AI in your business: where to start and what you actually get

The managing director has read three articles about AI this week. Half the staff already use ChatGPT on personal accounts, a salesperson has tried a meeting-notes tool, and finance has received a quote for an AI solution nobody fully understands. Nobody knows whether it works, what it costs, or whether customer data ends up somewhere it should not. That is where most businesses stand in 2026, and it is a perfectly good starting point. It just needs a plan that starts small and measures.

Where should a business start with AI?

Start where staff today spend time reading, writing, sorting or moving text, and where a person can quickly tell whether the result is right. That is typically customer-service emails, quotes, minutes, product descriptions, receipts and internal questions about procedures. These tasks have three things in common: they happen many times a week, they follow a pattern, and a mistake gets caught before it does harm. That is the perfect first AI task.

The next choice is how much you should build yourselves. There are three levels, and most businesses should move up them in order.

Which AI use cases work in which departments?

This table is based on the tasks we most often see deliver a noticeable gain in small and medium-sized businesses. Difficulty is mainly about how many systems the AI has to talk to and how costly a mistake is.

What we see in practice

The projects that succeed have one thing in common: a staff member who knows the task from the inside and wants to get rid of the dull part. The projects that stall usually start with the technology and look for a problem afterwards. So we always recommend that the first meeting is about a concrete working day. How many emails? How long per quote? How many times a week is the same information copied from one system to another? The answers drive both the choice and the business case. Our guide to digitising manual processes has a template for exactly that mapping.

How do you choose your first AI project?

For the scoring in step two, use four questions worth 1–5 points each. The task with the most points is your first candidate.

• Volume: Does the task happen many times a week, so even a few minutes saved each time adds up?
• Pattern: Do the cases resemble each other, so one good instruction covers most of them?
• Data: Does what the AI needs live in one place you can give access to?
• Risk: Will a person catch a mistake before it reaches a customer or the books?
• Owner: Is there a staff member who wants to test it and has time in the next four weeks?

What data do you need, and is it ready?

AI is only as good as what it has to work with. A model answering customer questions needs access to current product data, prices and policies. A model drafting quotes needs the price list and the standard wording you use. In most businesses that knowledge is scattered across folders, inboxes and people’s heads. So the first piece of work is often to gather it in one place and bring it up to date. That work pays off even if you never build anything with AI, because new staff benefit from the same thing.

What does AI cost for a small or medium-sized business?

Costs fall into three buckets: licences for ready-made tools, development of what you build yourselves, and ongoing usage with the AI provider. Usage is typically billed by the amount of text processed, and for most small and medium-sized businesses it is a smaller item than the licences. Licence prices change often, so check current pricing with the providers.

A worked example: customer-service emails at a wholesaler

Picture a wholesaler with 25 staff, where three people in customer service each spend around 10 hours a week reading, sorting and answering emails about order status, delivery times and returns. A solution that sorts the emails, looks up the order and prepares a draft reply lets the staff member approve or edit the answer in a minute. If that saves half the time, it frees 15 hours a week. At an internal hourly cost of DKK 400, that is roughly DKK 300,000 a year in time. A solution at DKK 60,000 plus running and usage costs pays for itself within six months, even on a cautious assumption. Run your own numbers the same way before deciding anything.

What are the risks of AI, and how do you manage them?

• Wrong answers. Language models can phrase something wrong very convincingly. Have a person approve anything that goes to customers until you have data on how often it gets it right.
• Data leaks. Confidential data in free tools is the most common mistake. Fixed with business agreements and a clear policy.
• GDPR. AI solutions that process personal data need a data processing agreement, a purpose and a deletion plan, like any other system.
• Legal requirements. The EU AI Act sets special requirements for uses such as recruitment and decisions about people. Check with a lawyer if in doubt.
• Vendor lock-in. Build so the model can be swapped. The AI market moves fast, and today’s best model is rarely the best one a year from now.
• Rising usage costs. Set a monthly cap with the AI provider and track usage per task.

When the task grows from drafting to taking actions in your systems, such as creating orders or sending invoices, you move into AI agents. That needs more guardrails, and we have collected them in our guide to AI agents for business. If a customer-facing chatbot is what you have in mind, start with an AI chatbot for your website.

How do you measure whether AI pays off?

1. Measure the time spent on the task for two weeks before the trial, so you have a baseline.
2. Measure the same time during the trial, including time spent correcting the AI’s mistakes.
3. Count how many outputs were approved without edits.
4. Ask staff whether they want to continue and what annoys them.
5. Add up all costs: licences, development, usage and internal time.
6. Decide on the numbers, and repeat the measurement after three months.

Internal time is the item most often forgotten. An AI project needs someone to write instructions, test and give feedback, and that typically takes a few hours a week for the first months. Include it, and your business case becomes credible.

How does Ceptiv help with AI in your business?

We build AI into the workflows you already have, on the same principles as everything else we do: a fixed price, you own the code and data, and one senior team that both designs and develops. We have more than 40 pre-built integrations, including e-conomic, Dinero and MobilePay, so the AI can work with your real data from day one. See what we typically build under AI solutions and automation. If you already have a prototype from an AI tool, read from AI prototype to production. Otherwise, describe the task and get a fixed price within 24 hours.