinary now: a supported model for utility invoices, generative extraction for documents nobody could template, questionnaires that return fields, and sign-off inside Microsoft Teams.
Relevant executive roles
The sustainability statement carries the same expectation of traceability as the financial statements
Specialist time moves from collecting and converting to boundary decisions, targets and the conversations that change the numbers
Site managers stop being a data source chased by email and start owning a few indicators with a definition attached
Common questions and objections
What is changing is scope and timing, not the need to produce figures that can be evidenced. The register, the sources, the formulas and the evidence chain are the same work whichever standard version applies, and they cannot be done quickly at the end.
They are not asked to. Most values come from systems that already hold them, and where a person is genuinely needed the request is a Microsoft Forms questionnaire with the definition attached, answered once rather than clarified three times.
The named owner of that indicator, exactly as today, with two differences: the value shows what it was built from, and the approval sits against the version signed rather than a document since edited.
When this is not the right solution
- A single-site company with a handful of indicators, where a well-kept workbook and a folder of invoices are cheaper than an automation
- Nobody can say what the reporting boundary is or which entities sit inside it. That is a group decision an automation can enforce but not invent
- Most of the scope depends on primary data suppliers do not yet collect. Questionnaires will chase it, but the gap is a contracting problem first
A question for the next management meeting
If a reviewer picked three figures from last year's sustainability statement this afternoon, how long would it take us to produce the source document, the formula and the approval behind each one?
Implementation approach
Delivery runs in stages, so it can be stopped at any point.
We deliver
- The indicator register for one reporting boundary: owner, unit, source, formula, factor version, evidence rule, validation rule
- Collection from the systems that hold data, with invoice and meter intake and extraction, plus the questionnaire path for sites and suppliers
- The validation layer: units, period boundaries, prior-period variance, plausibility bands and site-to-group reconciliation
- The chase and escalation cycle in Microsoft Teams, and the validation tasks for indicator owners
- The evidence chain, the sign-off flow, the assembled pack and the Power BI report
We need from you
- The indicator list and reporting boundary, in whatever form they exist
- A named owner per indicator group, and one person who owns the definitions
- Read access to SAP S/4HANA, the HR system, the invoice library and the energy mailbox
- The factor set you use and the rule for when a version is superseded
Stages
Discovery
Indicators, sources, owners, definitions, units and where each figure comes from today
Design
Register schema, validation and apportionment rules, evidence rules, escalation paths, security model
Build
Collection, extraction, validation, the Teams touchpoints, the evidence chain and reporting
Dry run
A closed historical period collected end to end and compared with what was reported by hand
Go-live
One period under supervision, then the remaining sites, then rule and band tuning
Enterprise. Effort is driven by the number of source systems and sites, how consistently the reporting boundary is defined, and how much data exists only at suppliers.
A correct figure with no trail behind it still fails the sample.
Send us five indicators and the detail behind them: how each is collected today, in which unit, from which document. You get back an assessment of what can be collected without a person, and where the evidence chain breaks.
Take five indicators back to sourceThe neighbouring process usually has the same problem
Your whole insurance programme is administered from one workbook and one person's mailbox.
View solution Management & planningThe management pack in Power BI, not fourteen Excel filesThe board pack should not depend on which analyst merged which spreadsheet on which day.
View solution Operations & qualityProduct compliance evidence, ready before it is asked forCustomers ask for the declaration; someone searches supplier emails and a shared drive to find it.
View solutionIndustries we deliver this in most oftenManufacturing & industryTransport & logisticsRetail & e‑commerceEnergy & utilities