AI tools for business: which ones do you really need?

The first question isn't 'which tool?', but 'where does our time go?'. That's the first principle of our APO method: process first, technology later. Turn that order around and you end up with solutions looking for a problem.

Take a process that comes back often, such as processing an order, drafting a quote or answering a customer question. Cut it into steps and give each step three numbers:

• Duration. How long does the step take?
• Number of people. How many people do it?
• Frequency. How often per month does it happen?

A forty-minute task, done by two people, eighty times a month: that's 1,280 hours a year. Now you know what's at stake. And now you can judge which tool, or which build, is worth it.

Often it turns out the time isn't spent on writing, but between systems: retyping, looking things up, forwarding, waiting for each other. A chat tool only helps a little there. A workflow or agent does help.

How to find those processes is covered in which processes to automate.

A chat assistant is the AI tool everyone knows. You type a question or instruction and get an answer. Writing, summarizing, translating, digging into an Excel export, rewriting a tricky email. It works for almost every role, which is why it belongs in the basics.

The well-known names are ChatGPT from OpenAI, Microsoft Copilot, Claude from Anthropic and Google Gemini, which is part of Google Workspace. They look alike, but differ in where they fit in. If your team works in Outlook, Word and Teams all day, Copilot is the obvious choice. If you work in Google Workspace, look at Gemini first. If you work with lots of separate tools, ChatGPT or Claude often fit better.

• Fits: everyone who writes, reads and analyzes. The gain is personal: every employee gets a little faster.
• Watch out: use a business version with a data processing agreement, not the free app. And pick one tool as the standard, not four.
• The limit: a chat assistant doesn't do anything by itself. Someone opens the chat, pastes, reads, copies and pastes back. The work itself doesn't change.

Plans, prices and where your data is stored are compared in ChatGPT, Copilot or Claude?

Do you keep pasting the same explanation, the same price list and the same example into the chat? Then it's time for a custom assistant. You put the instructions and documents in once, and after that nobody has to paste them again.

You can do this with a custom GPT in ChatGPT, a Claude Project or a Copilot agent in Microsoft 365. Think of an assistant that drafts quotes in your template, writes copy in your house style or answers questions about the staff handbook.

A step further is AI that answers from your own documents, with the source included. That's called RAG. The AI first looks up the right passages in your handbooks, contracts or product information, and only then writes an answer. That way you get answers based on your knowledge, instead of on what the model once read on the internet.

• Fits: recurring work with fixed context, and questions that keep coming from the same documents.
• Watch out: an assistant is only as good as the documents in it. Appoint someone to keep them up to date, and check who's allowed to see which information.
• Build it yourself or have it built: your team makes an assistant itself. Searching across thousands of documents, with permissions per user, is a build project.

Read more in build your own AI assistant and what is RAG?

A workflow takes over a process that runs between systems. Something happens, for example an email with an order comes in, and then the steps follow by themselves. Reading the PDF, looking up the customer in the ERP, preparing the order, having a colleague approve it. No more retyping.

AI can sit in every step. A language model reads the attachment, sorts the email or writes a draft reply. That's the difference with the chat: nobody has to open or paste anything. The work just keeps moving.

Well-known tools are n8n, Make and Zapier. Zapier is easy to get started with, Make is visual and n8n is flexible and can also run on your own server. They charge differently: per step, per module or per run of the whole workflow. With many steps and a lot of volume, that makes a real difference.

• Fits: work that comes back dozens of times a week, where someone moves data from one system to another.
• Watch out: who owns it, what happens when a system is briefly down, and where does your data go? A workflow nobody maintains will break at some point.
• Build it yourself or have it built: you can build a simple connection yourself. A core process, such as orders, invoices or customer questions, you have built and maintained.

Which tool fits you is covered in n8n vs Make vs Zapier. What we build with it is on n8n automation and AI automation.

An AI agent gets a goal and works out by itself which steps it needs. It reads a customer question, looks up the order in the ERP, checks the delivery time and prepares a reply. An employee approves and sends it.

Think of it as a digital colleague who reads fast, never gets tired and does exactly what the work instruction says. But also as a colleague in their first week: you don't hand them the keys to the safe straight away.

The difference with a workflow: a workflow follows fixed steps, an agent chooses by itself, within limits you set. That makes an agent a good fit for work with a lot of variation, where fixed rules fall short.

An agent needs access to your systems. That can be through a workflow, or through MCP: a kind of universal plug between AI and your software.

• Fits: work where someone gathers information from several systems and prepares a decision.
• Watch out: give an agent only the permissions it needs, have it log what it does, and put a human in the places where a mistake costs money or affects a customer.
• Build it yourself or have it built: agents in your core processes you have built, with someone who maintains them.

More in what is an AI agent? What we build ourselves is on AI agents.

Look at what you already have before you buy something new. More and more software comes with AI features built in. Your CRM, your ERP, your accounting package, your helpdesk and your meeting tool may already have a few, or will get them soon.

The advantage: no new tool, no new login, and the data stays where it already was. Sometimes it's included in your current license, sometimes in a more expensive plan or as a paid add-on.

• Fits: tasks that happen entirely within one system, such as a summary of a customer file or the minutes of a meeting.
• Watch out: check where the data is processed, whether the feature is on by default and what it costs extra. And expect such a feature to often work only within that one system.
• Build it yourself or have it built: you switch it on and set it up yourself, together with your supplier. If the process runs across several systems, you need a workflow or agent.

Simply ask your suppliers what's included in your current license. It costs nothing and keeps you from paying twice for the same thing.

A chat assistant makes people faster. A workflow or agent takes work over. That's the difference, and that's where the gain is.

With only a chat tool, the work doesn't change. Someone opens the chat, pastes, reads and retypes the result into another system. Only when AI is connected to your own systems, such as your mailbox, ERP or CRM, does the work keep moving without retyping. Also read connecting AI to your systems.

In our APO method, every application gets one of four build layers:

1. 01Prompts and Knowledge. Working better with the chat assistant. Your team does it themselves with AI, after training.
2. 02Skills and Projects. Custom assistants with your knowledge and way of working. Your team does this themselves with AI too.
3. 03n8n Workflow Automation. Processes that run by themselves, connected to your systems. NinA builds these.
4. 04Custom AI Build. Agents, search across your own documents and custom work for what no standard tool can do. NinA builds these too.

That's why a healthy setup often looks like this:

• One business chat assistant for everyone, with training and ground rules.
• A few custom assistants for recurring work, made by the team itself.
• One workflow platform for the processes that take up a lot of hours.
• Agents where the work has so much variation that fixed steps aren't enough.
• The AI in your existing software, where it's already there.

That's a handful of tools, not twenty. Every extra tool is an extra license, an extra place where your data lives and an extra thing to manage.

A demo always looks good. Ask these six questions before you sign.

1. 01Where does your data go? Choose a business plan with a data processing agreement, where the provider doesn't train on your data. Ask where the data is stored and processed: in the EU, in the US or in your own environment. More in using ChatGPT safely (GDPR) and using AI safely.
2. 02What are your people already using? First look at what's already running. Colleagues may already be working with personal accounts and free versions. That's called shadow AI: AI use the organization has no view of. A ban doesn't solve it. A good business alternative and clear agreements do.
3. 03Does it connect to your systems? A tool that can't reach your ERP, CRM, mailbox or documents stays an island. Ask which integrations exist and how permissions are handled. Read connecting AI to your systems.
4. 04Who manages it? Who owns it, who handles the accounts, who keeps the knowledge up to date and who gets the call when it breaks? Without a name next to each question, you don't have a tool, you have a risk.
5. 05Can your people work with it? A license without training is an expensive notepad. On top of that, since 2 February 2025 the AI Act has asked organizations that use AI to work on the AI literacy of their people. Read more about AI literacy.
6. 06Do you pay per user or per use? Chat assistants usually charge per user per month. Workflow tools and AI models charge per use: per task, per run or per token. The first grows with your team, the second with your volume. Work it out on your own numbers, and see what AI costs.

It rarely goes wrong on the technology. It goes wrong on the choices around it.

• Starting with the demo. Someone sees an impressive tool and then goes looking for a problem to fit it. Turn it around: first the process, then the tool.
• Too many separate tools. Every team picks its own tool. After a year your data sits in ten places, you pay twice and nobody knows what's running anymore.
• Licenses nobody uses. Everyone gets a license, nobody gets training. After three months, look at who really uses the tool, and for what.
• Nobody owns it. A colleague builds a handy connection with their own account. They leave, the account expires and suddenly something stops that nobody knew existed.

Why it goes wrong and how to prevent it is covered in why AI projects fail.

You don't have to figure this out alone. We always start with the work, and only then with the tools.

1. 01The AI Quick Scan. One day on site with you. We talk to management and the team, look over people's shoulders on the the work being done and map the processes. Within a week you have a business case per opportunity, a roadmap and advice on which package fits. € 6,000 one-off. More on AI Quick Scan.
2. 02The AI Partnership. A dedicated team builds the workflows and agents, connected to your systems, and trains your people. Package S (€ 4,500 per month, 30 hours), M (€ 7,500, 50 hours) or L (€ 12,000, 80 hours), for at least twelve months. With a guarantee: at least 2x back in the first year. More on AI Partnership.
3. 03Keeping it running. Maintenance and monitoring in a contract of € 245, € 445 or € 895 per month, including AI usage.

All amounts exclude VAT. Which chat assistant, which workflow platform and which model: that follows from the business case, not from a preference.