1 of 39

UX for AI products

Claire Lebarz

VP Data @Malt

claire.lebarz@malt.com

25/01/2024

2 of 39

Subtitle 1

SubSubtitle 1

3 of 39

Start with WORDS

4 of 39

As a user, what are your horror stories?

5 of 39

What are bad AI product experiences you’ve had?

As a host, reordering my photos without telling me why…

As a guest, showing me clearly irrelevant results…

6 of 39

Metrics are how we define if an experience is good.

Metrics reflect values.

Not all mistakes are equal. Metrics are proxies for what we care about.

What we care about evolves.

7 of 39

Model

Interface

8 of 39

Model

Data

Interface

Outputs

Inputs

Metrics

9 of 39

Multiple options: allow people to choose among the results a feature generates

10 of 39

Multiple options

Allow people to choose among the results a feature generates

Prefer diverse options
Differentiate between options
Learn from selection to improve ranking

11 of 39

Attributions: explanations that help people understand more about how your app makes decisions

12 of 39

Attributions

Attributions are explanations that help people understand more about how your app makes decisions

Avoid profiling people
Relate to objective facts, not subjective taste
Cite data sources

13 of 39

Confidence: a measurement of certainty for a result

14 of 39

Confidence

Confidence is a measurement of certainty for a result

Numbers are appropriate for statistical predictions
Translate confidence into easy to understand terms
Use range to help people understand risk

15 of 39

Limitations: happen when there’s a mismatch between people’s mental model of the feature and what the feature can actually do

16 of 39

Limitations

Limitations happen when there’s a mismatch between people’s mental model of the feature and what the feature can actually do

Manage expectations
Guide people to move past limitations
Suggest alternatives

18 of 39

Calibration allows people to provide essential info needed to engage in an experience

19 of 39

Calibration

Allows people to provide essential info needed to engage in an experience.

Be quick and effortless
Avoid the need for multiple calibration
Introduce, guide and confirm
Allow people to update information

20 of 39

Implicit feedback is info that arises from the interactions people have with your app

21 of 39

Implicit feedback

Implicit feedback is info that arises from the interactions people have with your app

Strive to identify people’s intention
Accuracy over speed
Respect people’s privacy
Make interactions more accurate and delightful

22 of 39

Explicit feedback allows your app to collect info by asking specific questions regarding your results

23 of 39

Explicit feedback

Explicit feedback allows your app to collect info by asking specific questions regarding your results

Prioritize negative feedback over positive
Clearly describe each option and its consequences
Act immediately and persistently

24 of 39

Corrections

25 of 39

Corrections

Allow corrections through familiar ways
Provide immediate value
Use corrections as implicit feedback

28 of 39

Model

Interface

Outputs

Inputs

Outputs

Multiple options
Attributions
Confidence
Limitations

Inputs

Calibration
Implicit feedback
Explicit feedback
Corrections

Recap

29 of 39