Home › Process Automation

Process Automation on Your Terms

In many companies the work between systems still runs by hand. Documents arrive by mail, and someone types them in. The same data goes into the ERP, the product system and a price list. We connect these steps: documents are read, checked against your rules and passed to the right system, and a person sees only what stands out.

Where the hours go

The supplier sendsInto a spreadsheetInto the ERPInto the price database a price list as PDF,48 pagesevery valuetyped by handevery itemcreated by handthe same valuestyped in once more BY HANDBY HANDBY HAND “every supplier writes it differently” THE SAME PATTERN, IN MANY PLACES orders and confirmationsdelivery notes and invoicescertificates and test reportsproduct data from suppliers THREE QUESTIONS WE ASK FIRST Which documents?From how many sources?How many hours a week?

An example from a large retailer with stores and online trade, as recorded in a meeting with the team. A supplier sends its product catalogue as a PDF, 48 pages. Someone types every value into a spreadsheet. Then every item is created by hand in the ERP, and the same values are typed once more into the price database. On top of that, as the team put it: “Every supplier writes it differently.”

The same pattern appears in many places: orders and order confirmations, delivery notes and invoices, certificates and test reports, product data from suppliers. Wherever a document arrives, a person transfers its content into one or more systems.

That is why we ask three questions in the first conversation. Which documents do you handle regularly? How many sources do they come from, such as mail, portals, post or file shares? And how many hours per week does your team spend collecting them and typing them in?

What runs by itself afterwards

DocumentsData between systemsSuppliers and partners TODAYTODAYTODAY from the mailbox, printed,typed intyped into ERP, product systemand price list, one after anotherevery supplier writes itdifferently AFTERWARDSAFTERWARDSAFTERWARDS read, classified and routedto the right system and personentered once, then it reachesevery system that needs itread, matched and normalised;queries drafted for release Fixed rules check every value. A person sees only what stands out, and decides.

Afterwards, three kinds of work run by themselves.

  • Documents are read, classified and passed to the right system and the right person: orders, delivery notes, invoices, certificates.
  • Data between systems is captured once and reaches every system that needs it, from the ERP to the product system to the price list.
  • Lists and data from suppliers and partners are read, mapped to your items and normalised. Where something is unclear, a query to the supplier is prepared, and a person approves it.

In all three cases, fixed rules check every value. The AI reads and maps. Whether a value passes is decided by rules you can read.

Two examples from our operations

The list arrivesAI reads itFixed rules checkInto SAP and PIM by mail, interfaceor web portaland pre-structures it,whatever the layoutevery single value,against your rulesautomatically, the oldprice database retired A person decides only what stands out; queries to the supplier drafted, then released minutes per supplier list, where it used to take hours 6 weeks until the business team ran it themselves SINCE THEN The business team develops the process further itself. The technology stood after two weeks; the process with the people in it started in parallel.

Supplier product catalogues at a large retailer with stores and online trade. Today the catalogue arrives by mail, through an interface or through a web portal. The AI reads it in and pre-structures it, whatever the layout. Fixed rules check every value, first for plausibility, then in deterministic quality checks. Fewer than 5 per cent of the values go to a person. The data flows automatically into SAP and the PIM, and the old price database has been retired. A catalogue that used to take hours is now processed in a few minutes. The technology was in place after two weeks. After six weeks the business team was working with it themselves, and they have developed the process further on their own since then.

New prices in every store overnight at an optician retailer with several hundred stores. In the past, a printed price list reached the stores once a year. Today, price changes reach every store overnight, and online and store prices run in sync. The first solution was live in six weeks and has been running since early 2020.

We name the customers in conversation.

On your terms

1234 It runs where you decideOpen source underneathSensitive documents stay with youYour team runs it managed by us or in your In-House Cloud,hosted or on site; you can change later Liquid.MX, open source; no lock-in,the processes belong to you read by models in your In-House Cloud; hybrid,the cloud sees placeholders only (Privacy Gateway) your business team configures and extends;we maintain the platform

Four things hold for every automation we build.

  • It runs where you decide: on the platform we run for you, or in your In-House Cloud, hosted or in your own data centre. You can change that later.
  • Open source as the base: the automation runs on Liquid.MX. Under it we run open-source integration tools; which one does the work depends on your task and your landscape. There is no lock-in, and the processes belong to you.
  • Sensitive documents stay with you: models in your In-House Cloud read them. If a task needs a cloud model, it runs hybrid through the Privacy Gateway, and only placeholders reach the Cloud.
  • Your team works with it: your business team configures the process and extends it, as in the first example. We maintain the platform underneath.

We use the same tools in our other topics. If an application is missing, we build it with Agentic Engineering and connect it to your data sources with Liquid.MX. If your suppliers' price lists should flow into your pricing, that is part of Agentic Commerce. How we build is described in our toolbox.

How we start

  1. The process check, free during our introductory period. You pick one process, upload three typical documents or lists and answer three short questions. Our agents run the first analysis on local models on our own machines.
  2. The pilot: one process, one measured result, at a fixed price. Six weekly sessions lead from the target picture to go-live. The first two sessions are free, and you decide on the pilot once the plan is on the table. At the end we measure the value against what we agreed at the start.
  3. Operations: the platform on a monthly basis, run by your team or by us. The next process goes through the same steps, only shorter.

Technical teams can also try it themselves with Liquid Upstart, free and open source under Apache 2.0, on their own machine. For technical readers there is also the white paper “From Prototype to Production”.

What we bring

  • Process & Document Automation: document-heavy processes such as invoicing, onboarding, compliance evidence and supply chain, automated on one platform. The process that runs in the pilot is the one that later runs in full operation.
  • Data Engineering & Pipelines: data flows on Liquid.MX that connect your operational systems, your analytics and your automations, observable from day one. For teams whose prototype has to go into operation.
  • SAP integration: if you are moving from SAP PI/PO to a new integration layer, see PI/PO to Cloud Integration.