Share
Explore

Two Medical Customer Support Conversational AI strategy

Two Medical wants to improve the customer service experience by allowing our customers to chat with a conversational AI-powered chatbot. The goal of this paper is to define the plan to take this initiative to market.

Goal: Improve customer support experience

What does this mean?
Better satisfaction
What satisfies our members?
Quicker time to resolution
Personalization (Knowing who they are, showing them relevant data, Etc.)
They don’t have to repeat themselves, even when passed from CSR to CSR


Learn, Build, Measure loop:

The end goal is to have a powerful conversational AI platform that leverages AI and ML to service our customers. Since we’ll have the opportunity to learn about how our customers interact with this type of technology along the way, we should use this to our advantage and take an iterative approach to our go-to-market.
image.png


Customer:

Americans in rural who use us for their Health Care. Here are some things we know about them:
Likely have to drive a long way to get medical help
Agriculture employs less than 5% of rural Americans
In 2017, the rural poverty rate stood at 16.4% compared to urban at 12.9%
Most popular web browsers in our demographic: Chrome, Safari, Edge
Average $$ spent on smart phone: $200


Opportunity:

Higher customer satisfaction
Faster responses to our customer questions
24/7 customer service
More efficient customer service team - Less time spent answering easy questions, more time spent solving the complicated issue
Consistent answers for our customers
Learn more about our customers in a more scalable way

How do we measure this?

Improved customer satisfaction increases our bottom line, and we can measure that with the following proxy metrics:
NPS scores/CSAT scores
Churn Rate
Retention Rate
Lifetime Value
Customer service hours saved is also an easy way to measure cost savings


Who is involved?

This document is available to each department head, and I’ve schedule time with them or a representative on their team to understand any concerns they may have. More specifically the following groups will help with the following tasks as we plan this initiative:

User Research:
Build persona of our “model” customer who stays for our average retention period
Understand the customer service team’s daily jobs and figure out what all we can automate with conversational AI
Engineering:
Acquire live chat transcripts
They can use those transcripts to begin testing various NLP and ML tools we will need to make this overall vision a reality
Design:
They will be responsible for designing the conversation flows, so their input with the tools we use will be crucial too
Data/Analytics:
Current retention rates
Churn rates
Cost Per Acquisition
Average number of customer service hours to keep our “model” customer happy
Internal Training Department:
Build training documentation to help the teams across the company understand what this initiative is and what the goals are to weigh in with their domain knowledge.
Go-to-Market taskforce
0
Name
Department
Responsible
Accountable
Consulted
Informed
1
Joe
Product
2
Stephanie
Customer Service
3
Josh
Customer Service
4
Kathy
Ops
5
Matt
Data/Analytics
6
Jessie
Engineering
7
Frank
Engineering
8
Stan
Design
9
Oprah
Design
10
Seth
Marketing
11
Stacy
CEO
There are no rows in this table
So everyone knows what sort of communication cadence to expect, here’s how I’ll be keeping you in the loop according to the Go to Market task force RACI chart above.
Responsible
: Daily communication and working on the initiative
Accountable
: Weekly updates, access to Slack channel where work is happening
Consulted
: Consultation as needed and monthly updates
Informed
: Monthly email updates on the progress

Rollout

I like to deliver value to the customer as soon as possible so that we can learn how they use the product and then continue to iterate to delight the customer. With that “learn, build, measure” approach in mind, here’s how we will roll this out:

Learn

Acquire a dataset (list of chat transcripts)
User Research will begin interviewing customers and customer service teams to start uncovering the most common use cases that we can automate.
Give engineering and design a period to explore the data set using various NLP and ML tools to see what will help us achieve our goal AND fit into our technology stack
Capabilities need to exist to roll out:
Capability to understand what our users are saying
Implement natural language processing/understanding tools
Capability to respond to our users
Begin building our machine learning model
Capability to predict customer behavior like churn based on their interaction with the chatbot.
Decide on the tools we’ll use
NLP/NLU
ML
Analytics
Code Languages
Databases
Etc.

At this point, we can begin building.

Build

We’ll take an iterative approach here. We’ll call it “Crawl, Walk, Run.”

Crawl part 1
: AI-assisted chat experience for customer service rep
Incorporate the NLP/ML in the existing chat tech.
As a customer service agent, when a customer sends me a message, a potential response is generated by our AI platform.
If the machine response is good, the CSR may send that response by clicking a button.
If not, the CSR can write their response. If the CSR types the response we’ll store the transaction in a DB so we can later work to solve these issues programatically.
Crawl part 2
: CSRs help train the ML model
When a CSR types their own response, the ML model flags that transcript for the team to analyze and train.
Walk
: Chatbot and voice experience.
Customers can chat with a chatbot/voice bot trained on customer interactions with Two Medical. This will fully automate many aspects of the customer service interaction, providing 24/7 customer service for a large number of issues.

Run
: Predictive Analytics + integrated marketing = less churn
Predictive analytics from customer interactions to predict when a customer might not be happy paired with proactive “save” campaigns to reduce churn even more.

Measure

Improved customer satisfaction increases our bottom line, and we can measure that with the following proxy metrics:
NPS scores/CSAT scores
Churn Rate
Retention Rate
Lifetime Value
Customer service hours saved
Additionally we’ll measure product usage of the bot
Total conversations handled
Most common intents triggered
Where people drop off in various conversation flows
Repeat visitors
Number of questions asked per session
Session length


Tactical details for the rollout

We’ll use agile scrum project management methodologies to plan and track the work
Sprints will be two weeks
Daily standups for the parties on the GTM taskforce chart marked “responsible.”
Bi-weekly check-in with GTM taskforce
Typical format: a pre-recorded video or email updating everyone on the status. Occasionally a meeting if decisions are requried.


Questions that need to get answered:

As we go through this process, questions will arise. We’ll track them and answer them here.
What data do we have from our customer service team now?
Do we have chat transcripts from our chat tool?
Yes
Do we know what the most common questions are?
Yes
Will we ever consider enabling a conversational AI-powered “IVR” or similar technology for phone support?
Maybe eventually
How many questions come into our customer service teams each day?
Are they able to answer them all?
What is our customer satisfaction score right now?
Why satisfaction?
Better long term brand trust
Proxy to retention



Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.