← overflowvoice.comOne-page version →
BetterFlow case study · public data · 2015–2025
We tested a simple claim on 10 years of public data across 8 hurricane-belt states: when a storm hits, homeowners rush to search for roofers — a sharp, short-lived surge that lands exactly when local roofers are too swamped to pick up the phone. The data backs it, hard.
Averaged across 221 major storms, urgent roofing searches (roof leakroof damageemergency roof repair) nearly double the week a storm strikes, peak immediately, and fade to normal within about three weeks. Planned-roofing searches (new roof cost) barely move — so the spike is urgency, not curiosity. Bigger storms drive bigger spikes. A placebo test on random calm weeks shows no spike at all, and the pattern holds in all 8 states. For a roofer, that three-week surge is a wave of ready-to-hire homeowners hitting a phone line that's already ringing off the hook.
event-study · 221 storms · weeks −8 to +8
Each storm is aligned to "week 0," and we track the search index for 8 weeks before and after, indexed so each series' own pre-storm level = 100. Emergency searches leap; planned searches stay flat.
Emergency-search lift in the storm week, by state (states with lower everyday search volume show the largest percentage jumps):
The biggest named storms produced the sharpest surges — the peak emergency-search index (0–100) in the landfall week:
| Storm | State | Landfall week | Property damage | Peak index |
|---|---|---|---|---|
| Hurricane Ian | Florida | 2022-09-25 | $15.4B | 49 |
| Hurricane Laura | Louisiana | 2020-08-23 | $10.7B | 8 |
| Hurricane Michael | Florida | 2018-10-07 | $8.9B | 7 |
| Hurricane Ida | Louisiana | 2021-08-29 | $8.6B | 25 |
| Hurricane Harvey | Texas | 2017-08-20 | $7.0B | 46 |
| Hurricane Helene | Florida | 2024-09-22 | $4.2B | 12 |
| Hurricane Florence | North Carolina | 2018-09-09 | $1.7B | 50 |
| Hurricane Irma | Georgia | 2017-09-10 | $77M | 31 |
✓ Placebo test. Running the same event-study on random calm weeks (no storm) produces no spike — the effect only appears around actual storms, so it isn't a seasonal or data artifact.
✓ Consistency. All 8 states are positive, from +55% to +292%. ✓ Clean data. 149,597 NOAA storm events and 92,848 weekly search readings, reconciled and tested at every step.
Using the +95% search lift as a stand-in for the inbound-call surge: a roofer fielding ~25 calls a week at a ~75% answer rate faces roughly double the volume in a major-storm week while the answer rate drops under the load — leaving ~20–25 missed calls that week. A voice agent that catches ~70% of the overflow recovers ~15–18 leads; at a ~25% lead-to-job rate and ~$10,000 per job, that's on the order of $40k–$70k in recovered revenue per major-storm week — and more for a named hurricane. These are illustrative assumptions, not measured results — the point is the shape: a big, brief, catchable surge.
This is a fully reproducible pipeline on 100% public data — no client data, no black boxes:
Data. NOAA Storm Events Database (every recorded storm, 8 states, 2015–2025) joined to Google Trends weekly search interest for ~19 roofing keywords (acquired via the DataForSEO API). Loaded into a Snowflake warehouse.
Modeling. A medallion pipeline (RAW → CLEAN → ANALYSIS) in dbt: storm damage parsed to dollars, events categorized into roof-relevant perils (wind, hail, tornado, tropical, heavy rain), searches grouped into emergency / hire / planned clusters, everything conformed to a state × week grain and tested at each layer.
Analysis. A pooled event-study across 221 major-storm episodes measuring the search lift relative to a pre-storm baseline, with placebo and per-state robustness checks.
Honest caveats. Google Trends is a relative interest index, not a call count (so the business figures above are illustrative). The rarest emergency terms are sparse week-to-week, so we read them as cluster aggregates. Flood damage (e.g. Harvey's) is excluded from "roof-relevant" because flooding damages homes from below, not the roof.