What lies behind “Let software understand you,” which AI functions are already running productively today, and where an honest gap still exists between vision and everyday work.
“Don’t teach users how to use software. Teach software how to understand users’ intent.” With this statement, Abacus co-founder Claudio Hintermann set a direction that reverses the relationship between people and ERP software. No longer does the person learn the logic of the software; instead, the software learns the person’s intent. Abacus Intelligence is the concrete expression of this idea, developed together with AI specialist Artificialy from Lugano. We’ll look at what is already working productively in Swiss SMEs today, what remains future music, and why the most interesting question is not what AI can do, but how it changes your way of working.
First, a clarification that is part of our consulting approach. There’s a lot of marketing around artificial intelligence, even from reputable providers. It’s worth distinguishing between three things: what software delivers productively today, what it shows in a controlled demo, and what is formulated as a vision for the coming years. All three are legitimate, but they have very different value for your investment decision. This article attempts to keep them separate.
The Core Idea: The Software Adapts, Not the Person
Traditional ERP software operates on a simple logic: it offers screens, menus, and processes, and people learn how to navigate them. Those who have worked with Abacus for 20 years know the program numbers by heart and navigate blindly. Those who are new need training, a manual, and patience.
Abacus Intelligence reverses this logic. Instead of people learning the language of the software, the software learns the language of people. Concretely, this means: you say or write what you want to achieve, and the system translates that into the necessary steps. “Record my working hours for today, 8 to 5, one hour lunch.” “What were sales in the third quarter compared to last year?” “Post yesterday’s laptop purchase.”
This is more than a gimmick. For many employees, the problem isn’t lack of competence, but the hurdle of finding the right function in powerful software. When this hurdle falls, it’s not the power users who benefit, but the occasional users. They are the majority in most SMEs.
What’s Already Running Productively Today
The most honest indicator that an AI function is more than a demo promise is named customer references with concrete numbers. For Abacus Intelligence, this is primarily the case in document recognition.
The DeepO technology, part of the DeepCloud ecosystem, recognizes incoming documents such as invoices and order confirmations, reads the relevant data, and prepares further processing. Abacus refers here to users such as the Rapperswil-Jona city administration, which according to its own statements processes a five-figure number of creditor documents per year via DeepO, as well as companies such as BKW Building Solutions and Transport AG Aarau. These are not demo scenarios, but production environments with real document volumes.
In addition, the following functions are available today and can be used in everyday SME operations:
| Function | What it delivers today | Maturity level in everyday SME use |
|---|---|---|
| Document recognition (DeepO) | Automatically recognize and read invoices and documents | Productive, with named references |
| Voice and chat control | Start processes via voice or chat, including Swiss German | Available, benefit depends on use case |
| Text generation | Keywords to flowing text, job postings, specifications | Practically usable, result requires review |
| Assistance with software questions | Chat help for using Abacus software itself | Useful, does not replace complex consulting |
| Data Analyzer | Generate KPIs and dashboards on request | Available, quality depends on data foundation |
| Actions and processes from chat | Trigger postings or requests directly in dialogue | Available, approval logic remains important |
| Regulation information | Answer questions about personnel regulations, home office, etc. | Useful, dependent on maintained content |
| AbaClik AI | AI chatbot in employee app, proactive notifications | Available, acceptance is the success factor |
Practical Example: Trading Company with High Document Volume
A trading company we advised on creditor processes processed around 600 incoming invoices per month before using AI-supported document recognition. The process was traditional: document by email or post, manual entry, account assignment, approval, payment. Pure data entry took around 35 working hours per month.
After introducing automated document recognition with subsequent account assignment logic, the picture changed significantly:
- Recognition rate of relevant fields for standard invoices: between 85 and 95%.
- Data entry effort: from around 35 hours to about 12 hours per month, with a large portion being review rather than input.
- Processing time per document until payment approval: from an average of 9 days to 3 days.
- Documents where AI had to pass (atypical layouts, poor scans, handwriting): around 10 to 15%, still manual.
The honest reading of these numbers: AI took over about two-thirds of the manual work, not everything. The remaining third hasn’t disappeared, but has shifted from pure data entry to reviewing AI suggestions and handling exceptions. Those expecting complete automation without human involvement will be disappointed. Those expecting significant relief with consistent control will get it.
Where the Honest Gap Between Vision and Reality Lies
The claim “Let software understand you” is a vision, and it’s a good one. But between a vision and productive everyday use, there’s always a distance. Three points we regularly address in conversations with SMEs:
AI is only as good as the data behind it. A Data Analyzer that delivers KPIs at the push of a button is impressive, as long as the data foundation is clean. If account assignments are inconsistent, cost centers have grown historically, and master data are incomplete, then even the best AI delivers questionable analyses—just faster and more convincingly presented. This is more dangerous than an obviously incorrect Excel cell because it looks more credible.
Responsibility remains with people. A posting triggered by voice command is still a posting with legal consequences. A job posting generated by AI can contain discriminatory language if no one reviews it. An employment reference automatically created remains a legally sensitive document. AI accelerates the draft; it doesn’t assume responsibility. This separation must be clearly anchored in processes.
Acceptance matters more than technology. A voice-controlled employee app is useless if employees don’t use it. From experience with self-service tools, we know: technology is rarely the problem; habit is. Those who introduce AI functions without bringing employees along have an expensive function that no one uses.
The Common Thread: Technology Complements People, It Doesn’t Replace Them
Abacus formulates its own position as technology and people complementing each other: you are competent in what you do every day, and AI enables you to become better at it. This aligns with a conviction we hold in every digitalization project: technology is only as good as the people who work with it.
Concretely, this means for Abacus Intelligence: voice control saves the accountant from searching for the right screen, but it doesn’t replace her professional judgment on whether a posting is correct. Text generation provides a draft for a job posting, but the decision about whom to hire and how to describe the company culture remains human. Document recognition processes hundreds of invoices, but the question of whether an unusually high payment to a new supplier is plausible belongs on a person’s desk.
Those who understand AI this way will win. Those who hope to save personnel through AI without thinking through processes will usually be disappointed and sometimes expensively surprised.
What Swiss SMEs Should Do Concretely Today
Abacus Intelligence is not a single product you buy, but a bundle of functions that are partly tied to modules and licenses. Four pragmatic steps for SMEs wanting to approach the topic:
Start with the document process. Automated document recognition is the AI function with the clearest and best-documented benefit. Those with a high volume of incoming invoices will find the fastest measurable leverage here. This is usually the most pragmatic entry point.
Data quality before AI. Before unleashing a Data Analyzer on your own numbers, an honest inventory of master data and account assignment logic is worthwhile. AI amplifies existing order and existing chaos equally.
Test small, don’t roll out big. Productively test a function in one team or one process, gather experience, then decide. A comprehensive rollout without prior testing is particularly risky with AI functions because acceptance is hard to predict.
Clarify responsibilities. Who reviews AI suggestions? Who approves? How are errors handled? These questions must be answered before production operation, not after. AI without clear control authority is not progress, but a risk.
Conclusion: An Honest Enthusiasm
Abacus Intelligence is one of the more substantial AI initiatives in the Swiss ERP market, not least because it builds on established software and has an area with DeepO that demonstrably runs productively. The vision “Let software understand you” is more than a slogan; it describes a real shift in the relationship between people and software.
At the same time, what applies to every new technology applies: value is not created by the function, but by how a company embeds it in its processes. Those who approach the topic with clear expectations, clean data, and a sense of remaining human responsibility can realize tangible added value today. Those who follow the hype and skip the homework produce expensive disappointments. The technology is ready. The question is whether the organization is too.
Would you like to know which AI functions make sense for your company today?
We’ll look at your processes, identify the areas with the clearest benefit, and tell you honestly where using Abacus Intelligence is worthwhile today and where it’s worth waiting. As an Abacus sales partner, we’ll accompany you from initial assessment to productive use.
