Why this exists. The finance deck led with "hygiene comp up as % of total revenue (+70bps)" — a scary headline. That denominator is contaminated by a $97K YoY drop in dentist production the comp plan doesn't touch. This rebuild isolates the metrics the plan actually controls, and nets out network-wide trend & seasonality with a control group the deck lacked.
01Verdict
Bottom line: Every productivity metric moved the right direction vs control — but with 5 offices and one post-month, none clears statistical significance. The plan is not making hygiene more expensive per unit of output (that was a dentist-revenue artifact). The lift is real-but-small and concentrated in revenue-per-visit (intensity), not visit volume. Richmond Hill — the only office with 3 months of exposure — shows the strongest, most consistent signal, suggesting the effect builds over time. Don't expand on this evidence; extend the window.
02Primary diff-in-diff — YoY May 2026 vs 2025
DiD = (treated change) − (control change). Cohort-aggregate (size-weighted) point estimate; permutation p-value from 10,000 random treated/control reassignments of office-level YoY changes.
| Metric | Treated 25→26 | Control 25→26 | DiD | Perm p | Read |
03Parallel trends — treated vs control
Monthly cohort aggregates, Jan 2024 – May 2026. Dashed gold line = pilot go-live (Mar 2026 Richmond Hill; May 2026 remainder). Stable pre-period gap → DiD assumption holds; post-period divergence → pilot effect.
Hygiene production / hygienist-day
The core lever — does paying 30% raise daily output?
Revenue per hygiene visit (RPV)
Intensity per visit — where the lift actually is
Visits per hygienist-day (VpD)
Throughput — flat = no extra patients seen
Hygiene % of total production
Rises partly from the dentist-revenue drop — read with care
04Richmond Hill — 3-month exposure (live since March)
The only office past one post-month. Mar–May 2026 vs Mar–May 2025, vs the same 46 controls. Ratio DiDs shown (level DiD omitted — 1 office vs 46-office aggregate isn't comparable).
| Metric | RH 25→26 | Control 25→26 | DiD |
|---|
Richmond Hill is the strongest case in the dataset: hyg production/day DiD +$213, RPV DiD +$24, VpD DiD +0.32 — all larger than the 1-month all-office cut. Consistent with an effect that compounds as hygienists adapt to the incentive. This is the argument for a full pilot window before any go/no-go.
05The compensation ratio — and the data gap
The exec success metric is "hygiene comp flat/down as % of revenue." Hygienist wage dollars are not in the Power BI ops model, so the comp ratio below is finance-reported (the deck), treated-only — it has no control group. Pulled onto the correct denominator (hygiene production, not total revenue).
Deck — all 5 offices, May YoY
Hyg wages ÷ hyg production: 40.7% → 39.8% −0.9pp
Deck — Richmond Hill, Mar–May YoY
Hyg wages ÷ hyg production: 31.7% → 30.8% −0.9pp
Gap to close before this is decision-grade: the comp ratio cannot be diff-in-diff'd without hygienist wage dollars by office-month for the 46 control offices (source: Workday / Paylocity, not PBI). Until then the −0.9pp is a treated-only pre/post — directionally good, but un-attributable to the plan vs network hygiene-wage trend. The production/productivity DiD above is control-adjusted; the comp ratio is not.
06What's needed for a real go/no-go
- Extend the window. ≥3 post-months on all offices (ideally a full seasonal cycle). The 1-month all-office cut is underpowered — perm-p 0.34–0.83 — and the "higher of hourly" floor can't be evaluated across soft months yet.
- Wage dollars for control offices. Pull hygienist comp by office-month (Workday/Paylocity) for the 46 controls to make the comp ratio a true DiD on the correct denominator.
- Decompose the RPV lift by procedure mix. The gain is intensity (RPV), not volume (VpD flat). Confirm it's legitimate perio/SRP need — not production-chasing — before scaling an incentive that pays 30% of production.
- Fix or formally drop Waverly. Its PBI data is broken (Apr = −$18, May = null) — exactly the deck's "data discrepancy." Reconcile the PMS feed or exclude it consistently across all reporting (don't silently drop a weak office).
- Validate the "not taxed as bonus" claim with payroll. A separate supplemental-wage check is still taxable wages; only the withholding method differs. Don't sell hygienists a benefit that may not survive payroll review.
07Method & limitations
Design. Treated = 5 pilot offices (Richmond Hill, Elite, Pryor, Rogers, Brentwood; Waverly excluded for broken data). Control = 46 SGA-East general-dentistry 4-Wall open practices in the same divisions (1 & 5), not in the pilot. Outcomes from Power BI ops dataset ([Net Production] split by Providers[Dr or Hyg], [Visits], hygienist-days = provider-days with >$100 production). YoY May 26-vs-25 nets out office fixed effects + seasonality; DiD nets out network trend.
Limits. (1) n=5 treated, 1 post-month for 4 of them → low power; point estimates are directional, not significant. (2) Comp ratio is deck-sourced, treated-only — not control-adjusted. (3) 6 remaining Parks-Pace offices sit in the control pool; if the pilot expands to them they'd contaminate the control. (4) Hygiene-%-of-production rises partly from the dentist-revenue drop — read productivity metrics (per-day, RPV, VpD) as the cleaner signal. (5) Permutation test assumes exchangeability of office-level YoY changes under the null.
Generated 2026-06-22 · Source: SGA Power BI bridge · Built by Scott Guest