Last Wednesday, I Was Still Stuck on a Customer Segmentation Job

At 9 p.m. last Wednesday, I was sorting more than 2,000 customer records in a Hangzhou café, trying to group them by purchase frequency and activity level. A standard clustering program had been running for 30 minutes without finishing, so I had to shut it down. I also caught myself wondering: are tools like this even useful to nontechnical founders?

What Is E-means? It Doesn’t Make Decisions for You

E-means is an improved version of K-means clustering. Its core advantage is reducing repeated calculations so nearby data points can be grouped faster. The paper reports up to a tenfold speed improvement at the same loss. Lin Xiao, a freelance consultant in Chengdu, once used it to divide 12,000 course-user records into six groups and identify dormant and high-frequency users the same day. She still reviewed every group before deciding whom to contact. In other words, E-means speeds up the computation; it does not decide whether a customer is worth reaching out to.

What It Took Me to Replicate This: Zero to Low Cost, About Two to Four Hours

Money: The code is free. If I use the author’s cloud-hosted service, the actual price will depend on the current page, so I’d set aside roughly ¥50. Time: Preparing the data took about one to two hours, and the first run plus validation took another one to two hours. Technical barrier: Spreadsheet skills are enough to start, but interpreting the results requires business experience. First step: Open the project page, enter the code repository, and download the run instructions. If I don’t want to install an environment, I can try the author’s online demo and upload anonymized sample data. I would never upload customer names, phone numbers, or addresses directly.

I’d Test It Based on the Stage I’m At

If I’m just getting started and serve only a few customers each week, I’d keep using Excel for segmentation rather than adding learning overhead just to be “more advanced.” Not trying it now is completely fine. If I already have one or two stable clients and have accumulated thousands of records but don’t know how to code, I’d first run a test with 500 anonymized records, then manually compare the groups. If I’m scaling and need to process tens of thousands of behavioral records every week, I’d put E-means into testing, compare its speed and grouping differences with my current process, and then decide whether to replace that process. I wouldn’t look only at whether it runs in ten minutes. I’d also check whether the groups reflect the way I actually understand my customers.