Five questions about data security and maintenance that save you from surprises later.
When you buy AI automation for the first time, the conversation easily focuses on what the automation can do. That matters, but what happens after rollout matters just as much.
In finance, automation usually handles confidential data such as invoices, salaries and tax information. It also runs as part of systems that change over time. That is why these five questions are worth going through before the contract is signed.
1. Where is our data processed and stored?
Ask which servers the automation runs on and in which country the data is stored. When personal data is involved, processing has to comply with the GDPR, and it is wise to keep the data inside the EU.
Also ask for a written description of what data the automation processes and how. A good vendor can describe this clearly before the automation is taken into use. If your processes are already discussed in the first meeting, a non disclosure agreement can be signed before it.
Our automations run on Google's servers in Finland and the data is stored encrypted. Customers also receive a data processing report that describes what data the automation processes and how.
2. Is our data used to train an AI model?
Many automations use a language model provided by another company. In that case it is worth finding out which model is used, who provides it and what happens to your data when it is sent to the model.
Ask directly whether your data can be used to train the model and whether it is stored on the model provider's side. The answer should be clear, and it should also be in the contract.
3. Who notices if the automation stops working?
An automation isn't finished when it goes live. Software gets updated and APIs change, so at some point a run will fail.
The most important thing is that the error doesn't go unnoticed. Ask whether every run is monitored, who gets alerted about an error and how quickly it is fixed. If the answer is that the customer will notice it themselves, the risk stays with you.
In the automations of the Pesonen accounting firm, for example, every run is logged in monitoring. If any step fails, an alert goes out immediately and we fix the problem.
4. What does maintenance cost and what does it include?
The price usually consists of a one time build fee and a monthly maintenance fee. The build fee alone doesn't tell you the total cost, so ask for both in writing.
Also ask what maintenance includes. Are bug fixes included in the price, and how much do small changes cost? At Pesonen, adding a new client company is one row in a Google Sheets spreadsheet and requires no code changes.
Finally, ask for a payback calculation. If the vendor can't estimate how much the automation saves, it is hard to know whether it is worth it.
5. What happens if the partnership ends?
This is rarely asked, but it is worth agreeing in advance. Ask who owns the automation and its documentation, and how your data will be deleted or returned to you if the contract ends.
A clear answer also tells you how the vendor approaches the partnership. When things are agreed in advance, they don't have to be sorted out afterwards.
You can go through these questions with us too. In a free AI roadmap we work out which of your processes are worth automating, and we can sign a non disclosure agreement before the interview. More on how we process data is in our privacy policy.