Why don't business owners buy AI, and what do they actually buy?

Business owners don't buy AI because AI is not an outcome: they buy more leads answered, less rework, better customer service, reliable financial reports and less mess in the operation. AI only comes in when it is the simplest piece that delivers one of those outcomes, and the project starts with the problem, not the tool.

Have you ever received a proposal to "implement artificial intelligence in your company" and finished reading it without knowing what would change on Monday? That is the point of this article. When the offer talks about the technology and not the problem, the person paying the bill is left with no criteria to decide.

In The Hitchhiker's Guide to the Galaxy, by Douglas Adams, a group of hyper-intelligent beings builds the supercomputer Deep Thought to get the answer to life, the universe and everything. After seven and a half million years of computing, the answer is 42. The computer itself explains that it seems meaningless because those who ordered the calculation never knew what the question was.

An AI project without a question ends the same way. The company hires a chatbot, a dashboard or an agent, receives something that works, and nobody can say whether it was worth it, because no one ever agreed on what was supposed to improve. The system answers. Nobody asked the question.

Clayton Christensen, Scott Cook and Taddy Hall recall, in a 2005 Harvard Business Review article, the line Professor Theodore Levitt used to repeat to his students at Harvard: "People don't want to buy a quarter-inch drill. They want a quarter-inch hole!"

The article's argument goes beyond the line. According to the authors, when people need to get a job done, they "hire" a product to do it. The best-known example in the text is a fast-food restaurant that, instead of asking customers how to improve its milk shake, sent a researcher to spend the day in the restaurant to understand what job people were trying to get done when they bought one.

Swap the drill for AI and the logic is the same. Nobody wakes up wanting a language model. People want the patient who sent a message at 10 p.m. to get an answer, the ERP spreadsheet to become a number they understand, the salesperson to stop copying data from one system to another.

At DOUVRAS, we start from a short list. Business owners don't buy AI. They buy one of these five outcomes.

• More leads. Visitors who become conversations, contacts that don't get lost and follow-ups that happen on the right date. This is where websites and landing pages connected to WhatsApp and a simple CRM come in.
• Less rework. Information stops crossing the company through manual copying, memory and scattered messages. That is the territory of automation and integrations.
• Better customer service. Fast answers to repeated questions, with clear limits and a handoff to a person when the case calls for it. This is where an AI agent or assistant usually makes sense.
• Financial reports. Knowing the month's real profit without building a spreadsheet by hand, as we showed in the article on an automatic income statement from the ERP.
• Less mess. An internal system, dashboard or portal that organizes what today lives scattered around, built as custom software.

Notice that only one of the five has AI in the name of the service. In the others, it may appear as one piece, for example to classify a lead or suggest the category of a transaction, but it is not the product.

Our method has a principle that sums this up: architecture before the tool. The technology is chosen after the problem, the constraints and the expected outcome are understood. Another principle completes it: adding components, models or automations is only worth it when it increases capacity, clarity, security or efficiency.

In practice, every project starts with a diagnostic conversation that defines four things.

• The bottleneck. Where exactly the work gets stuck today.
• The dependencies. Which systems, people and data the flow touches.
• The evidence of completion. What needs to be true at the end to say it worked.
• What stays out. The limit of the scope, said out loud before starting.

The budget only comes after that cut. And the success criteria are commercial, not technical: adoption, reliability, reduced cost or rework, increased revenue or capacity, operational clarity and sustainable maintenance. If none of them moves, the project did not work, even if the AI is impressive. We wrote about the risk of the opposite path in how to automate without automating chaos.

There are cases where AI is exactly the simplest tool: reading free text, answering questions from company documents, summarizing, classifying. Even there, the problem we see often is the same one described on our agents service page: there is interest in AI, but no definition of source, cost of error, human escalation and accountability.

That is why, when an agent enters the project, it comes with a knowledge base of declared origin, tests, guardrails, an escalation flow, cost logging and shutdown criteria. That is what makes the system auditable. And we take one principle seriously: AI and automation should expand people's capacity without erasing responsibility, context or the possibility of review.

So my company shouldn't use AI? It should, when it solves a problem you can name and measure. The mistake is starting with the tool and looking for the problem later.

How do I know if an AI proposal is good? Check whether it says which bottleneck it solves, how the outcome will be measured, what happens when the AI gets it wrong and what stays out of scope. If one of those answers is missing, the project's question is missing.

Do I need a big project to start? No. The safest path is to start with a small, well-defined flow, measure it, and only then expand.

DOUVRAS Labs works with websites, custom software, automation and integrations, AI agents and technical rescue, and the starting point is always the same: bring the problem before the feature list. In the diagnostic conversation, we turn "we want to use AI" into a bottleneck, an evidence of completion and a scope with limits.

And in your company, what is the question AI should answer, before someone decides the answer is 42?

Douglas Henrique, DOUVRAS Labs.

Christensen, Clayton M.; Cook, Scott; Hall, Taddy. Marketing Malpractice: The Cause and the Cure. Harvard Business Review, December 2005. hbr.org

Adams, Douglas. The Hitchhiker's Guide to the Galaxy. Pan Books, 1979.

Phrases from The Hitchhiker's Guide to the Galaxy. Wikipedia. wikipedia.org

DOUVRAS Labs. Method and epistemological contract. douvras.com