Home · Solutions · IT & services
Solution · IT & servicesMachine-detectable failures found the week they appear, not in next spring's audit
Accessibility checked every week, not once before the audit
Robots crawl every published page on a schedule, log each machine-detectable accessibility failure against the page and its editor, and keep the evidence the statement review needs.
Executive summary
The statement is written once. The site changes every day, and nobody is checking in between.
We build a scheduled sweep of everything the entity publishes and an inventory of who owns each page.
Failures are found within days of publication rather than within a quarter, while they are cheap to fix and the editor remembers the page.
SharePoint page inventory and evidence archive; Power BI accessibility report; the statement-review extract
Business problem
Digital accessibility
A public entity publishes continuously and checks occasionally. The Polish act of 4 April 2019 binds public bodies to WCAG 2.1 at level AA and requires an accessibility statement reviewed every year by the end of March, and after every significant change. Publishing, meanwhile, is decentralised: a dozen organisational units post announcements, tenders and attachments against their own deadlines.
What accumulates between reviews is ordinary rather than exotic. An image goes up without alternative text, a heading level is skipped because the editor wanted smaller type, a scanned attachment carries no text layer. None of it is visible from the inside, and each closes the page to somebody. The coordinator holds accessibility alongside another job and hears months later; the secretary signs the statement and answers the complaint.
What breaks at scale is the evidence. A complaint asks what the page contained, when it went up and what was done about it. Those answers exist only if somebody wrote them down at the time, and a sweep squeezed between other duties records almost nothing.
How it works today
Whatever platform sits underneath, this is the shape we find before automation.
- PersonAn editor publishes a notice and attaches the PDF that arrived with it
- WaitingNothing looks at the page again until the quarterly sweep or a complaint
- PersonThe coordinator opens pages one at a time against a checklist
- Risk of errorThe sweep stops when the time does, and the part skipped is not recorded
- PersonFindings reach the units as one spreadsheet, with no owner per page
- Risk of errorCorrections are not re-checked, so a fixed page and an ignored page look identical
- Risk of errorWhen a resident complains, what the page contained is reconstructed from memory
Why the current process costs more than it appears
Nobody planned this work; it accumulated.
- Checking by hand never catches up with publishing. About 190 pages a month are created or edited here, and a quarterly sweep describes a site that has moved on.
- Between sweeps the estate is unmeasured, so the office cannot answer what a resident asks: is the page I need readable today.
- A scan is the cheapest thing to publish and the dearest to correct: by the time anyone notices, the source file is archived and the page is linked from elsewhere.
- Evidence is the line nobody budgets for and the first thing an auditor asks about. Without a dated record, a complaint is answered with an assurance.
Cost of inaction
A page that fails an accessibility rule still loads, still looks right to the editor who published it, and still counts as published. That is why nothing forces this to change: the cost falls on people who cannot read the page, and most do not write in. The statutory penalties are real, up to 5,000 zł for a missing statement or an inaccessible bulletin and up to 10,000 zł for a persistent breach, but they are the smallest part of it.
What compounds is the evidence gap. Every quarter without a record makes the next statement harder to write honestly, and one that says "partially compliant" without listing what is wrong invites the question of how the office knows.
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 350 staff in eleven organisational units, with three subordinate units publishing to the same public information bulletin; Microsoft 365 across the office.
Three public sites, about 2,400 pages; roughly 190 created or edited a month, of which about 70 carry an attachment, mostly scans.
One coordinator walks the estate with a written checklist once a quarter. Some quarters allow a few hundred pages, others fewer, and findings go out as one spreadsheet.
About nine minutes of checking per page. A full sweep of 2,400 pages is 360 hours a quarter, which no part-time role has, so it becomes a sample nobody records.
Robots crawl the three sites on a schedule, run the checks a machine can run, test every attachment for a text layer, and file each finding against the page and its owning unit.
In the modelled case, failures surface within days instead of within a quarter, the coordinator's hours move from finding problems to deciding about them, and the review starts from a record. These figures follow from the assumptions above and nothing else.
Proposed solution
We build a scheduled sweep of everything the entity publishes and an inventory of who owns each page. The robot walks the sitemaps of the main site, the bulletin and the services portal, queues every page new or changed, and runs each through the accessibility test set in UiPath Test Cloud. Attachments get a second, deterministic check for extractable text and a declared language. Findings then land where the work belongs: each editor gets their own in Microsoft Teams with the rule in plain language, while the coordinator gets the trend by unit and the evidence pack for the annual review.
The split between machine and person is the part we will not blur. An automated check can prove that a page fails a rule; it cannot prove that a page is accessible, which is what the W3C's own guidance on evaluation tools says. The robot issues no verdict: it reports failures with their evidence, the pages a person must still read, and what was not checked. Conformance status stays a human judgement.
UiPath Test Cloud accessibility (WCAG) and web testing with evidence in Test Manager; UiPath Orchestrator time triggers, queues and audit log; UiPath Action Center notifications in Microsoft Teams; Microsoft SharePoint with Microsoft Purview retention labels; Power BI reporting
The crawl, the page inventory and owner map, the rule set in business language, the attachment check, routing per editor, the re-check after each correction, and the statement-review extract
The read path into each publishing platform, from a sitemap or export where one exists and from the pages themselves where it does not; the mapping of a page to its unit and editor
How the automated process works
- AutomationAn Orchestrator time trigger starts the weekly sweep; the robot walks the three sitemaps and queues every page new or changed
- AutomationEach page runs the accessibility test set in UiPath Test Cloud: alternative text, contrast, heading order, form labels, language and link text
- AutomationEvery attachment on those pages is tested for extractable text, a title and a declared language
- SystemEach finding goes to the page record on SharePoint with the failing element, the rule, a screenshot and a timestamp
- AutomationFindings are grouped by owner, and each editor receives only their own list in Microsoft Teams
- PersonThe editor corrects the page and marks it done; the next sweep confirms or reopens it
- PersonThe coordinator works the pages flagged for judgement and records each decision
Human-in-the-loop model
Automation handles
- Crawling, queueing, running the rule set and re-checking every page after a correction
- Opening published attachments and testing them for a text layer, a title and a language
- Filing each finding against a page, a unit and an editor with its evidence and date, and keeping the coordinator's pack current
People decide
- Whether a page is accessible. The robot reports failures and gaps in coverage; nothing it produces is a conformance verdict
- What the statement says. The status, the inaccessible content and the reasons are signed by a person who carries responsibility
- Everything a rule cannot judge: whether alternative text describes the picture, whether a heading structure makes sense, whether a document is understandable
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
- the main site and its sitemap
- the public information bulletin
- the services portal
- published attachments
Automation layer
- UiPath Orchestrator
- UiPath Robots
- UiPath Test Cloud
- UiPath Action Center
Target systems
- SharePoint page inventory and evidence archive
- Power BI accessibility report
- the statement-review extract
Human touchpoints: editor findings in Microsoft Teams; the coordinator's review queue in Action Center; the weekly summary in a Teams channel
Technologies used
runs the accessibility (WCAG) and web checks on every crawled page and stores each result with evidence
Acrawl the sites on a schedule, queue pages, retry and log every run
Aeditors complete their own findings in Microsoft Teams, without a new system or login
Apage inventory, owner map and evidence archive with versions and retention
Acarries the weekly summary and the review queue
Aopen findings by organisational unit and age, and the coverage behind each review
AIllustrative economic model
What it is worth, with the arithmetic shown.
This prices the quarterly sweep spread evenly across the months: 2,400 pages a quarter is 800 a month, nine minutes is what one page costs at a desk, and €23 an hour is a fully loaded cost of an administrative post. Nothing was measured in an office, and the arithmetic shows released capacity, 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
- Failures are found within days of publication rather than within a quarter, while they are cheap to fix and the editor remembers the page
- The coordinator's time moves from finding problems to deciding about them, so the part that needs judgement gets the hours
- Each unit sees a short list of its own pages, so accessibility becomes a task somebody can finish, not a standard somebody is told about
- Residents reach the same information by the same route whatever they read it with, which is what a complaint is really about
- The estate can grow without the rota growing with it, because a schedule absorbs more pages and a part-time role cannot
The management view
- Coverage becomes a number the office can state out loud: how many pages, on what date, against which rules
- Accessibility stops being one person's worry and becomes a line each unit owns, with an age against every open finding
- The obligation survives a change of coordinator, because what the office knows lives in a system, not a notebook
Board-level KPIs
Security and governance
An auditor should be able to reconstruct every decision.
- The sweep reads only what any visitor can read, through a dedicated account with read rights and nothing else. It never writes to a page
- Secrets sit in the credential store rather than in a workflow, the process runs under its own identity, and every run records what was requested and when
- Findings, screenshots and the page inventory stay in the office's Microsoft 365 tenant under a Microsoft Purview retention label, and the robots run from the EU region of UiPath Automation Cloud
- Checking and publishing stay separate: the robot reports, the editor edits, and the segregation an auditor expects is kept
Why now
The statement is reviewed once a year by the end of March and again after every significant change, and a site a dozen units edit every week is in significant change all year
A complete quarterly sweep costs 120 hours a month of an administrative post, about €33,120 a year of capacity the office does not have and so does not spend
Accessibility (WCAG) checks are a standard test type in UiPath Test Cloud, and Action Center delivers a finding into an editor's Microsoft Teams. Both are configuration, not development
Relevant executive roles
signs the statement and answers the complaint, and can do both from a record, not an assurance
stops spending the quarter finding problems and starts deciding which ones matter
owns the publishing platform without owning the content, and gets a dated picture of what it is doing
Common questions and objections
No, and no honest tool can promise that. An automated check proves a failure; it cannot prove accessibility, which is what the W3C's guidance on evaluation tools says. You get every machine-detectable failure with its evidence, plus the pages a person must still read.
They receive fewer items than today, because the list is only their own pages and says what to change. Today they get a spreadsheet for the whole office, months late.
A separate regime with a different addressee. Its Polish implementing act, applied since 28 June 2025, addresses economic operators rather than public bodies; an office stays under the 2019 digital accessibility act.
When this is not the right solution
- A small estate that rarely changes. A few dozen stable pages are cheaper to check with a checklist than under a scheduled sweep
- A platform with no field for alternative text and no control over heading levels. Fix that first, or findings pile up against editors who cannot act on them
- No page ownership. If nobody can say which unit owns which area, findings have nowhere to go, and the owner map comes first
A question for the next management meeting
Since our accessibility statement was last reviewed, how many of the pages published on our three sites has anybody actually opened and checked, and where is that written down?
Implementation approach
What we deliver, and what we need from you to start.
We deliver
- An inventory of the three sites: page counts, publishing patterns, attachment volumes and the owner of each area
- The scheduled crawl and the change detection that keeps each run to what moved
- The test set in UiPath Test Cloud, each rule in language an editor can act on, plus the attachment check
- Findings routed per editor in Microsoft Teams, with a re-check after each correction
- The evidence archive on SharePoint and the coordinator's report
We need from you
- The sites in scope and their sitemaps or an export, where one exists
- The organisational units and editors, and who owns which area of each site
- Your current statements and last findings, so the rule set starts from what you know
Stages
Discovery
Sites, page counts, publishing patterns and the owner map
Design
Rule set, escalation thresholds, routing, evidence model and retention
Build
Crawl, test set, attachment check, SharePoint archive and the Teams touchpoints
Validation and go-live
A full sweep of the real estate, false positives tuned out, then scheduled runs
Quick win. Effort is driven by the number of publishing platforms, whether a sitemap exists for each, and how clearly pages map to editors.
The statement was signed in March. The site has changed every day since.
Send us the sites in scope, with a sitemap for each if you have one. We run one full sweep and come back with the machine-detectable findings ranked by rule and by organisational unit.
Run one sweep across your sitesThe neighbouring process usually has the same problem
Gaps in the bulletin are found by a resident, a journalist or an inspector, never by the office itself.
View solution IT & servicesEvery release regression-tested, including the phoneRegression is a checklist two people click through on the last afternoon before a release.
View solution Legal & complianceThe archive that knows what it may destroy and whenCases close, files do not: categories unchecked, transfer lists retyped, appraisal postponed another year.
View solutionIndustries we deliver this in most oftenPublic sector