After launch: what maintenance to agree with your AI software development company in Mallorca and the Balearics

Most conversations with an AI software development company are about the project itself: what gets built, how long it takes and what it costs. But with software that relies on artificial intelligence, launch day isn't the finish line. It's the day it starts dealing with real data, real customers and real mistakes.

Traditional software that nobody touches tends to keep doing the same thing. An AI system can get worse without a single line of code changing: the documents it receives change, the model provider updates something, customers start asking different questions, or volume simply grows. That's why maintenance is worth agreeing before you sign, not when the first problem shows up.

This article covers what that agreement should include, how to tell whether it's being honoured and which questions to ask, with businesses in Mallorca, Ibiza, Menorca and the rest of the Balearic Islands in mind.

A hypothetical example

Picture a hypothetical accounting firm in Palma that commissions a tool to read the invoices its clients email in. The tool pulls out supplier, date, net amount and VAT, and queues each invoice for a person to check and book.

For the first few months it works well. Then high season arrives: far more invoices from hospitality clients, receipts photographed on a phone, invoices in English and German from foreign suppliers. Nobody has changed the code, yet the tool starts making more mistakes. A good maintenance agreement is there precisely for this.

Five things that wear out in AI software

1. Answer quality

This matters most and gets measured least. Ask for the project to include its own test set: a collection of real cases (anonymised if needed) with the correct answer for each. For the accounting firm, that could be a few dozen varied invoices with the data that should be extracted.

That set lets you check, regularly and after every change, whether the system is getting more or fewer cases right than before. Without it, "it works fine" is just an opinion.

2. The AI model

Most tools run on third-party models. Those models get updated, repriced or retired. The agreement should make clear:

• Who keeps an eye on the provider's notices about versions being withdrawn.
• That the test set is run, and you're shown the results, before any model switch.
• Whether the software can change provider without being rebuilt.

3. Usage costs

On top of maintenance, AI software has a variable cost: every document or conversation processed uses up credit. For a seasonal business, and the Balearics are full of them, that usage can jump in summer. Ask for:

• A monthly dashboard or report showing usage and what it cost.
• Limits and alerts, so a spike or a bug doesn't turn into a surprise bill.
• A clear answer on who pays for usage and whose name the provider account is in.

4. Data and instructions

The instructions given to the model, the examples it uses and the reference material (price lists, catalogues, policies) all change as the business changes. Decide who can update them and how. If your team can edit part of that content from an admin panel, you won't need the developers for every small tweak.

5. Integrations

AI software rarely works alone: it connects to email, your CRM, accounting, WhatsApp or your website. When one of those changes how it connects, the integration breaks. The agreement should say whether fixing that is covered or quoted separately.

What the maintenance agreement should cover

You don't need a fifty-page contract. It just needs to be specific about:

1. Regular quality reviews. How often the test set is run and who sends you the results.
2. An issue log. Where your team reports problems and how quickly they get a response, separating "the tool is down" from "this invoice was read wrongly".
3. What's in and what's out. What counts as maintenance (adjusting instructions, updating the model, fixing an integration) and what is a new feature to be quoted.
4. Cost control. Usage limits, alerts and who receives them.
5. Security and data. Where documents are stored, for how long, and what gets sent to the model provider.
6. Exit. What you get if you stop working with that company: code, instructions, test set and documentation.

How to tell the maintenance is actually happening

An agreement nobody follows up on becomes a fee paid out of habit. A few simple signs it's working:

• You get a short report on a regular basis, even if it only says nothing has changed.
• When something in the business changes (a price, a new language, a new type of customer), there's a clear way to feed it into the system.
• Errors your team reports become new cases in the test set, so they don't come back.
• Before any model change, you're shown how it compares with the old one.

For the accounting firm in our example, the fix for the high-season problem wouldn't be "rewrite everything". It would be adding phone receipts and foreign-language invoices to the test set, adjusting the instructions and confirming with numbers that accuracy is back where it was.

Questions to ask before you sign

If you're comparing proposals, these questions separate companies quite clearly:

• How will you show me the AI is still working well in six months?
• What happens if the model provider retires the version we use?
• How will I see what I'm spending on AI each month?
• What can I change myself without calling you?
• What do I take with me if I switch provider tomorrow?

Vague answers here usually mean vague maintenance later.

How much maintenance do you need?

Not every project needs the same level. As a rough guide:

• Internal use, low volume (an assistant drafting texts that a person always reviews): occasional reviews and on-demand support may be enough.
• Daily use in a business process (invoices, quotes, bookings): regular quality reviews, cost control and a fast response to errors.
• Direct contact with customers (a chatbot or automatic replies in several languages): all of the above, plus monitoring conversations and a well-defined human handover, especially in high season.

How we approach it at Derek Solutions

When we build AI-powered software for businesses in Mallorca and the Balearics, maintenance is part of the conversation from day one: what we measure, how we keep usage costs under control and what stays in your name. You can see how we work on our AI automation in Mallorca page and, for larger projects, on custom software development in Palma.

Already running an AI tool, or about to commission one? Email derek@dereksolutions.com or call +34 680 283 973 and we'll go through what kind of maintenance makes sense for you.