• Ariyh
  • Posts
  • Explain your AI recommendations

Explain your AI recommendations

People are up to 17.6% more likely to say they will buy when you explain how your AI made a recommendation (e.g. “Because 500 other people in your neighborhood are going to this concert”)

Author

Thomas McKinlay
April 23, 2024

New to Ariyh? This is a 3min practical summary of a scientific study 🎓 Join 27,214 marketers who use science, not flawed opinions 📈 Subscribe here

The team at Modash interviewed 5 influencer marketers in B2C brands to learn how they find influencers.

  • How they filter inbound leads

  • How they search using software

  • How (and if) they work with talent managers

  • ...and more.

Want to sponsor Ariyh? Here’s all you need to know.

📝 Intro

You’re updating your ecommerce store and want to test out an AI plug-in that can make personalized product recommendations to your customers.

The AI plug-in offers two interface options:

A. Simply give an auto-generated recommendation

B. Give a recommendation and provide an explanation why it’s recommending that specific product

You’re unsure which option to go with. You want to keep the interface clutter-free, but the explanation could add some legitimacy.

Here’s why leaving some space for that explanation is worth it.

P.S.: Are you recommending products in person? Suggest pairings of different products (e.g. a rug to match a sofa), people will consider you a more credible expert and will be more likely to buy.

Want to access hundreds more insights like these? Explore Ariyh insights here.

People are more likely to trust and follow AI recommendations when they understand them

Topics: Website/App | Ecommerce
For: B2C. Can be tested for B2B.
Research date: December 2023
Universities: Chongqing Normal University & Shanghai University of Finance and Economics

📈 Recommendation

When giving AI or algorithm-based recommendations, always explain why the recommendation was made (e.g. because people with similar tastes bought it).

This is especially important for functional, practical products (e.g. cleaning supplies, tools) compared to emotion-driven products (e.g. perfume, chocolate).

People will be more likely to follow the recommendation and buy.

🎓 Findings 

  • People are more likely to say they would buy products recommended by AI or algorithms when there is an explanation of why the AI made those recommendations (e.g. “Check out this cream, it’s the most repurchased item in our store!”).

  • Scientists ran 4 experiments and found that when given an explanation (vs no explanation) of how the AI algorithm worked, people were: 

    • 12.5% to 17.6% more likely to say they would buy a pair of sneakers

    • 54.6% more likely to click "See more details" for recommended t-shirts (vs clicking "View other items" instead)

  • The effect is strongest for products bought for their functionality (e.g. insect-repellent candle), compared to products bought for pleasure (e.g. decorative candle).

🧠 Why it works

  • AI assistants are a relatively new experience for most shoppers.

  • Since we don’t always understand how AI works, we’re skeptical of its choices. 

  • So when a business is transparent about how and why its AI recommendations are generated, it makes us understand it better.

  • We then trust the AI recommendation more and become more willing to buy.

Join the Essential Community for Marketers

Stay ahead of the curve and join the thousands of marketers shaping the future of marketing with the American Marketing Association. As the largest community-based marketing organization, you’ll find award-winning content, professional certifications (PCM®), industry-leading training events, and vibrant local chapters. Members get access to an abundance of resources, on-demand courses, and the most essential community in marketing.

This announcement was sponsored. Want your brand here? Click here.

✋ Limitations

  • The experiments focused on sneakers, shirts, and face creams, which are generally quite cheap. The study did not test any expensive or premium products, for which people might be more likely to do more in-depth research (e.g. reading reviews) before buying.

  • The study collected data using a Chinese survey website Credamo. It's unclear whether the effects are the same in other cultures (e.g. Chinese consumers may be more likely to trust AI compared to Western consumers, so the effect could be even stronger).

🏢 Companies using this 

  • Online stores tend to not explain how their algorithms work as it can make the page look busy and clutter the user interface. 

  • Most do not go beyond basic sentences like “Goes well with” or “Other people bought this too”, which don't usually provide a sufficient explanation:

    • Clothing retailer ASOS product pages have a “Similar items” recommendation section

    • Hershey’s product pages have a “You might also like” section

    • Amazon’s “Discover similar items” and “What do customers buy after viewing this item” sections

TripAdvisor does a good job of explaining how nearby places are recommended using an information bubble.

⚡ Steps to implement 

  • Include AI recommendations in your online store. People especially appreciate AI recommendations for practical products (e.g. toothbrushes). 

  • Explain how these AI recommendations are made and what data is used (e.g. “Other dog owners love this toy”).

  • Be mindful not to clutter the user interface. For example, you could use a ‘See why you got this recommendation’ near the AI recommendation which expands and explains the recommendation.

  • Adjust your AI’s recommendations according to different markets and cultural preferences. People in interdependent cultures (e.g. South America, Asia) are more interested in which products are ‘Top Rated’ (vs which are ‘Bestsellers’). 

  • For emotional and enjoyable product (e.g. cinema) AI recommendations, focus on other users’ preferences (e.g. “Try it! Most popular this week”) rather than functional qualities (e.g. “The most ergonomic choice”)

  • For functional products, the AI recommendation should be based on the product features (e.g. “Ultra lightweight choice”), or both.

🔍 Study type

Online experiments.

📖 Research

🏫 Researchers

  • Changdong Chen, Chongqing Normal University 

  • Allen Ding Tian, Shanghai University of Finance and Economics

  • Ruochen Jiang, Shanghai University of Finance and Economics

Remember: This is a new scientific discovery. In the future it will probably be better understood and could even be proven wrong (that’s how science works). It may also not be generalizable to your situation. If it’s a risky change, always test it on a small scale before rolling it out widely.

Rate today’s insight to help me make Ariyh's next insights 🎓 even more useful 📈

How was today’s insight?
Loved it | Great | Good | Meh | Bad