1 of 17

Greg Linden

Make Data Useful

2 of 17

“[Help] people find and discover anything they want to buy online”

- Jeff Bezos

3 of 17

Help people find and discover anything online

4 of 17

“Encourage experimentation ... as much of it as possible”

- Jeff Bezos

5 of 17

What is the goal?

Revenue

Traffic

Visitors

Retention

Customer acquisition

Time on site

Clicks

6 of 17

Measure

Learn

Optimize

Iterate

7 of 17

“Worry about the data first before you worry about the algorithm”

- Peter Norvig

8 of 17

[Banko and Brill, 2001]

9 of 17

On the design of recommendation features

10 of 17

Design to encourage forgiveness of mistakes

11 of 17

Help when help is wanted

The goal of recommendations should be “first, do no harm.”

No matter what, never get in people’s way. If they want to search, let them search. Don’t throw recommendations in their face and interfere with their mission.

But, it is good to help at key times. For example, when someone is about to checkout, it is a good time to offer a couple more items for their cart. The timing is right.

It is a little like the impulse buys at the grocery store checkout lane, but this is as if the impulse buys magically rearranged themselves based on what you have in your cart.

Other examples of good timing include the “Customers who bought” on detail pages (where people are looking for which item they want and might need help finding related items) and the personalized home page (where some customers arrive without a clear mission, so it is a great opportunity to show them a subset of the catalog related to their past behavior)

12 of 17

“Discovery is when something wonderful that you didn't know existed, or didn't know how to ask for, finds you.”

- Fortune Magazine

13 of 17

Discovery of the new and surprising using the recent

14 of 17

“Users really respond to speed.”

- Marissa Mayer

15 of 17

+500 ms -20% traffic

@ Google

+100 ms -1% sales

@ Amazon

16 of 17

Speed matters!

17 of 17

Measure, then optimize toward a goal
Big data first, then algorithms
Encourage users to forgive you
Help when help is wanted
Surprise and focus on the new
Speed matters