Five predictions for 2027: when software starts doing the buying
By Dr. Terrence E. Brown, Professor of Technology-Based Entrepreneurship, KTH Royal Institute of Technology · November 10, 2026
For most of the history of marketing, the customer was a person. Someone saw an ad, compared a few options, asked a friend, and decided. That assumption is about to break.
In 2018 my doctoral dissertation described the consumer digital twin: a digital model of an individual consumer, built from data, that can be asked how that person would think and act. At the time it was a research idea. Today, AI agents are starting to search, compare and buy on behalf of real people, and payment networks such as Visa are building for it. An AI agent is a digital twin with a wallet. It does not only predict what you would do. It acts and pays.
That changes who companies are selling to, what they spend money on, and what founders need to prove. Here is what I expect in 2027.
1. AI agents start shopping for real people, and many ads will have nobody left to persuade
In 2027, agents buying on people's behalf move from demos to everyday use. Brands that machines can't read will lose sales without ever seeing them go.
An agent does not watch a commercial or scroll past a banner. It reads prices, product data, reviews and return terms, and it compares them. Advertising was built to persuade a person at the moment of attention. When the shopper is software, that moment often never happens.
The risk for companies is not that they lose a sale they can see. It is that they are never on the shortlist, and nothing in their dashboards tells them why. Marketing moves from persuading people to being chosen by machines.
2. The first big product flop blamed on "AI customers"
Startups will make synthetic customer research standard practice, and at least one prominent launch will be built on AI-customer enthusiasm that real buyers never show. The lesson: simulations test ideas, they don't prove demand.
AI customers are useful for testing ideas, challenging assumptions, exploring customer reactions, and identifying weaknesses in a value proposition. However, they are simulations, not real customers, and their responses should never be mistaken for evidence of actual customer needs, willingness to pay, or market demand.
My 2018 dissertation warned that twins built on bad data give faulty predictions, because the records they are built from contain many errors. The same holds for today's synthetic panels. They are fast and cheap, which is exactly why some teams will mistake them for the market. The founders who win will use AI customers to find the flaws in an idea, and real customers to decide whether to build it.
3. Product data becomes the new brand budget
Companies shift money from advertising to clean, consistent product information, return terms and reviews, because that's what agents compare.
If agents choose on facts, the facts become the marketing. A clear price, an accurate specification, a return policy that says what it means, and reviews that match the product will do work that a campaign used to do.
This is also an opening for startups. If agents compare offers on facts, a small company with a clear, honest offer can beat a big brand's advertising budget. Brand-building matters less; product data quality matters more.
4. Trust becomes the scarcest asset
As AI floods every channel with content, both people and their agents will favour sources with a track record they can check. Founders' personal credibility will count for more, not less.
Trust is moving. The shopper trusts the agent, and the agent trusts whoever's data is accurate and consistent. Agents working from bad data will make bad choices on people's behalf, and the companies feeding them that data will pay for it.
At the same time, when anyone can produce endless content at no cost, content stops being a signal. A record that can be checked becomes the signal: who has been right before, who stands behind what they say. For founders, that makes their own credibility a business asset, not a personal brand exercise.
5. Smaller teams, harder evidence
AI lets two- to five-person startups reach revenue that used to take twenty people. Investors will judge teams less by headcount and more by real customer evidence.
Headcount used to be a rough proxy for ambition and capacity. When a small team with AI tools can do the work of a much larger one, that proxy stops meaning much. What remains is the question investors should always have asked first: do real customers want this, and will they pay?
What to do now
The common thread is simple. Software is becoming both a research tool and a customer, and neither replaces the real thing.
- ·Make your offer readable to machines. Prices, product data, reviews and return terms are now marketing.
- ·Use AI customers to find weaknesses, not to prove demand. Test with simulations, decide with real buyers.
- ·Watch for gamed agents. Agents optimize what they are told, often price. Sellers will try to game them, and shoppers may lose discovery: the product they didn't know they wanted.
- ·Treat trust as an asset you build on purpose. Consistent data and a checkable record will matter more each year.