Hundreds of staff keying the same transactions into systems that will not be replaced this decade.
A legacy ERP with no API, an Excel queue and a team that copies fields between two windows all day.
You need a robotic process automation consultant who will also keep the bots running after go-live.
Claims, portals and mainframe screens where a human still does the typing.
Numbers from EY, Deloitte and HfS research on RPA programmes — the licence is the small part; design and support decide whether the bots survive.
of first RPA projects fail, according to EY’s experience across 20 countries.
EY, Get ready for robots ↗of companies in a Deloitte survey managed to scale beyond 50 robots.
Deloitte, via DO Supply ↗of the total cost of RPA is the licence; the rest is implementation, infrastructure and support.
HfS Research, via Blueprint ↗is the typical cost of one bot before ongoing maintenance is counted.
Blueprint ↗of RPA proofs of concept lead nowhere, says IRPA AI’s Frank Casale; bots built without IT lose long-term support.
CIO.com ↗of adopters say RPA exceeded expectations on accuracy, speed, flexibility and compliance — when it is done right.
Deloitte, via DO Supply ↗Put your monthly volume into the calculator and see hours and dollars.
RPA services — robotic process automation — give a software robot the work a clerk does in a system that has no API: open the queue, read the form, copy the fields into the right screen, check them, press Save, keep a screenshot, pass the exception along. The bot does it the same way, every time, around the clock. The difficulty is not building the first bot; it is building one that survives a screen change, scales past a pilot and has someone to fix it — which is where most first RPA projects fail.
Wireclad works as a robotic process automation consultant that builds custom workflows around the systems you already run — SAP, Dynamics, Oracle, Citrix sessions, web portals, Excel queues — rather than selling a platform. Where a system has an API we use it, through n8n or Python, because it is cheaper to keep alive. Where it does not, UiPath, Power Automate Desktop or Playwright drive the screen. A language model reads the incoming PDFs and explains the exceptions; a health check watches every bot and raises an alert before the queue backs up.
Transactions are posted within minutes of arriving, with evidence attached. Clerks handle the exceptions instead of the typing, and each exception comes with its reason. Bots are monitored, versioned and covered by change control, so a system update is a planned fix rather than a surprise. Payments and financial approvals still go through a person.
The run path of a typical RPA job. Each queue item travels it on its own; anything the bot cannot read or verify stops and goes to a person with the evidence.
A row in the Excel queue, an email with a form or a file in SharePoint — every item becomes a job with its own log.
OCR plus a language model pull the fields from the PDF or scan into the schema the target system expects.
Totals, codes, dates and statuses are checked against your rules before anything is typed anywhere.
Where the target has an API, the job goes to an integration node. Only true legacy screens get a UI bot.
SAP, Dynamics, Oracle or NetSuite record created through the official interface — no screen scraping to maintain.
UiPath, Power Automate Desktop or Python with Playwright logs into the portal or Citrix session and keys the fields.
The bot reads the screen back and compares it with the input; a mismatch stops the job before it is committed.
Screenshot and the resulting record ID are stored per item for audit and reconciliation.
Daily reports from bank, insurer and government portals fetched by a headless browser into SharePoint or S3.
Pandas compares the two extracts by key and lists every mismatch.
The model classifies why an item failed — missing field, changed layout, business rule — and writes the context for a person.
The exception becomes a task with the screenshot and the reason, in the queue or in Teams.
A failed login, a changed selector or an empty queue raises an alert in Slack, Teams or PagerDuty, and the daily counts go to the operations lead.
Typical minutes per transaction for a shared-services clerk. Your own numbers go into the calculator below.
Based on 10.5 min by hand and 30 sec with the flow per item, from the table above.
The bots read, key, verify and report. People choose the processes, own the exceptions and approve anything that moves money.
Stable, rule-based, high-volume work first; the choice and priority are yours.
Anything the model cannot explain goes to a specialist with the screenshot and the input.
Bots prepare; a person approves before money moves.
System updates are assessed for their effect on the bots before they are rolled out.
Bot logs and evidence are reviewed on a schedule you set.
Credentials in a vault, scoped access and HIPAA or SOX requirements met by design.
Flip a switch to hand a step to the flow or take it back.
Benchmarks from EY’s Get ready for robots and Deloitte’s RPA survey: lower cost per transaction, bots that survive system updates, and exceptions that reach a person with the evidence attached.
We map the process as it runs today, count the minutes and agree what the flow must never do on its own.
A working flow on your real data, in a sandbox. You see every run and every exception.
Edge cases, approvals and alerts, then the switch-over — with the old way kept as a fallback.
Monitoring, fixes when a vendor changes a format, and a monthly report of hours saved.
An API integration talks to the system directly and does not care how the screen looks; it is cheaper to run and rarely breaks. A screen bot is for systems with no API — mainframes, Citrix sessions, old portals. We check for an API first and build UI bots only where there is no other door.
Industry estimates put a single bot at $5,000–$15,000 before support, and the licence at only a quarter to a third of the total. We scope each process separately and keep the design simple, so the maintenance stays small; exact savings depend on your volumes.
The health check raises an alert the first time a login or selector fails, and the queue holds the items instead of losing them. Change control means we usually know about the update before it ships; when we do not, the fix is a selector, not a rebuild.
Not necessarily. Python with Playwright, Power Automate Desktop and n8n cover most cases without a per-bot licence. UiPath or Automation Anywhere make sense when you already own them or need their orchestration at scale.
It depends on the volume and the hours it replaces, so we do not promise a number. EY reports 25–40% lower cost of manual operations; the calculator above with your own transaction counts gives a realistic range.
Yes, and it usually should be. OCR with a language model reads the PDF or scan into structured fields, validation checks them, and only then does the bot or the API post the record. The model also explains exceptions so a person can resolve them quickly.
The bot gets its own account with the minimum rights, credentials live in a vault, every action is logged with a screenshot, and payments still need a human approval. That is what HIPAA and SOX auditors ask for.
Describe it in two sentences — we reply within a day with a workflow sketch.
hello@wireclad.com →