Business processes and their automation

How we look at it

A business process rarely happens where it is documented. The real flow lives in inboxes, spreadsheets and manual bridges between systems, and that is where we measure the cost. In one of our analyses the starting point was counting four manual re-keyings of data between systems, plus the order lines lost because they arrived as free text in a message. Only a list like that settles what to automate first, and exposes the steps that are cheaper to delete than to accelerate.

What we have built

We build systems where the process is executable, not described. Document flow: a BPMN engine with an approval matrix that enforces segregation of duties, SLA timers and escalations. Invoices: full integration with Poland's national e-invoicing system under the FA(3) schema, retrieval per tax ID, a parser hardened against XXE attacks, XSD validation, tax-ID checksum control, treasury whitelist checks, split-payment handling, deduplication, plus a JPK_V7M(3) return generator. Debt recovery as a legal process: from reminder, through an interest note with a versioned central-bank rate and the statutory 40, 70 or 100 euro recovery compensation, to a claim filed with the electronic order-for-payment court over its SOAP interface. Production planning to ISA-95 across four levels, S&OP through RCCP and MPS down to detailed scheduling, with a rolling horizon, plan-freezing zones, and a CP-SAT constraint solver as the decision-maker. ERP integrations go exclusively through an idempotent outbox, because a duplicated accounting event is not a technical glitch but an error in the books.

What follows

Changing the process stops being an IT project and becomes a change to a flow you can see in the data. Every transition leaves a trail, so the process can be accounted for and shown to an auditor. The usual side effect of measurement is a shorter process: once the cost of each step is visible, steps get deleted instead of automated.