Summary

  • Can ChatGPT help Customer Success be more productive?

    • Answer: Yes, if used consistently.

  • There is already a lot of online content about using ChatGPT to write emails and other corporate fodder. We’ll review this briefly, but my perspective is these use cases offer limited, incremental value until you get really good at prompt engineering.

    • This article focuses on higher-value use cases for the smaller, adaptive startups our team advises for and invests in.

  • It will take time for CS teams to establish the habit—cue, routine, reward—of consistently using ChatGPT in their day-to-day. Like most user-facing software, workflow is king. Things like using ChatGPT in Slack and/or adding ChatGPT to users’ Bookmark Bar in Chrome might help keep it top of mind.

Advanced Data Analysis

I’m going to skip straight to the good stuff. For $20/month it’s worth upgrading to ChatGPT “Plus” to gain access to Advanced Data Analysis and Plugins. You can then hover over GPT-4 and select “Advanced Data Analysis” or “Plugins". It looks like this:

Now you’re ready to cook.

As the name implies, ChatGPT Advanced Data Analysis can interpret, clean, and analyze 100MB-or-less data sets in minutes.

It won’t replace the best BI analysts and dashboards, but it’s better than the average ones. So far I’ve used it to analyze personal spreadsheets (goals, budgets), product usage trends, and government data.

For example, I fed ChatGPT a spreadsheet with my personal goals, attainment, activities, etc. and it swiftly analyzed them, identified insights, and generated the below charts.

It had a little trouble with an undefined header which it missed during cleaning, but all the underlying data was spot on. This will save me 2-3 hours per quarter going forward. Not life-changing, but certainly useful.

Here are the file types ChatGPT Advanced Data Analysis can work with—in descending performance (best to worst):

  • Data (.csv, .xlsx, .tsv, .json, etc.)

  • Code (.py, .js, .html, .css, etc.)

  • Text (.txt, .csv, .json, .xml, etc.)

  • Document (.pdf, .docx, .xlsx, .pptx, etc.)

  • Image (.jpg, .png, .gif, etc.)

  • Audio (.mp3, .wav, etc.)

  • Video (.mp4, .avi, .mov, etc.)

Warning: do not upload personal or sensitive customer information. OpenAI is not perfect—no software company is.

Sample data

If you want some sample data to play around with, here you go:

  • Amplitude 10-Q (pdf)

  • Fake employee data (xls, csv)

Prompts with Plugins/Extensions

ChatGPT Plugins and (browser) Extensions help fill in the feature gaps that OpenAI has deprioritized, e.g. internet access, real-time information, automation, image creation, uploading PDFs, etc.

Here are a few of the most-valuable use cases I’ve found for Customer Success, including a) the plugin/extension used, b) prompt, and c) perceived value.

1) Use case: Save time by uploading a lengthy document for ChatGPT to read and summarize for you.

a) Chrome extension: ChatGPT File Uploader (adds the green upload button below)

b) Prompt: Please summarize this document for me: key trends, business risk, and opportunities.

c) Perceived value: ChatGPT helps me understand important customer documents quicker, e.g. 10-Q, product announcements, press releases, even blog posts. Humans read about 280 words/minute. The Amplitude 10-Q is 43,847 words, so it would take ~160 minutes to read completely. 10-Qs have lots of fine print that are often skimmed, so let’s round down to 2.5 hours—just to read it. ChatGPT read/summarized the 10-Q it in 51sec + it took me 1min 31sec to read for a total of 2min 22sec. So from a text-based comprehension standpoint, ChatGPT helped perform the task in 1.5% of the time—inversely, a 98.5% time savings.

To try this: 1) download the PDF, e.g. Amplitude’s recent 10-Q linked above, and 2) upload it to ChatGPT for analysis. The prompt/completion looked like this:

Full response at bottom of article

Caveats: having worked on Wall St. in a past life, I also performed my own analysis of the 10-Q. I would rate the quality of ChatGPT’s summary a B/B+. It captured most—but not all—the salient points. ChatGPT also did not compute independent financial analysis of reported metrics. This would take ChatGPT—and a CSM, analyst—more time to develop deeper comprehension.

2) Use case: create v1 visual content for internal and customer use.

a) Plugin: Show Me

b) Prompt: Show me a diagram of "New Customer Onboarding" that could be used in B2B SaaS to help guide customers to learn, use and adopt a new software.

c) Perceived value: What ChatGPT provided. Here we get a glimpse of ChatGPT’s ability to create reusable business assets which are a bit more sophisticated than “write me an email.”

Caveats: the visuals will likely be generic and off-brand. Additional time would be required to customize and brand the assets in-line with your company’s product and brand guidelines.

Other example prompts

I promised to briefly touch on the widely-circulated ChatGPT use cases for Customer Success. Here are a few prompts that can be useful:

  • Write an email to a B2B startup customer who’s usage recently dropped 25%.

  • Create a QBR outline for an airline customer.

  • Role play a tense customer call who wants to downgrade due to budget reasons.

  • What are some under-appreciated customer discovery questions?

  • What are some novel change management tactics for larger, slower enterprise customers?

  • Generate a python script for a mass email.

  • Proofread this email for clarity and brevity.

Use case framework

I’ve been playing around with a way to better structure and simplify the near-infinite ways to use ChatGPT for work applications. So far this is the best I’ve come up with:

Use C.A.S.E. framework:

  • C: Create & Coach

  • A: Ask

  • S: Summarize

  • E: Execute

Here’s a visual broken out by function:

A note on fine-tuning

Fine-tuning involves training a model on new data to improve knowledge in specific areas, e.g. a company’s intranet, or proprietary data set. You will hear a lot about this in the coming months.

For ChatGPT Enterprise/API-based customers, it’s worth noting that fine-tuning inherently changes the base model. While fine-tuning is a powerful way to expand the training set—and by extension, use cases—there are serious cost/benefit considerations, e.g. do you have in-house AI/ML engineers, ample time, money + the opportunity cost of ongoing maintenance?

For enterprise customers early on in their AI journey, my advice would be to focus on expanding & optimizing prompt engineering before jumping to advanced fine-tuning.

Crawl, walk, run.

Go deeper

It’s no surprise there is a lot of content on using the fastest-growing app in history. I’ve tried to capture some of the better Customer Success-related articles here, in descending usefulness:

Full ChatGPT analysis of Amplitude 10-Q filing

The value for the CSM here is they don’t have to read an 87-page document which could take several hours. Instead, you get a capsule summary in several seconds. Tradeoff: the CSM might miss an undisclosed insight in the actual data/numbers, along with semantic context.

Recap

Overall, ChatGPT seems likely to increase productivity for Customer Success. Like most software, workflow is king. It will take time for teams to establish the habit (cue, trigger, reward) of consistently using ChatGPT in their day-to-day, e.g. using ChatGPT in Slack, adding ChatGPT to users’ Bookmark Bar in Chrome, etc.

We hope you found these prompts and use cases valuable.