Ethical AI CX: Responsible Personalisation in ecommerce (2/3)
As anticipated in the first post on Ethical AI CX, centred on Consent as Communication, the second use case will delve on Responsible Personalisation: shifting from inferences and surveillance to conversational explicit profiling experiences.
Getting to know your customers in a trustworthy way
Stop annoying your customers with products they already purchased or not interested any longer. If you don’t personally like it as a shopper, please be mindful rather than creepy. Personalisation is not purely about relevance, serendipity and conversion, but also, and more importantly, about transparency, agency and trust.
Let’s apply common sense. Sharing intimate preferences, interests, health data, personal images, voice, etc implies trust. Let’s ask, let’s explain and let’s make it fun and convenient when it comes to getting to know your customers. Let’s turn profiling into a progressive conversation, evoking trust through transparency and nurturing a sense of relationship or belonging. As my colleague Angel Maldonado constantly reminds us, common sense always wins.
This post elaborates 3 use cases on responsible personalisation that elevate explainability, privacy and bias as their main ethical principles.
1. Mission-based Progressive Contextual Profiling
Let’s use AI to make searching, shopping and profiling a seamless experience:
Mission-based recipe suggestions + progressive profiling. Once avocado is added to the cart:
- “Want to make some tacos? or Ceviche?”: A selection of recipes are recommended, based on the last action performed + items in cart.
- “Add all ingredients”CTA is displayed for improved convenience. No need to check your cart.
- “Add “Mexican Food” to My Preferences” for privacy-first explicit profiling and explainability: Yes <> Check My Preferences <> Don’t Ask again — Great for customer who value time and convenience <> Opt-out — always easily accessible throughout the whole process.
Frequently bought together items: Once avocado is added to the cart, Cilantro, Lime etc are displayed as an automatic “bundle” for easy basket building.
Progressive explicit profiling works very well due to familiarity. To some extent, we’ve all used it in the past with music apps, media and other sites where our taste and affinities are shared.
2. AI Informed Profiling > My Preferences AI Assistant
Once customers understand how profiling works, progressively built as part of the shopping experience, they may find it too interruptive and decide to “jump” subsequent prompts. As shared before, always allow to choose “don’t ask again” and also design stand-alone profiling experiences using AI to power them.
AI Assitant for Explicit Profiling
Here’s some quick tips:
- Use AI to infer most likely preferences based on past purchases, aggregated search journeys and favourite lists, etc
- Make it easy to change preferences. Always think about your customers being in control of the automated output.
- Use Confidence Display. Use numerical values or alternative copies (ie Best Match) to signify Preference Trust Scores. See more examples in this great post from Irinanik.com: “Don’t Build trust with AI, calibrate it”.
- Make it fun and entertaining. A form with check boxes — as used in the design above for simplicity- is not the best formula. Only there for simplicity. Be creative: tinder-like swipes, icons, curated lists, quizzes, gift-finders, etc..
- Make profiling during Onboarding very light. It’s not wise putting your customers off very early in the journey. Build trust first, then apply the magic. ;)
- Balance progressive with stand-alone profiling experiences.
- Lists: Allow your customers to self organise their preferences in lists, making them shareable, making them public, etc… as we’ll surely done in the past with our music lists in Spotify.
- Preference insights: Show the weight of your affinities in your past and new orders.
3. Preference Activation & Explainability > My Cart
Once profiling has been completed, or at least started, let’s look at some examples on how to activate those preferences in ways that are explainable.
Improve your cart with preference based product substitutions Nutritional:
- “Other Gluten free options”
- Savings: “Similar item in Promotion”
4. Preference Explainability & Bias
Introduce explainability tags and feedback mechanisms, such as favourite and opt-out icons for signalling positive/negative feedback on automated preferences.
Try seeking consistency in profiling by using normalised signals and experiences.
- Favourite icons. Simplify the declaration of favourite signals (ie. items, filters, preferences…) with normalised iconography, such as the heart for increased understanding and explainability.
- “Why is this here”- Add explainations of the preference applied to the item displayed.
Lastly, try adding preference opt-out directly in the shopping experience.
- Feedback & Opt out: If there’s a preference that you dont like any longer, consider adding opt-out options in-context.
Take-aways
- Personalization can be done in ways where your customers not only are aware, understand and consent sharing their affinities and interests, but also enjoy doing it.
- Design profiling as part of the shopping experience, where positive signals such as search queries, add to carts, engagement with recipe suggestions, etc act as triggers for preference building.
- Use AI to make profiling convenient and fun. Apply AI to assist in completing those boring profile forms. Add lists, shareability etc for customers to enjoy curating how they want to shop.
- Always allow customers to provide feedback to automated inferred profiling suggestions
Personalization is about understanding and control.
Embrace personalization. Trust is earned, not given or exchanged.
The next and final post of this Ethical AI CX Series will focus on AI Personal Shopper.
