Data science and analytics consulting for sustainability-focused companies. Measure what matters. Zola 2024-04-20T00:00:00+00:00 https://polycultureresearch.com/atom.xml 2024-04-20T00:00:00+00:00 2024-04-20T00:00:00+00:00 Devon https://polycultureresearch.com/projects/customer-segmentation/ <p>An e-commerce company wanted to move beyond basic demographics to truly understand their customers. Behavioral clustering revealed five distinct customer types with different needs.</p> <h2 id="the-challenge">The Challenge</h2> <p>An e-commerce company selling sustainable home goods wanted to move beyond basic demographic targeting. Their email campaigns treated all customers the same, and their product recommendations weren’t driving the engagement they expected.</p> <p>Goals:</p> <ul> <li>Understand distinct customer types in their base</li> <li>Personalize marketing and product recommendations</li> <li>Identify highest-value segments for acquisition focus</li> </ul> <h2 id="the-approach">The Approach</h2> <h3 id="phase-1-data-preparation">Phase 1: Data Preparation</h3> <p>Gathered behavioral data across purchase behavior, browsing behavior, engagement, and customer context.</p> <h3 id="phase-2-clustering-analysis">Phase 2: Clustering Analysis</h3> <p>Applied k-means clustering to find natural groupings, revealing five distinct customer types:</p> <ol> <li><strong>Conscious Newbies (23%)</strong> — Recent customers exploring sustainable living</li> <li><strong>Mission-Driven Loyalists (18%)</strong> — Long-term customers with high LTV</li> <li><strong>Gift Givers (21%)</strong> — Seasonal purchase patterns, higher AOV</li> <li><strong>Category Specialists (25%)</strong> — Deep engagement in 1-2 categories</li> <li><strong>Bargain Hunters (13%)</strong> — Only purchase on sale, lowest LTV</li> </ol> <h3 id="phase-3-segment-activation">Phase 3: Segment Activation</h3> <p>Developed detailed persona documentation, email content recommendations, and product recommendation logic for each segment.</p> <h2 id="the-outcome">The Outcome</h2> <p><strong>Immediate impact:</strong></p> <ul> <li>Email click rates increased 34% with segmented content</li> <li>Product recommendation CTR increased 45%</li> </ul> <p><strong>Strategic insights:</strong></p> <ul> <li>Mission-Driven Loyalists were 4x more valuable than average but under-invested in retention</li> <li>Gift Givers represented untapped referral potential</li> </ul> <p><strong>Acquisition shift:</strong></p> <ul> <li>18% improvement in new customer LTV over 6 months by reallocating spend toward channels that brought in higher-value segments</li> </ul> 2024-03-05T00:00:00+00:00 2024-03-05T00:00:00+00:00 Devon https://polycultureresearch.com/projects/carbon-accounting/ <p>A B Corp manufacturer needed accurate Scope 3 emissions data for investor reporting. Built a carbon accounting system that tracked emissions across 200+ suppliers.</p> <h2 id="the-challenge">The Challenge</h2> <p>A certified B Corp manufacturer needed accurate Scope 3 emissions data. Their investors and customers were increasingly asking about supply chain sustainability, and their B Corp recertification required robust environmental metrics.</p> <p>The challenge was significant:</p> <ul> <li>200+ suppliers across multiple countries</li> <li>No standardized data collection from suppliers</li> <li>Multiple product lines with complex bills of materials</li> <li>Need for auditable, investor-grade reporting</li> </ul> <h2 id="the-approach">The Approach</h2> <h3 id="phase-1-emissions-mapping">Phase 1: Emissions Mapping</h3> <p>Mapped entire value chain to identify emission sources across Scope 1, 2, and 3. Scope 3 represented over 80% of total emissions.</p> <h3 id="phase-2-data-collection-system">Phase 2: Data Collection System</h3> <p>Built a tiered approach: direct measurement for top 20 suppliers, activity-based estimates for medium suppliers, and spend-based estimates for smaller suppliers.</p> <h3 id="phase-3-calculation-engine">Phase 3: Calculation Engine</h3> <p>Built a system that tracked emissions by supplier, material category, and product line with full audit trail.</p> <h3 id="phase-4-reporting-and-insights">Phase 4: Reporting and Insights</h3> <p>Delivered annual emissions reports aligned with GHG Protocol standards, product-level carbon footprints, and supplier sustainability scorecards.</p> <h2 id="the-outcome">The Outcome</h2> <p><strong>Data quality:</strong></p> <ul> <li>95% of emissions now measured or estimated with activity data (vs. 30% previously)</li> <li>Passed external audit for B Corp recertification</li> </ul> <p><strong>Business impact:</strong></p> <ul> <li>Identified $200K+ annual savings through logistics optimization</li> <li>Two major suppliers committed to emissions reduction plans</li> <li>12% reduction in emissions in first year</li> </ul> 2024-02-15T00:00:00+00:00 2024-02-15T00:00:00+00:00 Devon https://polycultureresearch.com/posts/welcome/

<p>I’m excited to launch this blog as a space to share insights at the intersection of data science and sustainability.</p>

2024-02-10T00:00:00+00:00 2024-02-10T00:00:00+00:00 Devon https://polycultureresearch.com/projects/churn-prediction/ <p>A subscription business was losing customers without understanding why. Predictive modeling identified at-risk accounts and the factors driving churn.</p> <h2 id="the-challenge">The Challenge</h2> <p>A B2B SaaS company with 500+ customers was experiencing higher-than-expected churn. They knew they were losing accounts, but couldn’t predict which customers were at risk or intervene before it was too late.</p> <p>Their customer success team was stretched thin and needed to prioritize their time. They wanted to:</p> <ul> <li>Identify at-risk accounts before they churned</li> <li>Understand the leading indicators of churn</li> <li>Create a systematic approach to customer health</li> </ul> <h2 id="the-approach">The Approach</h2> <h3 id="phase-1-data-exploration">Phase 1: Data Exploration</h3> <p>Built a historical dataset of churned vs. retained customers going back 18 months, with features capturing behavior in the 90 days before churn/renewal.</p> <h3 id="phase-2-feature-engineering">Phase 2: Feature Engineering</h3> <p>Created features across usage patterns, engagement signals, and business context.</p> <h3 id="phase-3-model-development">Phase 3: Model Development</h3> <p>Tested several approaches and chose a gradient boosting model for its balance of accuracy and interpretability. Key focus was on practical utility, not just accuracy.</p> <h3 id="phase-4-deployment-and-integration">Phase 4: Deployment and Integration</h3> <p>Delivered weekly risk scores for all accounts, top 3 factors contributing to each account’s risk, and integration with their CRM.</p> <h2 id="the-outcome">The Outcome</h2> <p><strong>Model performance:</strong></p> <ul> <li>Identified 73% of eventual churners in the “high risk” category</li> <li>60+ day lead time for most flagged accounts</li> </ul> <p><strong>Business impact:</strong></p> <ul> <li>22% reduction in churn rate over 6 months</li> <li>CS team focused on accounts that actually needed attention</li> <li>Identified two product issues that were driving churn for a specific customer segment</li> </ul> <p><strong>Key insight:</strong> Champion user departure was the strongest single predictor of churn.</p> 2024-01-15T00:00:00+00:00 2024-01-15T00:00:00+00:00 Devon https://polycultureresearch.com/projects/zero-to-series-a/ <p>A fast-growing consumer app needed to go from spreadsheets to a real data stack before their Series A. Built complete analytics infrastructure in 8 weeks.</p> <h2 id="the-challenge">The Challenge</h2> <p>A consumer mobile app was preparing for their Series A raise. They had strong product-market fit and impressive growth, but their data situation was a mess: metrics lived in spreadsheets, different team members had different numbers, and investors were asking questions they couldn’t answer confidently.</p> <p>They needed:</p> <ul> <li>A single source of truth for all key metrics</li> <li>Self-service dashboards for the team</li> <li>Reliable data for investor due diligence</li> <li>Foundation to build on post-funding</li> </ul> <p>The timeline was tight: 8 weeks until investor meetings.</p> <h2 id="the-approach">The Approach</h2> <h3 id="week-1-2-discovery-and-design">Week 1-2: Discovery and Design</h3> <p>Started by mapping out all existing data sources: the mobile app’s event tracking, payment processor, CRM, and marketing platforms. Interviewed each team lead to understand what questions they needed to answer.</p> <h3 id="week-3-5-infrastructure-build">Week 3-5: Infrastructure Build</h3> <p>Set up a modern data stack:</p> <ul> <li><strong>Snowflake</strong> as the data warehouse</li> <li><strong>Fivetran</strong> for automated data ingestion</li> <li><strong>dbt</strong> for transformations and data modeling</li> <li><strong>Mode</strong> for dashboards and ad-hoc analysis</li> </ul> <h3 id="week-6-7-dashboard-development">Week 6-7: Dashboard Development</h3> <p>Created three main dashboards: Executive Overview, Growth Dashboard, and Revenue Dashboard.</p> <h3 id="week-8-training-and-documentation">Week 8: Training and Documentation</h3> <p>Trained the team on how to use the dashboards and documented everything.</p> <h2 id="the-outcome">The Outcome</h2> <p><strong>Immediate impact:</strong></p> <ul> <li>Single source of truth for all metrics</li> <li>Team could answer investor questions in minutes instead of hours</li> <li>Data issues caught automatically before affecting decisions</li> </ul> <p><strong>For the Series A:</strong> The founder reported that data quality and accessibility was specifically called out by investors as a sign of operational maturity. They closed their round 3 weeks after initial investor meetings.</p>