Conversation Design @ Salesforce

CHALLENGE

As .

SOLUTION

Provide detailed guidelines for the central Conversation Design team, and bite sized guidance for all other Salesforce teams and stakeholders, who don't necessarily create copy, but will need to be aware when they should bring in a conversation designer.

MY ROLE

Led the effort to create a unified Salesforce style guide and glossary
Launched an improved version of the Salesforce Conversation Design Guidelines, including a process section that details the Conversation Design Framework

01 | USER RESEARCH

What do users need?

People use the Yahoo! home page for many things, but not everything all at once.

Pains

Navigating the home page can be confusing or overwhelming
Too many products offered—how can another product, a chatbot, simplify Yahoo!'s user experience and still provide value?

02 | COMPETITIVE ANALYSIS

What's out there?

Popular chatbots at the time, such as Sephora's shopping assistant or Hipmunk's virtual travel agent, were built for e-commerce or customer support. This didn't exactly align with the host and range of services Yahoo! had to offer, however.

In Hipmunk's bot, there's a chunk of 4 utterances from the bot. In smaller phones this would cause the user to scroll just to understand what's being said. It would have been nice to see more turn taking.

Sephora's bot offers images of products, which is great for clarity. This is a great way to confirm that the product is what the user wants.

03 | VALUE PROPOSITION

What could be better?

Finance, News and Sports are Yahoo!'s most successful products. We decided that starting with these made the most sense, and would keep things simple. Then we added Weather and a Tamagotchi-style pet called MonkeyPet for fun.

These bots were launched in Facebook messenger and Kik.

Note: After Verizon Media bought Yahoo!, there was a change in priorities and the project was deprecated.

04 | CHALLENGE

How to manage multiple bots?

Our greatest challenge was building a consistent brand voice across a range of topics and user intents.

We knew that Finance Bot’s users would be different from MonkeyPet’s users, so it was imperative to identify and differentiate the voice and tone of each bot as well. For example, the Finance Bot should read as professional with short and complete sentences, while MonkeyPet should exude fun and excitement through lots of exclamation points and emojis.

And how do we do this?

Process

Create bot personas:
-Use data to Identify Yahoo! users into segments, which we'd later map to each bot
-Write a golden set of include/exclude keywords specific to each bot
Create bot playbook:
-Come up with a list of use cases for each bot
-Define "success" and "areas of improvement" in bot-user interactions.
Edit AIML
Monitor user engagement
-After launching the bots, we'd analyze conversation length, user sentiment, retention rate, etc
-Gather user feedback through survey prompts offered at the end of the conversation
Iterate, iterate iterate!

CONVERSATION REPAIR | Error Handling

At some point, users will come across the bot's limitations. When that happens, it is important to apologize and provide the user with a path forward, whether that be a call to action or a reminder of the bot's function and abilities.

Weather:

I'm sorry, I can’t see that far into the future. Type Help, to learn more about what I can do.

MonkeyPet:

Me no support that yet! 😞 and me hungry

For MonkeyPet, the sorry is expressed with the emoji, thereby staying true to its voice.

Conversation Design

ONBOARDING

Finance:

News:

Bot should introduce itself as a bot. The user may feel distrust if they first thought they were talking to a human, and later find out it's a bot

This is the Yahoo Finance bot! Type a company name or ticker to get started.

CTA, moves conversation forward.

Hi! I'm the Yahoo News bot. I can find stories for you, tell you about what's trending, and provide top news twice a day.

List 2-3 options, to minimize cognitive load