How to get started with AI in your business: four steps from current state to production

The best way to get started with AI in your business is to start with the operations, not the tools. First assess where you stand today, then choose one to three well-defined use cases where AI saves time or reduces errors, set guidelines for how AI may be used and build it into the systems you already have, in short steps. This guide walks through the four steps and the mistakes that most often leave AI work stuck in pilots.

Short answer

• Step 1: Assess the current state. How work flows, what data exists, which systems AI needs to work in and who owns the question.
• Step 2: Choose the right use cases. Start where the work is recurring, time-consuming and based on information that is already digital.
• Step 3: Set guidelines and ownership. Decide what employees may use AI for, which data may be used and who owns each solution.
• Step 4: Build it into existing systems. Build AI into the workflow you already have, test it with the people who do the work and measure the impact.

Where do you stand today?

Most companies have already started with AI in some way. The next step depends on where you are. These are the four stages we see most often:

Step 1: Assess the current state

Before you choose tools, you need an honest picture of where you are starting from. Review four dimensions:

• Process. How does work flow today? Where does time go, and where do errors or waiting occur? Map the workflows that involve the most people, such as orders, purchasing, customer tickets or reporting.
• Data. What information exists, where is it and in what condition? AI can't do more than the data it gets. If the most important information lives in emails, PDFs and spreadsheets, that is an important insight in itself.
• Systems. Which systems does the work happen in, such as monday.com, Visma Net, Fortnox or your e-commerce platform? Can they be connected via API? That determines how AI can be built in.
• Organization. Who owns the AI question? What skills exist? What do the people who will use the solutions think?

The result should be a list of where AI could add value, and a picture of what is missing for it to work.

Step 2: Choose the right use cases

The most common mistake is starting too big, or with whatever sounds most impressive. A good first use case is small, well-defined and easy to measure.

Typical first use cases are interpreting incoming documents and entering the data into the right system, sorting and answering incoming tickets, summarizing customer history before meetings and producing recurring reports. For SKG Storköksgrossisten, a Swedish foodservice wholesaler, we built AI-driven document scanning, where incoming documents are interpreted automatically and the data lands directly in monday.com, with no manual entry. Read the case.

Evaluate each candidate with three questions: how big is the impact, how big is the risk if AI gets it wrong, and how easy is it to implement? Start with what scores high on impact and low on risk.

Step 3: Set guidelines and ownership

AI work without guidelines leads either to no one daring to use AI, or to everyone using it in different ways. Decide early:

• What AI may be used for, and what it may not be used for.
• Which data may be used in which tools. Personal data and business-critical information require special handling under GDPR and your own requirements.
• Who owns each solution, and who follows up that it works.
• Where a person needs to approve before the AI does something, for example before an invoice is posted or an email is sent to a customer.
• How employees are trained. The EU AI Act also requires that staff who use AI systems have sufficient knowledge of them.

The guidelines don't need to be long. One page that everyone understands is better than a document no one reads.

Step 4: Build AI into the systems you already have

AI does the most good where the work already happens. An AI solution that sits next to your regular systems often becomes one more tool to keep track of. Instead, build it into the workflow, for example as an AI agent that reads incoming emails and updates the ticket in monday.com, or an automation that interprets supplier invoices before they are approved in Visma Net.

• Measure before and after. Note how long the work takes, how often errors occur and how long lead times are before you start. Otherwise you can't show the impact.
• Build in short steps. Get something into production within weeks, test it with the people who do the work and adjust.
• Keep a person in the loop where a mistake would be costly, and step back gradually once the solution has proven it works.
• Plan for ongoing management. The business changes, and AI solutions need to change with it.

Common mistakes

• Starting with the tool. "We bought licenses, what should we use them for?" rarely leads in the right direction.
• Too many pilots at once. It spreads the effort thin, and nothing makes it all the way into operations.
• No owner. A solution without an owner stops being used when the first person who championed it moves on.
• No measurement. Without numbers from before the rollout, you can't know whether it paid off.
• Forgetting the people who do the work. They know where things actually get stuck, and without them the solution won't be used.

When do you need outside help?

Many companies can take steps 1 to 3 on their own, especially if someone internally has the time and interest. Outside help adds the most value when you have tested but not gotten anything into production, when AI needs to be built into several systems or when no one internally has time to drive the work. That is when an AI consultant can do the current-state assessment, create the roadmap and build the first solutions together with you.

Frequently asked questions

Where should a company start with AI?

Start with the operations, not the tools. Assess the current state across four dimensions (process, data, systems and organization), choose one to three well-defined use cases where AI saves time or reduces errors, set guidelines and then build AI into the systems you already have.

Do we need an AI strategy before we start?

You don't need a long strategy, but you do need a direction: which use cases you prioritize, which guidelines apply and who owns the work. A simple roadmap and a one-page set of guidelines are often enough to get started the right way.

Which AI tools should we use?

It depends on what you want to do. General AI assistants work well for individual employees' writing and analysis. For AI to take over work in your workflows, it needs to be built into the systems you already use, for example as an AI agent or automation in monday.com or Visma Net.

How long does it take to get started with AI?

A current-state assessment and a prioritized roadmap can be done in about a week. A first well-defined use case can often be in production within a few weeks, if the data and systems can be connected.

What does it cost to get started with AI?

It depends mainly on how much needs to be built and how many systems are involved. Assessing the current state and choosing use cases mostly costs time. If you build in platforms you already have, the cost is mainly in the work, not in new licenses.

What is an AI agent?

An AI agent performs well-defined work in your systems, such as handling incoming tickets, interpreting documents or updating records, with clear permissions, a log of every action and human checkpoints where needed. Read more about AI agents for businesses.

Next steps

Start by answering two questions: where does time go today, and what information is already digital? The answers usually point to the first use case. If you want help with the current-state assessment, you can book a free mapping session.