Laxmi
Sharma

Telecom companies were losing their highest-value customers silently — no system existed to identify who was about to leave or why. Retention teams were reactive, not proactive.

The most profitable customers are the highest-risk. Month-to-month contract holders churn at 42.7% — versus just 2.8% for 2-year plans. Churned customers paid 21% higher monthly charges. Revenue and risk moved together.

A full ML prediction pipeline (7,032 records) producing a churn-risk list exportable directly to a CRM or retention team — plus a 3-page Power BI dashboard with contract-type and tenure filters for ongoing monitoring.

• High recall matters more than accuracy for churn — missing a churner costs more than a false alarm
• Business framing beats statistical framing when presenting to stakeholders
• Feature engineering (tenure buckets, charge ratios) improved model signal more than tuning

Banks were approving loans using aggregate credit scores that masked behavioral risk. A $312M portfolio had no segment-level default visibility — decision-makers couldn't see which borrower profiles were silently failing.

Credit score alone is a weak predictor. Loan grade, income-to-loan ratio (LTI), and credit history length together outperform it. LTI greater than 4x correlates with a 35% default rate. 23% of the portfolio was flagged high-risk — actionable, not theoretical.

An end-to-end SQLite pipeline (ingestion → cleaning → segmentation → KPIs) and a Power BI dashboard deployable by any credit risk or MIS team with zero additional setup — built to be handed off, not demoed.

• Pipeline design matters: building for handoff forces cleaner code than building for yourself
• Segmentation before modelling reveals patterns aggregate stats hide
• In BFSI, a false negative (missed default) has asymmetric cost — precision-recall trade-offs are business decisions

Comparing 14 global companies across 8 sectors over 15 years manually is analyst-hours wasted — and error-prone. No scalable way existed to surface cross-sector trends or flag early stress signals without rebuilding every time.

Leveraged sectors showed diverging risk trajectories post-2016 — a stress signal visible only in multi-year cross-sector view. Apple's ROE hit 61.3%. Built a custom Financial Strength Index (FSI): ROE×0.30 + Net Margin×0.30 + Current Ratio×0.20 − D/E×0.20, scoring all 14 companies in one comparable metric.

A 3-page Power BI report — revenue trends, profitability deep-dive, leverage vs risk scatter — mirroring equity research output quality. Refreshable with new data in minutes. No rebuild required.

• Composite indices (like FSI) force clarity on what "financial health" actually means — good for stakeholder alignment
• 15-year longitudinal data reveals structural trends that quarterly snapshots mask
• Scatter plots (leverage vs. risk) communicate risk-return trade-offs faster than tables ever will

Retail management tracked top-line sales ($2.3M) but had no visibility into where margin was quietly being destroyed — by discount strategy, category, or region.

$92,000 in recoverable loss identified through discount-depth segmentation. Specific subcategories were discounted beyond the margin threshold — generating revenue while destroying profit.

A management-ready dashboard with drill-down by region, category, and discount tier — enabling the pricing team to act on findings immediately without touching the source data.

• MIS reports must answer "so what?" — not just show the number, but show the action it demands
• Discount analysis requires unit-level granularity — averages hide the worst offenders

Operations teams tracked fulfilment and revenue separately — no integrated view connected supplier quality, lead time, and revenue outcomes in one place.

Supplier rating averaged 2.86/5 — below threshold — with a direct correlation to revenue concentration in high-performing SKUs worth $582K. Underperforming suppliers were concentrated in specific product routes.

An operations dashboard linking supplier scores, delivery timelines, and revenue by product and region — giving procurement and ops a single source of truth.

• Cross-functional data (supplier + revenue + logistics) tells a richer story than any silo alone
• Operational dashboards need refresh logic built in from day one — not retrofitted