As Fivetran has grown, so has the difficulty of understanding our customer experience across hundreds of connectors and thousands of accounts. When the product team kicked off an initiative on churn mitigation, we realized that while we had plenty of data, we didn’t know what was driving customer dissatisfaction and, worse, lacked a systematic way to determine the answer.

We had a problem of both variety and scale. Key customer signals were scattered across multiple systems. Conversations in Gong, tickets in Zendesk, and notes in Salesforce each told part of the story, but had to be pieced together and analyzed. Our PMs were spending too much time gathering and attempting to make sense of information, and not enough time actually mitigating churn. We needed to surface insights quickly and make them easy to act on.

You have to get the data, analyze it, and then act

To act on the data, we first had to get the data and turn it into insights. To do that, we built a simple workflow to synthesize customer signals and automatically highlight churn risks. It looks like this:

1. Ingest: Source data flows into BigQuery through Fivetran
2. Summarize: A lightweight JavaScript script calls the OpenAI API to generate a short summary of each conversion, request, or ticket, including the issue, severity, and potential churn risk
3. Synthesize: The summaries, paired with relevant account data and context, are added to a Google Sheet reviewed weekly by PMs
4. Alert: When a high-risk case is detected, a Slack alert automatically notifies the right stakeholders.

With this setup, our PMs gained near real-time visibility into churn signals using existing, off-the-shelf tools and without any engineering effort. What had been manual, time-consuming, and often futile became automated, quick, and practical.

This workflow required procedural changes for the PMs, as well. To ensure our signals turned into action, we established a lightweight rhythm: PMs would review the weekly insights sheet, add action items into a column, and coordinate next steps with support and sales teammates.

We couldn’t take for granted that the AI would always produce the correct conclusion. Sometimes, the action item would be to improve the output of the sheet, including through prompt engineering. This kept our feedback loops continuous and interventions timely.

While our workflow was originally built to help with churn mitigation, it also changed how the product team’s relationship with customer feedback. By connecting scattered signals and making them actionable, we helped PMs spend less time identifying and scrutinizing problems and more time solving them.

Look for small but impactful opportunities to use AI

Generative AI is an unparalleled tool for retrieving and synthesizing information, especially at scales that are impractical for humans. Our workflow here leverages the natural language processing abilities of generative AI without even requiring a RAG architecture. Your AI usage doesn’t need to be big or complex to make an impact. Our simple, lightweight system and spreadsheet have helped the Product team diagnose better, act faster, and serve customers more thoughtfully.

Your organization’s operations almost certainly offer similar opportunities to use AI. Any tasks that rely on generating, summarizing, and evaluating text or code are potentially strong use cases for leveraging the core abilities of generative AI, even with limited external context fetching. Some concrete examples include:

• Drafting reports and summaries from tickets
• Summarizing, evaluating, and drafting follow-up communication
• Generating descriptions and documentation
• Conversational analytics
• Proposing simulations and “what-if” scenarios

In all of these use cases, your AI projects will require considerable human supervision. AI complements, rather than replaces, human labor. Even so, our own use case demonstrates that a contained, well-defined AI project can lead to productivity gains even if it produces imperfect outputs (and besides, humans aren’t perfect, either).

It will take some experience, insight, and imagination to incorporate AI into existing operations effectively. But regardless of how the future of AI shapes up, there are plenty of opportunities to use AI productively today. 

[CTA_MODULE]