Listening Where Customers Already Are

Overview

My team runs formal studies, surveys, moderated interviews, usability tests, but we had a gap: what customers say when they are not talking to us. Business customers discuss their telecom experiences openly on Reddit, in communities for wireless carriers, IT administration, and small business operations. I built a tool that systematically collects those conversations, classifies each one, and surfaces patterns in a dashboard, so the team has an ongoing, shared view of unmoderated customer voice instead of the occasional manual scroll. I built it with an AI coding assistant and no coding background, and the value is in the research decisions, not the code.

The problem

The signal was there, but we had no systematic way to collect, categorize, or track it over time. Checking Reddit by hand was something any of us could do on a given afternoon, but that is sampling, not monitoring. It does not build a dataset, it does not surface patterns across weeks, and it does not give the team a shared view of what customers are saying. I wanted a tool that could do the collection and categorization, so I could spend my time on interpretation and application, which is where the research value actually lives.

What I built

I described what the tool should do in plain language, and an AI coding assistant generated the parts I do not write myself: the scraping, the analysis engine, the database, and the dashboard. I made the research decisions. I chose the communities where business customers and prospects actually post, and defined search terms that catch the abbreviations and informal names a narrow keyword search would miss. The tool pulls the last thirty days on each run and only fetches new posts, so it builds a cumulative dataset rather than re-scraping the same content.

For classification, I defined categories grounded in the questions my team works on. Each post is tagged on four dimensions: sentiment; customer status, meaning current, former, prospective, or non-customer; theme, from network coverage and customer service to pricing, billing, onboarding, switching, business features, devices, support, and competitor comparison; and whether it is about the business or the consumer side. Posts with negative or mixed sentiment also get their specific complaint sentences pulled out for quick scanning. The dashboard has summary metrics, a filter for every dimension, theme and sentiment charts, and a scrollable feed of post cards that link back to the original thread, plus a PDF export for colleagues who do not run the tool themselves.

What it surfaces

The customer-status tag lets me separate a current customer’s friction with a product they already use from a prospect weighing us against competitors. Those are different research problems, and treating them the same would lead to misplaced priorities. The theme tags show whether complaints cluster, which points me toward where the strongest pain is, even without a formal study; I treat those as directional signals for where to dig deeper, not findings on their own. And the business-versus-consumer filter is especially useful, because most public conversation is consumer-focused and the business voice, fleet management, multi-line administration, business portal experiences, otherwise gets buried under consumer noise.

What I learned

The barrier between a researcher having an idea for a tool and that tool existing has dropped dramatically. That changes what a team can do, because tools that used to need an engineering request or a vendor purchase can now be prototyped by the researcher who understands the problem best. But I am careful about the framing. The tool automates collection and applies rules-based classification. It does not do research: it does not decide what matters, it does not interpret a pattern, and it cannot tell when a post that looks negative is actually a sign of high expectations from a loyal customer. Those judgments require a researcher.

I was also deliberate about security and data governance. Because I work on a corporate machine with specific policies, I evaluated every sharing and deployment option against what was permitted before pursuing it, and chose to share the tool through live demos and screen recordings rather than external deployment. When researchers start building tools, we need to bring the same care to data handling that we bring to participant privacy in formal studies.

Reflection

Social listening expands whose voice gets counted. Reddit users are not a representative sample of all business customers, but they include people who would never sign up for a research panel or respond to a survey, some of them posting precisely because the formal feedback channels did not work for them, or because they did not know those channels existed. Building a tool that captures those voices, even imperfectly, is one way to make the research practice more inclusive about who gets heard. That is the same commitment that runs through the rest of my work.