AI automation for companies: processes, marketing and chatbots

We automate repetitive office work: retyping documents, schedules, settlements, reports and answers to common customer questions. We start with the process that takes the most hours.

up to 100 h a month saved at Piuur Schoon
12 s to import 23,653 line items from Comotel at ABC Szalunki
5 languages ALU-GUM shopping assistant, no subscription
Our services

What do we automate?

Three areas where we have delivered automation for clients

Business process automation

Schedules, settlements, retyping documents between programs and printouts that someone still does by hand in Excel or in a notebook. See also order and document workflow.

  • Document imports from a warehouse program or ERP
  • Settlements calculated by the system in real time
  • Schedules repeated automatically week to week
  • Statements, reports and quotes generated as PDF
  • Business rules written into code and covered by tests
Next.js PostgreSQL PHP Python

Measurement, reporting and marketing automation

Campaigns can only be automated once sales are measured to the last euro. We set up measurement, reports and automatic data feeds to your sales channels.

  • Sales events in GA4 with Consent Mode v2
  • Refunds reported to GA4 server-side
  • Looker Studio reports
  • Product feeds for Merchant Center and Ceneo
  • Google Ads changes made through the API, with a copy of the previous state
GA4 Google Tag Manager Looker Studio Google Ads API
Benefits

What does automation change?

Time goes back to the business

At Piuur Schoon, scheduling, phone calls and manual counting took up to 3 hours a day. With the system that is up to 100 hours a month saved.

Numbers that match

At ABC Szalunki the system loaded a full year of documents from Comotel: 2,136 of 2,136 documents matching its control totals, 23,653 line items in 12 seconds.

No fee per query

The ALU-GUM shopping assistant answers in five languages with no subscription and without sending customer data to outside services.

AI automation

Where does AI automation pay off most?

AI marketing automation

Google Ads campaigns with automated bidding learn from conversion data. If sales are not measured, the algorithm optimises for the wrong goals. First we set up sales events in GA4 with Consent Mode v2 and server-side refund reporting, then product feeds for Merchant Center and Ceneo that update automatically.

We make campaign changes through the Google Ads API, with a copy of the state before each change. More on campaigns: SEO and Google Ads.

AI automation in B2B

In B2B companies most time goes into retyping documents between programs, calculating settlements and sales staff answering the same questions about stock and prices. At ABC Szalunki the system imports documents from Comotel and calculates statements using 31 business rules: 2,136 of 2,136 documents match.

When customers order regularly, a B2B platform with prices and stock taken straight from the ERP works better than emails and phone calls.

Process

How do we implement automation?

Every step ends with a result you can check

1

Process description

We talk to the person who does the work today. We write down the steps, the exceptions and how many hours it takes. At Piuur Schoon a one-hour conversation was enough.

2

Rules in writing before code

We write down the calculation rules and check them against the client's examples. At ABC Szalunki we asked 55 questions before writing any code, then reconstructed the spreadsheet formula and proved it with a calculation.

3

First version on your data

We run the solution on a copy of real data and compare the result with what a person calculates today. Only a version whose results match goes into daily use.

4

Development from real requests

After launch we collect what the system did not handle. The second version of the ALU-GUM assistant was built from the list of questions the first one could not answer.

Projects

Automation delivered for our clients

Store for 5 markets

ALU-GUM: the Felek shopping assistant

Problem: The client wanted a chat like a competitor's, built on a rented service with a subscription.

Solution: Our own assistant that recognises the intent of a question, copes with typos, maps everyday car nicknames to the right model and searches the wheel catalogue.

Result: Answers in five languages with no fee per query. Unanswered questions go to a panel for the owner.

See the project: ALU-GUM
Services, Netherlands

Piuur Schoon: schedules and settlements instead of paper calendars

Problem: Up to 3 hours a day spent on scheduling, phone calls and counting hours and pay by hand.

Solution: One online schedule on everyone's phone, visits repeated every week, and hours, pay and cash from clients settled by the system.

Result: Up to 100 hours a month given back to the business.

See the project: Piuur Schoon
Construction, in progress

ABC Szalunki: rental settlements without retyping into Excel

Problem: Every issue and return note from Comotel was retyped into a spreadsheet by hand to calculate rental.

Solution: The system loads documents from Comotel and calculates statements using 31 business rules.

Result: 2,136 of 2,136 documents matching, 23,653 line items loaded in 12 s, a PDF statement in 1.3 s against a 3 s requirement.

See the project: ABC Szalunki
FAQ

Frequently asked questions

The questions we hear most often before an automation project.

Where do you start with automation in a company?

With the process that takes the most hours and repeats every day or every week: retyping documents, schedules, settlements, reports. First we describe how the work is done today and write down the rules. Code comes only once we know which result has to match.

Does a store assistant have to use a paid language model?

No. The shopping assistant we built for ALU-GUM works without a language model: it recognises the intent of a question, copes with typos and answers in five languages, with no fee per query. When a task does need a language model, we choose one that fits the task and the budget.

Does company and customer data go to external services?

Only if we agree on it. We build solutions so that data stays in your store or system. If we use an external language model, we say up front which data goes to it and keep that to a minimum.

How much does an automation project cost?

We bill hours worked at €52 net per hour. After a conversation and a description of the process we give you a quote with the scope of work, so you know up front what you are paying for.

How much does an AI chatbot cost?

A simple chatbot that answers questions about delivery, returns and payments from store data usually starts at €780 net. An assistant that searches the product catalogue, handles several languages and collects unanswered questions in a panel costs €2,080-5,190 net. Our own engine has no per-query fees. If a task needs a language model, we list its API cost separately in the quote.

What is marketing automation?

Handing repetitive marketing work over to systems: collecting sales data, reports, product feed updates and campaign changes. Google Ads automated bidding strategies, based on machine learning, are only as good as the conversion data they receive. That is why we start by measuring sales down to the cent and only then automate campaigns.

How long does implementation take?

It depends on the process. The scheduling and settlement system for Piuur Schoon took one month. The shopping assistant was part of the three-month ALU-GUM store project. Larger systems, such as the settlement system for ABC Szalunki, are rolled out in stages, and each stage ends with a working part.

Will automation replace employees?

It takes retyping, counting and answering the same questions off their hands. People handle what a system cannot: customers, exceptions and decisions. At Piuur Schoon the time saved means the business can take on more clients without the admin growing with them.

Which process takes up most of your team's hours?

Describe it in a few sentences. We will tell you whether it can be automated, where to start and roughly what it may cost.