Red flags in an AI software development company’s proposal: what to question before signing in Palma de Mallorca and the Balearic Islands
Published on

“So how long would it take you?”
“With AI, six weeks. Turnkey, everything included.”
“And what if we change something later?”
“Don’t worry, the AI adapts on its own.”
This dialogue was repeated to us, almost word for word, by the manager of a holiday-rental management company in Palma de Mallorca who came to us for a second opinion. On her desk was a proposal from an AI software development company: an attractive price, a fourteen-page PDF and plenty of pretty screenshots. There was nothing illegal in it. The important things simply weren’t written down.
A proposal is not just a price: it is the first sample of how that supplier will work with you. If you have already gone through how to choose an AI software development company and sent a one-page brief, the next step is to read what comes back with a magnifying glass. We won’t repeat selection criteria here. Instead, we show you how to spot the phrases that sound good and usually bring trouble, whether your business is in Palma, Ibiza, Menorca or anywhere else in the Balearic Islands.
Proposal translator: what they say versus what it usually means
Almost every problematic proposal we see repeats the same formulas. None of them is bad on its own; the problem appears when the phrase replaces a fact. Use this table as a translator and, above all, keep the third column: it is the question that forces the supplier to be specific.
| What the proposal says | What it usually means | The question that clarifies it |
|---|---|---|
| “The AI is trained on your data and learns by itself” | Nobody has defined which data is used, who cleans it or who reviews the mistakes. | Exactly which data will you use, where is it hosted and who validates the answers before they reach a customer? |
| “Fixed price, everything included” | Vague scope: every change will be an extra, or quality will be cut to make the price work. | What is excluded, how many rounds of changes are included and what is the rate for extra hours? |
| “Delivery in 6 weeks”, with no phases | The schedule was worked backwards from the sale, not forwards from the work. | What will I see working in week 2 and in week 4? |
| “We use the most advanced AI models” | Usage costs have not been estimated and you will end up paying the API bill. | How much will the AI cost per month at my real volume, in high season and in low season? |
| “We work with our own platform” | Lock-in: the code or the data may not be yours, or you may not be able to take them with you. | Who owns the code, the repository and the data on the day we stop working together? |
| “Support included” | An email address with no response time. | What response time do you guarantee on a Saturday in August? |
| “99% accuracy” | A demo figure, measured with hand-picked examples. | What data was it measured on, and how will we measure it with real cases? |
| “Integration with your systems” | It assumes your PMS, ERP or CRM has a usable API, without having checked. | Have you already reviewed our system and its API? What happens if it has none or it is limited? |
Traffic light: when it is amber and when it is red
Not every red flag justifies dropping a supplier. Many proposals are written in a hurry, from templates, and a vague phrase does not always hide bad faith. What separates a warning from a real problem is how the company reacts when you ask it to be specific.
Amber is an imprecise phrase that can be fixed with a written answer. “Support included” is fixed with an annex stating hours and response times; “fixed price” is fixed with a list of exclusions and a rate for changes. If the supplier adds it without complaint, the flag is cleared.
Red is when the supplier refuses to be specific or is annoyed that you asked. The four most common red flags we see in the Balearic Islands are: not clarifying who owns the code, not letting you export your data, not estimating the monthly AI cost, and not showing you anything working until the end of the project.
A rule of thumb we use with our clients: three ambers without a written answer equal one red. Not because each one is serious, but because together they say a lot about how problems will be handled once the project is under way.
A real proposal, annotated line by line
To see it in practice, here are anonymised excerpts from a proposal we reviewed for a yacht-charter company with bases in Palma, Ibiza and Menorca. They wanted an AI assistant that would answer booking enquiries in four languages and coordinate skippers and cleaning crews across the three islands.
“Single phase: development and roll-out of the multichannel AI assistant. Timeline: 9 weeks.”
Our note: a single phase means seeing nothing useful until week 9, right when the season starts. We proposed splitting it: a prototype in week 3 using real questions from the previous summer, followed by a period in shadow mode, where the assistant drafts answers but a person sends them.
“The client will provide the information required for training.”
Our note: this sentence shifts all the risk onto the client. Which information, in what format and by when? In this case, the Ibiza and Menorca bases kept their rates in different spreadsheets, with different boat names. Unifying them was two weeks of work nobody had budgeted for.
“Hosting and maintenance: on request.”
Our note: “on request” in a proposal usually means a surprise on the invoice. With three bases and peaks in July and August, the volume of AI queries was five times higher than in March. We asked for a monthly estimate with an August peak scenario and a low-season one.
“The supplier retains the intellectual property of the platform and grants the client a licence to use it.”
Our note: red, as worded. It is reasonable for the generic components the supplier reuses to remain theirs, but the project-specific code, the prompts, the configuration and the data should belong to the client, with a documented export procedure.
Outcome: the supplier accepted three of the four changes and negotiated the fourth. The client signed with phases, a written monthly cost estimate and an exit clause. The final price rose by 11%, but every “on request” and open-ended item disappeared. That 11% turned out far cheaper than an August without the assistant.
Frequently asked questions the proposal should answer (and almost never does)
Who owns the code, the prompts and the data?
You do, except for the supplier’s generic components, and it must be in writing. Ask for repository access from day one, not at the end. If the answer is “we’ll talk about it later”, that is a red flag.
How much will the AI cost each month, on top of development?
A serious proposal separates the cost of building from the cost of running: hosting, AI model usage, monitoring and maintenance. For seasonal businesses in Mallorca, Ibiza or Menorca, insist on two figures: high season and low season.
How will we know the AI is working well?
With metrics agreed before starting and measured on your real cases: percentage of correct answers, enquiries handed over to a person, and time saved per week. A “99% accuracy” claim with no method is worthless.
Where is the data hosted and how is GDPR complied with?
The proposal should state where the data is stored, which AI providers process it, whether it is used to train third-party models and who signs the data processing agreement. If you handle guest or patient data, this is not optional.
What happens if we want to switch suppliers?
There should be an exit plan: handover of the code, documentation, data export and a transition period. Nobody signs thinking about breaking up, but the suppliers who have this section clear are usually the ones who need it least.
How to ask for changes without derailing the negotiation
Spotting red flags doesn’t mean picking a fight. The most effective approach is to send the proposal back with specific questions and a deadline. You can copy and adapt this message:
“Hi, thanks for the proposal. Before we decide, we need four points in writing: (1) what we will see working in weeks 2 and 4; (2) an estimate of the monthly AI and hosting cost at our enquiry volume in high season and in low season; (3) confirmation that the project-specific code, the prompts and our data belong to us, with a documented export; (4) support response times in July and August. With that, we’ll compare and get back to you this week.”
A good supplier replies within a couple of days, without drama. If the reply is vague or irritated, you already have your answer. At Derek Solutions we include these four points by default in our custom software development in Palma proposals, and we apply them just the same when we work with businesses that need app development in Ibiza or in Menorca.
Already have a proposal on the table? Get a second opinion
You don’t have to hire us for anything. Send us the proposal you’ve received, even if it is from another company, and we’ll send it back marked with the traffic light: green, amber or red for each section, with the exact question you should ask on every doubtful point. We do this for businesses in Palma de Mallorca, Ibiza, Menorca and the rest of the Balearic Islands, usually within 48 working hours and with no commitment.
Write to derek@dereksolutions.com · +34 680 283 973 and tell us in one line what you want to build and which island you work on.