You can build trust and positive rapport with customers when you take the measures to protect their information, confirm actions of high consequence, avoid unsolicited content, and meet the expectations customers will have of your skill.
Checklist for designing trustworthy skills:
▢ Where there are potentially multiple customers or accounts, confirm the intended listener, and/or that there aren’t other people present, before revealing potentially sensitive information
▢ Follow the requirements for protecting sensitive information with a PIN code
▢ Surface only relevant messages to the customer
▢ Don’t surface options that will result in an error if selected
▢ Don’t interrupt an experience with an offer to take an unrelated action
▢ State a clear value proposition for actions your skill asks the customer to perform
▢ Ask customers to confirm actions that have a high consequence for error
▢ Use implicit and contextual confirmations
▢ Don’t reveal sensitive customer information on the screen
▢ Display information on-screen that is relevant to the context of the verbal responses
▢ Test your skill on a range of devices with a screen, without a screen, and on the go
▢ You skill name should be an accurate representation of its functions
▢ Your skill’s description, in the skill store and actual experience, should be an accurate representation of its functions
▢ Don’t explicitly or implicitly promise future functionality
▢ Don’t require additional confirmation from the customer to exit when they use the stop/exit command unless significant data or progress will be lost
▢ Don't change Alexa’s name or personality, or treat Alexa as a brand representative
▢ Alexa should use “we”/“us” pronouns to refer to herself and the customer, not herself and a third party
Remember that Alexa devices are often in communal environments such as the living room. If your skill performs tasks that involve sensitive information, take steps to reduce the chances your skill might blurt out that information intended for one individual to everyone in the household. This might include a confirmation step before revealing the information. For example:
Customer: Alexa, ask ShopBuddy what’s in my cart?
Alexa: Was that the ShopBuddy cart for Jonathan, or for Cody?
Customer: For Cody!
Alexa: Ok. There are seven items in Cody’s shopping cart, including Super Blocks Mega toy set for $179.99. You can hear the rest, or check out. What would you like?
The ephemeral nature of conversations means that, unlike in a screen-based interface, an action might be taken unintentionally, or it may be difficult for a customer to un-do what has been done. While they should always be able to ask Alexa to “repeat,” misunderstandings can occur. The consequences of such an error might include loss of progress in a process or having to start a task over, spending money unintentionally, or losing a game. There are a few ways you can use confirmations to prevent these errors and their trust-breaking consequences:
These confirmations require customers’ verbal approval to continue. It adds an additional step in the conversation but protects against high-consequence failures such as monetary impact, data or game progress lost. The explicit confirmation acts as a safety net. It should be used when the consequences for an error are high (financial impact, lost progress). For example:
Customer: I’m ready to check out.
Alexa: Ok. Your total for three items is $36.47, including taxes and fees. Are you ready to place your order with Toys 4 You?
These confirmations re-state what customers said, but don’t ask them to confirm it. The customer can still take action to stop the interaction and correct the mistake. Use this confirmation when the likelihood of a mistake is low, and the consequences of a failure are minor.
Customer: Alexa, ask Trip Planner to start my vacation on June ninth.
Alexa: Okay. You're leaving June ninth. What day are you returning?
These confirmations let customers know that the skill understood them and that they are moving to the next step in the skill workflow. The confirmation doesn't explicitly re-state the utterance, but includes enough information to imply that Alexa understood correctly based on the next response it delivers. Use these for faster-paced conversations with low probability and minor (or no) consequences for errors.
Alexa: I have trivia about People, places, and animals. Which would you like?
Alexa: Ok. First question: what animal is the only marsupial in America?
Despite your best intentions, when you surface information that is unrelated to a customer's primary task, customers may perceive this as promotional content that puts business needs over their needs. Customers lose trust in skills that deliver unsolicited messages, especially those messages that stand in the way of the experience they expected, interrupt an experience in progress, or aren’t relevant to them at all.
Watch the following interaction where the “skill” breaks the customer’s trust by offering numerous irrelevant unsolicited upsells:
Everything your skill offers the customer should be relevant to them (see above, Be Contextual), and where the benefit or relevance to your customer isn’t immediately obvious, you’ll need to include a strong value proposition. Any time a skill asks a customer to act, always include a value proposition for doing so. State a clear value proposition when your skill asks the customer to perform an action that will result in an action being taken on their behalf (setting notifications, sending something to their phone), accessing their personal information (such as financial data), or to make a purchase (in-skill purchasing upsells).
Recall that in our hypothetical MyCapital example above, we clearly told the customer the value of linking their account.
Customer: Alexa, open the MyCapital skill
Alexa: Welcome to MyCapital. I can help you check your MyCapital card balance, pay your bill, check your credit score, and more. To start, you’ll need to link your MyCapital account. I’ve sent some information to your Alexa App to help you.
Avoid setting false expectations about what your skill can do. What will a customer expect a skill to do, either from its name or the brand that your skill represents? Does your mobile app function completely differently than your skill? Does your skill name or description set high expectations for functionality that your skill can’t deliver?
Do: When a feature is limited, be honest about it, and inform the customer of what they can do.
Alexa: I can't track calories by month, but I can tell you about your calorie intake yesterday, today, or record a food to your food journal. What would you like?
Don’t: Don’t promise and avoid implying existing or future capabilities. ‘Yet' implies that a feature might be available at a future date when in fact it might never be supported.
Alexa: Sorry. I can't do that yet, but I'm always learning.
Additionally, consider expectations customers already have of voice assistants, such as an experience stopping immediately when the customer says “Stop.” You skill also shouldn’t change Alexa’s name or re-brand Alexa as a spokesperson. See guidelines for using other voices in Improve Your Audio.