New customers took about 30 days to set up their first feature. I designed an AI agent that does that setup with them in a chat, on day one.
Since an agent could work through our APIs, together with my team we pitched the chat experience to our VPs. I designed the experience and built its front end, working with a PM, a product owner, two engineers and two data scientists.
What are the results of this work?
53 of 114 beta customers had a feature set up on day one.
That’s every customer who finished the optional onboarding. Before, it took around 30 days.
Satisfaction averaged 4.4 out of 5, and ease of use 4.6.
From an in-product survey about the onboarding, with 28 responses. 82% rated their satisfaction 4 or higher.
The situation
New customers signed a contract and landed in an empty product.
There was no onboarding or training, only a help centre that assumed you knew what to search for.
Customers took about 30 days to set up a first feature, and our data analytics team found that customers who engage in their first 30 days are more likely to stay. Most never found the features that would have made their job easier.
The goal
Get new customers to value on day one, inside the month that decides whether they stay.
Customers needed the product set up for their business without learning how it’s built first. Verizon Connect needed less early churn.
Decision 01
With context from sales, the agent sets up what each customer needs.
Every new account needs some basic setup, like create users. My first question was which features would help each customer get their job done.
So the agent starts from what sales already knows. An LLM from our data science team turns the sales notes into context, and the agent picks the features that match.
I also considered guiding people through the existing screens. Each part of our product is built differently, so a walkthrough per area wasn’t realistic. New users valued not having to hunt for settings, though the two who knew the product best asked for the walkthrough we dropped.
Decision 02
The agent always asks before it acts.
In a chat, the agent can run ahead and lose the user. So before it creates anything or moves to the next step, it waits for a yes, and customers can change anything by asking.
In testing, this kept people engaged. They learned how to talk to the agent as they went, and felt confident in what it set up.
“There are sometimes stuff you just wouldn’t think about, so it’s good that it does it for you, or it lets you know about them and you can edit them.”
Decision 03
Users can change anything the agent recommends.
Based on what we know about each customer and their needs, the agent recommends a set of features, and fills in their settings with the values most customers use. New customers don’t know yet what fits their fleet, so the agent gives them a starting point.
Every recommendation stays open to change. Customers can adjust a setting, change a value or remove a feature that doesn’t fit their needs, just by asking.
The outcome
Users get value from day one, where it used to take a month.
The beta showed that an agent can do the setup with a new customer on day one. It also showed that chat alone isn’t enough for everyone. Some customers wanted to see more of the product itself as they went, to learn how it works.
We’re now evaluating a mix of chat and other guidance.
53 of 114 beta customers had a feature set up on day one.
That’s every customer who finished the optional onboarding. Before, it took around 30 days.
Satisfaction averaged 4.4 out of 5, and ease of use 4.6.
From an in-product survey about the onboarding, with 28 responses. 82% rated their satisfaction 4 or higher.