We read the newest studies from the top business schools and kept the ones that matter for growing a business. Every finding links to the study and to the people who did it.
AI search
8.8
extra website clicks per 100 searches with AI answers hidden
Google's AI answers are taking clicks from websites
Researchers at Penn and Northeastern had 1,100 people search Google in their own Chrome browsers for ten days in March 2026, while a browser tool changed what each group saw. Hiding Google's AI Overviews and AI Mode added 8.8 clicks to outside websites for every 100 searches, and people rated their searches just as good. Sending every search to AI Mode instead cut clicks by 18.8 per 100 searches and lowered trust in what they found.
What it means for you. Plan for fewer visits from Google searches that show AI answers, and judge search by calls and leads, not visits alone.
AI shopping agents reward ratings and snub Sponsored tags
50%+
more AI picks after a 0.1 star rating gain
Columbia and Yale researchers built a test store showing eight products per page and let leading AI shopping agents choose, while randomly changing prices, ratings, badges and positions. A product picked 10% of the time rose to between 15% and 20% when its rating was just 0.1 stars higher, while a Sponsored tag cut it to about 8% to 9%. The agents also crowded onto a few favorite products, and those favorites shifted when the models were updated.
What it means for you. If AI assistants start shopping for your customers, strong ratings and clear, accurate listings may decide whether you get picked.
By Amine Allouah, Omar Besbes and colleagues
Columbia Business School and Yale University, with the startup MyCustomAI · 2026 · Proceedings of the ACM Web Conference 2026 (WWW '26)
Researchers from Columbia Business School and Zhejiang University studied randomized field experiments that a large cross border shopping site ran across seven tasks in 2023 and 2024, covering millions of shoppers. An AI chat assistant that answered questions before purchase lifted sales 16.3% compared with an automatic reply saying no agent was available. AI written product descriptions added about 2%, and AI rewritten Google ad titles showed no clear gain.
What it means for you. Put AI to work first where customers wait for answers, such as chat on your site, and expect smaller gains from AI written copy.
Losing tracking data makes each customer cost more
31%
higher cost per customer won without offsite tracking data
Researchers from Kellogg, Chicago Booth and Meta ran a randomized experiment with more than 70,000 advertisers on Facebook and Instagram. Without offsite data, meaning signals about what people do on other websites and apps, the median cost to win one extra customer would rise from $38.16 to $49.93. Small advertisers had the most to lose.
What it means for you. Keep your website tracking and conversion data connected to your ad accounts, because ads without it cost more per customer, especially for small businesses.
Stanford researchers studied a randomized banner ad test for one of the biggest US auto insurers, run with about 300,000 people on a major website. Seeing the ad lifted visits to the insurer's site from 0.2% to 0.8% that day, and the lift lasted up to two months. Rival insurers got 11% fewer visits the next day.
What it means for you. Regular ads keep you in mind when people are ready to buy, even if they already know your name, and make rivals less likely to come to mind.
Economists from Duke, Goethe University Frankfurt, Humboldt University Berlin and the Indian School of Business ran a randomized trial with a German online furniture seller, which offered a pay later option to randomly chosen shoppers. Offering buy now, pay later lifted sales by 20%, mostly from customers with weaker credit. The merchant's gains clearly outweighed its costs.
What it means for you. If you sell bigger ticket items online, a pay later option at checkout may win buyers you would otherwise lose, so test it against the fees.
Duke University Fuqua School of Business, Goethe University Frankfurt, Humboldt University of Berlin and Indian School of Business · 2025 · Journal of Financial Economics
The right loading animation keeps visitors waiting
2 in 3
stayed through a 30 second wait with moderate animation
Stanford and Babson researchers ran experiments with more than 1,400 people, plus tests using real Facebook ads, changing how fast the loading animation moved during a wait. Moderate speed animations made waits feel shortest. In a 30 second wait, more than two thirds of people who saw a moderate animation stayed to finish, versus under a third with a static image and about half with a fast animation.
What it means for you. Make pages fast first, but where a wait cannot be avoided, a steady, moderate loading animation keeps more visitors than a frozen screen or a frantic spinner.
fewer unsubscribes when daily emails became every other day
Researchers from NYU Stern, Columbia and Boston College ran a one week test with 603,153 customers of a major North American online retailer, sending one group coupon emails every other day instead of daily. Fewer emails cut unsubscribes by 59% but lowered short term revenue by 5% to 8%. Because customers who unsubscribe go on to spend 36% less, the authors estimate that tailoring email frequency to each customer raises lifetime value by 6.8% to 8.3%.
What it means for you. Send promotions less often to customers who are tiring of them; a little short term revenue buys customers who stay and keep spending.
clicks per post, relaxed brief versus strict brief
Researchers from Stanford, the University of Lucerne and the University of Basel studied 984 real influencer contracts, then ran a field experiment with 1,495 Instagram influencers for a cosmetics brand. Detailed rules on exactly how to show the product made influencers less willing to sign up. Posts made under the brand's strict brief drew about 14 clicks each, versus about 53 when the rules were relaxed.
What it means for you. When you hire creators, give them the goal and the key facts, then let them show your product in their own style.
earned per dollar by minority workers under five stars
Researchers from Yale, Rice and the University of Toronto studied 69,971 customer ratings on a home services platform that switched from five stars to a thumbs up or down. Under five stars, racial minority workers got top marks a little less often (83.4% vs 86.9%), and they earned 91 cents for every dollar white workers earned for the same work. After the switch, top marks were nearly equal (95.4% vs 95.9%).
What it means for you. If you ask customers to rate your staff or crews, a simple thumbs up or down can cut hidden bias that five star scales let in.