For marketers to succeed with their customer marketing efforts, it’s essential to understand which customers are happy, which are at risk of churn and which present cross-sell and upsell opportunities.
Your customer data is full of clues to help you understand which customers fit into each of these buckets. You just need to mine that data to find the clues. Sounds simple, right? It hasn’t always been that way. Then along came generative AI.
If you’re like most people, you hear the term “genAI” and you think about content creation. But its capabilities are expanding and users are experimenting with more use cases.
For this article, I used the Google Gemini generative AI application to help me analyze customer data, identify customers that met certain criteria and receive recommendations and messaging to use with those customers.
Creating sample customer data with Google Gemini
Before I could analyze customer data, I needed to create it. I built a scenario with Gemini to essentially fabricate data in a spreadsheet for customers of a small B2B software company
I told Gemini the columns I wanted in the spreadsheet and provided ranges for some of the data in order to prevent some customers having lifetime revenue several multiple higher than others, for example.
The columns I chose to create were:
- Company name.
- Number of licenses for 2003 and 2004.
- Total time logged in the app (2003 and 2004).
- Average time logged in the app (2003 and 2004).
- Lifetime value.
- Customer supports costs (2023 and 2024).
- Average monthly support costs (2023 and 2024).
One of the memorable moments from this part of the exercise was when I decided to add year-over-year data so I could run comparisons. Gemini added 2023 and 2024 lifetime values for each customer. I pointed out that there is only one lifetime value, and Gemini immediately apologized for the error and fixed it.
All told, Gemini created sample data for 150 customers. Here is the sample data for eight of them.
Let’s dig into customer support costs
You can learn a lot by analyzing support costs. Customers with high support costs are at risk for churn. And customers that cost a vendor more in support than they bring in in revenue are bad deals that need to be fixed.
That’s why my first query was to ask Gemini how many of the 150 customers had higher support costs than revenue. Fortunately, none of them did, which is good news. Next, I wanted to know which customers had the highest support costs over the two years of data we created.
Note that for each year the same companies had the highest support costs, which could be an anomaly in our customer data, a misunderstood prompt or it could be a sign of a larger problem with the customers in question.
What do I say to customers whose support costs are killing us?
Since Gemini is a generative AI platform, it’s more than capable of taking the data and analysis and drafting relevant communications. So I asked Gemini to draft an email to send to customers with high support costs offering them resources to help.
Gemini’s initial draft cited the high support costs in the email, which I asked it to correct because it’s not wise to tell your customers “You’re costing us a ton in customer support!”
The second draft was closer to what I was looking for:
Dear [Customer Name],
We wanted to reach out to express our appreciation for your continued partnership with [Company Name]. We value your business and are committed to providing you with the best possible support.
To help you get the most out of our products and services, we’d like to offer you access to additional training and resources. These may include:
[Training Course 1]: [Brief description of the course]
[Training Course 2]: [Brief description of the course]
[Knowledge Base Article]: [Link to a relevant article]
These resources can help you and your team troubleshoot problems more efficiently, train new users effectively, and maximize the value you get from our products.
By taking advantage of these resources, you can potentially reduce the need for support interactions and optimize your overall support experience.
Please let us know if you’re interested in learning more about these resources or if you have any other questions. We’re here to support your success.
Let’s find those churn risks and get ’em fixed
I asked Gemini to identify in the data the five customers most at risk for customer churn. It pointed out that it was capable of hypothesizing which customers were most at risk, but that additional data, like historic churn rates for example, would help identify risks more accurately.
Here’s a sample from the five customers it identified as churn risks and why.
Identifying churn risks is only half (or even less) of the battle. So I asked Gemini to help me outline a retention strategy for these customers.
A loaded question: Who are the ‘best customers’?
Everyone in customer marketing would love to know which of their customers are “the best customers.” But a lot goes into defining a “best customer.” And, as I expected, when I put the question of which customers from our dataset were “the best” and why, Gemini reminded me we’re working with a relatively simple dataset.
It could do an even better job of answering the query, it said, if it had information on:
- Customer satisfaction ratings.
- Product usage patterns.
- Churn history.
- Referrals made.
Nevertheless, Gemini took a shot at identifying the best customers based on the data we had around customer lifetime value (CLTV), support costs and engagement with the product.
What I learned from this exercise
The natural language capabilities of Gemini and other genAI applications are getting better. I didn’t need to create complicated prompts or ask Gemini to play a role. I simply asked it to do what I wanted it to do.
More than spitting out answers, however, Gemini added useful suggestions, such as additional data for our dataset that would be helpful, or suggestions around strategies.
I found Gemini’s role in this exercise to be part simulator and part mentor. We were using fabricated data, and while the exercise was fictitious, it was also very real. This could have been actual customer data and the results and suggestions would likely hold up. Even as a simulation, it made for a great thought exercise.
The suggestions and areas for improvement Gemini offered were similar to working with a more experienced mentor. Gemini was right in many cases. I didn’t add data like customer satisfaction scores, for example, or referrals. Nor did I take a shot at adding customer acquisition costs. That’s the type of feedback a more experienced marketer might deliver in a case like this.
My current plan is to keep the data on the 150 fictional customers and add to it. I’ll continue to ask Gemini to give me insights and suggestions. I can’t wait to see what I learn along the way.
Dig deeper: Meet my research team: Gemini, ChatGPT and Perplexity
