A Practical ML Segmentation Story: 20K Customers, 500 Reactivations, $300K Margin

The Indonesia marketing team had a problem that will feel familiar to anyone who has tried to do serious analytics in spreadsheets: they had plenty of customer data, but the workflow couldn’t scale.

Their SHARE app captured rich ordering and promotion data across roughly 20,000 workshops. In theory, that dataset could answer critical questions—Which promotions actually change behavior? Which customers are worth reactivating? What kind of campaign works best for which segment? In practice, the team was stuck in a loop of manual downloads, Excel-based slicing, and repetitive analysis that consumed time and still left a lot of insight on the table.

The bigger issue wasn’t effort—it was capability. Excel can help you summarize what already happened, but it’s not built to surface patterns across tens of thousands of customers, especially when you want the patterns to be stable enough to drive action. And without machine learning, it was hard to go beyond basic descriptive cuts into something more strategic: clear customer archetypes the marketing team could plan around.

A pragmatic approach: “automation where it matters”

Before building anything, I worked with the local marketing team to pressure-test the use case. The insight was important, but the expected usage frequency didn’t justify an expensive, fully automated pipeline from day one. So we designed a solution that was intentionally practical:

• Data could still be manually exported (keeping effort and complexity low)

• Model runs could be triggered on demand (when the team actually needed refreshes)

• Outputs needed to be simple enough to trust and operationalize

With product development leads across Software Engineering and Data Science, we built an ML-driven segmentation workflow using K-Means clustering. The goal wasn’t “AI for AI’s sake.” It was to take a large, messy customer base and group it into logical clusters based on meaningful behavioral and commercial attributes — clusters that could translate into distinct promotional strategies.

Making the model usable (and trusted)

Segmentation only creates value if teams understand and use it. So we ran multiple working sessions with marketing stakeholders and the Data Science function to explain:

• what each cluster represented,

• why certain workshops grouped together,

• and how to interpret the model’s outcomes in plain business terms.

That step—translation and trust-building—was essential. Without it, the model would have remained “a DS artifact” rather than a marketing tool.

Finally, we visualized the segmentation outputs in a lightweight PowerBI view so the business could track outcomes, review changes over time, and incorporate the insights into planning cycles.

Results: from archetypes to action

Once the clusters were in place, the marketing team moved from broad, generic promotions to targeted programs by customer archetype. This unlocked two clear outcomes:

• Reactivation of 500+ inactive workshops through tailored campaigns

• Approximately $300K incremental margin over 12 months

Just as importantly, the project proved a broader point internally: machine learning could create real commercial leverage beyond manual analysis. That credibility helped the capability get included in the local digital roadmap—shifting it from a one-off initiative to a repeatable part of the ecosystem.

In the end, this wasn’t a story about building the most sophisticated platform. It was about building the right level of intelligence—at the right cost and complexity—so a business team could make better decisions, faster, and with measurable impact.

Key takeaways

• Spreadsheet-driven analytics hit a hard ceiling when datasets grow and pattern-finding becomes the real need.

• Customer segmentation becomes valuable when it produces actionable archetypes, not just clusters on paper.

• The “right” solution isn’t always full automation — on-demand modeling can deliver strong ROI with lower complexity.

• Trust and adoption are built through explanation and stakeholder working sessions, not just dashboards.

• Targeted campaigns by segment drove reactivation (500+ workshops) and ~$300K incremental margin in 12 months.