SAP B1 blog

Just one more thing to learn, then we can get started.

There is still one thing that you would have to learn beforehand. But then! You can guess it: This phase doesn't end. On the other hand, AI changes too quickly. And it fits on too many different types of jobs, so that no one from the outside could tell you which application is right for you.

That depends on three things: what you’re good at; what annoys you; and what a good result looks like in your world. Nobody knows that better than you do.

Ultimately, AI requires two ingredients for everyday work: your expertise and a working knowledge of AI. You already have the first ingredient. The second ingredient cannot be obtained through reading alone; it must be acquired through practice. Even when some aspects of the tools are still unfamiliar.

The family outing

Let’s take a family outing. The AI suggests ten destinations in just a few seconds. Impressive. Unfortunately, the AI doesn’t know that no one wants to get up early on Sunday. It also doesn’t know that one of the family members can’t walk very far. And especially not that the lunch must be on the table before the children get hungry. Otherwise, the mood is set, and that lasts for the rest of the day.

It is these details that make any daily schedule something that the family actually uses. And it is these details that make you immediately notice when the AI is off-kilter. The AI provides the possibilities. It provides the judgment.

With AI in everyday work life, it is no different.

It’s the same way at work. You know why the usual response to customers sounds so cold. You know which figure in the monthly report everyone looks at first. And you know what the boss asks as soon as he sees the first slide. This knowledge is not in any manual or language model.

AI can do thousands of things. That’s why meaningful applications look different for everyone. Your experience helps you find the one thing among the thousands of things that matters to you.

Every attempt shows the next one.

Back to the trip. The first plan is obviously too full. So tell the AI what to include and why. The second version is better. In the meantime, you have learned something: the AI can compare options, take multiple conditions into account simultaneously, and revise a plan without starting from scratch. Next time you go on a trip, you might even attach the booking confirmations and plan around that day accordingly.

So each attempt shows four things:

• what the tool can do,
• what it needs from you,
• where it still makes mistakes,
• and what you could try next.

With new models and features, it’s similar. They only become understandable when you have something of your own to test. An old task for a new model. An image on which you have previously failed. Or a small task for an agent.

Agents: let me just read it first

Speaking of agents. Many people put the topic off because they think they need to fully understand it first. You don’t have to. Give the agent a single folder and a first task in which they can only read: Tell me what’s in it, what tasks you could take on and what you need from me for that.

While the agent is working, very specific questions arise from the unknown parts. And for concrete questions there are concrete answers. Those who work with an ERP system are familiar with the principle, by the way: First, let them look at things, then gradually trust more. More about this can be found in the article AI in ERP, but under control.

Prompt 1: Where does AI help in your daily work?

Ask an AI: „What can AI do for me as a bookkeeper?“ Then you get the usual list: compiling documents, designing emails, collecting ideas. Thanks too. That was already in the last twenty articles on the topic.

It works better the other way around. Tell the AI about your day at work yesterday and let it look for details. This prompt helps with that:

Help me find a meaningful AI experiment for my last working day.

Interview me, always one question at a time. Please let me go through the day from the first to the last task first. For each task, find out: what came in and what I made of it; what I had to look for, compare, decide on, or correct; what took longer than expected or remained unresolved.

After that, show me three moments that are worth testing with AI. Use only the work I mentioned. Prefer tasks that have been repeated, gathered information, required a judgment that I can explain, or created something that I can test.

Let me choose one of them, request the real material, and help me finish the smallest version that is usable right now. In the end, I want to have something that I can judge, not just a proposal.

An example: When onboarding new colleagues, one might notice that they keep explaining the same customer situation over and over again. This can be used to build a practice exercise that includes the current policy and the places where newcomers regularly get stuck.

For AI in everyday work with SAP Business One, this means: Don’t ask what AI can do for accounting. Instead, tell the story of yesterday’s day. Which incoming invoices came in? Which query about open items had to be answered? Which analysis took longer than planned? It is from these very details that the prompt identifies its three candidates.

Anyone who regularly works with AI can make it more efficient. If the AI already knows your patterns from its memory or the course of the conversation, simply ask directly: What should I no longer do manually?

Prompt 2: The tool becomes self-explanatory during operation

The second prompt flips the usual order. First learn, then use? No. You open the model, the agent, or the function you don’t yet understand, and together you complete something. The explanation comes when the work needs it. Not a minute sooner.

I am using [model, agent, function, or tool] for the first time. Help me learn by completing something.

Ask me first: „What is one thing that you can or can judge that this tool should help you create, investigate, change, or organize today?“

Always ask only one question until the task is clear. Choose the smallest version that uses a skill that precisely identifies this tool. Tell me what we will end up with, and lead me step by step.

Explain unfamiliar terms only when we encounter them. Do what you can do immediately yourself. If I have to click, add something, or select something, tell me exactly what needs to be done. Pause before anything is sent, published, purchased, deleted, or access rights changed, unless I explicitly ask you to do so.

Finally, give me the result and three notes: what the tool did well, where my judgment was still needed, and what I can do well now that I can try it out next.

Pay attention to the built-in brake. The tool has many functions, but not everything. Sending, publishing, buying, deleting, and changing permissions remain the responsibility of the human being. Everyone who works with an ERP system knows this: the system can make suggestions. Only someone who is prepared to accept them can make a release.

And if a new announcement has sparked your interest? Then first check whether the news is relevant to your work at all. If the answer is yes, include exactly the one appropriate skill in this prompt. Leave the rest of the tool for now.

AI in everyday work: Work says what comes next

Admittedly, a blog post about AI that recommends reading less about AI is somewhat ironic. Of course, new models and tools need translation. But the best translation is still the result of your own efforts.

The right time to look for an answer is when you have a specific question. Until then, use what you have already learned. The result will then show you what is missing: attach a file, sharpen an instruction, or choose a different tool. And already the next learning round has a purpose.

Therefore, your expertise is not an obstacle to AI in everyday work situations. It is precisely the part that no AI can provide.