Overview
Hostfully is a San Francisco based, venture-backed SaaS company for the vacation-rental and short-term-rental market. It runs two products: a Property Management Platform (PMP) subscription and a digital guest-experience product called Guidebook. As the business scaled, leadership wanted a rigorous, data-backed view of who their customers really were, which ones drove recurring revenue, and where pricing and packaging could be improved. They engaged Winston Francois for a fixed-scope customer-segmentation analysis.
The mandate:
- Understand which customer segments and product tiers actually drive monthly recurring revenue
- Replace assumptions with a defensible, data-backed segmentation of the active customer base
- Leave the team with a repeatable model they could re-run in-house on new data
The challenge
- Customer data was fragmented across Salesforce (Account and Opportunity), Stripe, and Braintree, with no single customer-level view.
- The raw dataset held more than 600 fields with inconsistent naming, duplicates, and no shared data dictionary, so nobody could easily separate signal from noise.
- It was unclear which product tiers and customer types were carrying revenue, and why the upsell motion from Starter to Pro to Premium was not behaving as expected.
- The team had no repeatable way to re-segment cohorts, such as churned users, without starting the analysis from scratch.
The approach
1. Unify and clean the data
- Pulled raw exports from Salesforce, Stripe, and Braintree and reconciled them into one customer-level dataset.
- Filtered to active subscribers, removed churned and canceled records, and resolved duplicate and mismatched fields flagged during cleaning.
- Worked with Hostfully's analyst to build a glossary defining the significant fields and calculations.
2. Explore, then segment with K-means
- Ran exploratory data analysis on both products, testing results with and without outliers and using Winsorized data to keep findings honest.
- Applied K-means clustering to produce three PMP segments and two Guidebook segments, each profiled by field averages, counts, and the features that most influence MRR.
- Assigned every customer to a cluster so the team could target segments directly.
3. Deliver readouts and a self-serve model
- Packaged Jupyter notebooks (cleaning, transformation, EDA, and segmentation), two executive summaries, detailed process documentation, and a Power BI file with dedicated PMP and Guidebook pages.
- Built the notebooks as a repeatable script so Hostfully could re-run the analysis on future snapshots, including churned cohorts, without WF.
- Led a findings readout and a follow-up session focused on business implications and a pricing-tier exercise.
The results
| Metric | Change | Impact |
|---|---|---|
| Data unified | 600+ fields, 3 systems | A single, analyzable view of the active subscriber base |
| Segmentation delivered | 5 segments (3 PMP, 2 Guidebook) | Every active customer mapped to a segment for targeting |
| Revenue drivers surfaced | Premium & Starter drive MRR | Framed a concrete pricing and packaging question, including a possible new tier, as Pro and Pro+ lagged |
| Repeatable model handed off | Notebooks + Power BI | Hostfully can re-segment cohorts such as churned users in-house |
The analysis did what good segmentation should: it gave Hostfully a defensible, customer-level map of its business and pointed leadership straight at a pricing and packaging conversation, backed by a model the team now owns and can re-run.
The work was a fixed-scope analytics engagement, so the value shows up as clarity and validated direction rather than a campaign performance number.
“Winston Francois helped us analyze our customer base and growth strategy, and while their findings largely confirmed our assumptions, their thorough approach gave us the validation we needed to move forward confidently. They're on the premium side, but you get what you pay for: high-quality strategic thinking and a fantastic partnership for early-stage startups.”Margot SchmorakCo-Founder & CEO, Hostfully
What we learned
- Segmentation is only as strong as the pipeline beneath it. The heaviest lift was unifying and cleaning fragmented data across three systems, not the clustering itself.
- Confirming a team's assumptions has real value. Validation gave the client the confidence to act on pricing decisions.
- Handing over a repeatable, self-serve model extends the impact of a project well past the final readout.
- The clearest strategic signal was in pricing and packaging, not acquisition, which is exactly where the data told us to look.