The Business Times reported on 6 November 2026 that Databricks believes the current debate about AI’s return on investment works in its favour. Databricks sells tools that help companies manage their data. According to the report, the company says conversations with businesses now revolve around making AI work.
The claim, and the part of it that is true
The pitch is simple. If the return on AI depends on data, then the company that manages your data is the one you need. That is a convenient conclusion for a data company to reach, so it deserves a test.
Is there a real need behind it? Yes. Two years into this cycle, most businesses I meet have tried a chatbot, a copilot or a content tool. The demo was impressive. The result on a normal Tuesday was thin. The model was fine. What it had to work with was poor, incomplete or missing.
What changed is the question. In 2024 the question was which model to use. In 2026 the question is what to feed it. Databricks is reading that shift correctly.
What did not change is where the useful data comes from. For most businesses it does not sit in a warehouse waiting for a pipeline. It sits in the owner’s head.
What “data” means when you are not Databricks-sized
For Databricks, data means pipelines, lakes and governance across thousands of tables. That is a real problem for a bank or an airline.
For a shop with twelve staff, the data that decides whether AI pays back is smaller and far more boring. Your prices. Your refund policy and its limit. Your delivery cut-off time. Which product you would recommend for which problem. What you do when a customer asks for a discount. When a person needs to step in.
Take a clinic that wants AI to handle bookings on WhatsApp. The tool can be excellent and still fail on day one, because nobody told it the clinic takes no new patients on Saturdays, that a first consultation needs a deposit, or that a patient asking about a lump gets a person, not a slot. None of that is in any database. It is in how the front desk has always done things.
A marketing team hits the same wall from the other side. Ask AI to write about your product and it produces fluent, generic copy, because it was given a product name and nothing else. Give it the actual positioning, the actual numbers, the customers you turn away and why, and the output changes completely. Context is the product. The quality of what goes in decides the quality of what comes out.
This is why a platform cannot supply the return for you. It can hold your data and move it around. It cannot know your refund limit until you write it down.
The technology is the easy part
I learned this building AeroChat. The technology was the easy part. Changing how people think and work was the hard part, even for a tech team.
Every workflow a merchant runs on AeroChat starts as a description in their own words: what to check, what to offer, where the limit is, when to hand over to a person. The system turns that into steps and connects to the merchant’s own data and backend systems so it can act, not just answer. We take care of the connections and the hard parts underneath. We never write the merchant’s refund policy for them, because we cannot. A merchant who has written their rules down clearly gets a workflow that runs the way they want from the first week. A merchant who has not spends that week discovering what they never decided.
The same holds for every AI tool I use in my own work. I use Claude for almost everything, and the difference between a useful output and a wasted hour is almost always the brief, not the model. When the output is wrong, the fix is usually in what I gave it.
What a business should do with this
Do not buy a data platform because Databricks says the ROI conversation is a data conversation. For a business your size, the data that matters fits in a document.
Start with that document. Pick one workflow where AI would save real time or catch real sales, such as order enquiries, bookings or product recommendations. Write down the ten questions customers actually ask there and the answer you would give each one. Write the rules: the refund limit, the delivery cut-off, the discount you will and will not give, the moments a person takes over. Use plain words. If two staff would answer differently, settle it now, in writing.
Then give that to the tool, run it, and read what it does. Where it goes wrong, fix the document before you blame the software. Most of the return on AI is decided in that document, long before any platform gets involved.
Source: Business Times: Databricks thinks AI discourse around ROI plays into its favour