Sample. Facts, counts and results are illustrative and describe no real employer or class. The structure, disclosures and limits are exactly what a delivered memo contains.

Settlement Analytics Memorandum

Off-the-clock work — common proof, sampling plan and aggregate damages for a putative class

To	[Counsel], [Firm]
From	Karim Souidi, M.Sc. (Econometrics), NorthLaw.ai
Matter	[Named Plaintiffs] v. [Retailer] — sample file NL-2026-0004 (FLSA collective / Rule 23 state class)
Date	25 September 2026
Purpose	Settlement and certification-strategy support. Privileged; prepared at the request of counsel. Not prepared for filing or testimony.

1. Bottom line

The employer's own systems record the alleged violation. Across a stratified random sample of 400 shifts from 46 stores, 71% show the employee logged into the point-of-sale system before clocking in, by an average of 9.4 minutes (95% CI 8.6 – 10.2). The pattern is present in every store (range 58% – 82%) and varies little between them, which supports common proof: the question "did this happen, and how much" can be answered for the class from records, not from 3,200 individual testimonies.

Aggregate unpaid time over the three-year class period is about 170,000 hours, worth $4.0 million in wages at the blended regular and overtime rate (95% CI $3.4M – $4.6M), before liquidated damages, interest and penalties, which are for counsel to apply. The employer's strongest arguments are a de minimis defense and the claim that pre-login time was not work; the first is quantified below (it removes about a fifth of the damages, not the case), the second is factual.

2. Questions presented

Can the existence and extent of pre-shift unpaid work be established for the class with common, representative evidence?
What is a defensible sampling plan, and what does the sample show?
What is the aggregate damages figure and its uncertainty?

3. Data and sampling plan

Three record systems were produced: the timekeeping system (clock-in/out), the point-of-sale system (first and last login per employee per day), and payroll. The class comprises 3,200 hourly sales associates across 46 stores, 2023 – 2025, about 5,100 employee-years and 1.06 million shifts. Matching the two systems for every shift is feasible and is recommended for the merits; for this memo, a sample was drawn to estimate the pattern and its variability quickly.

Design element	Choice	Reason
Frame	All 1.06M shifts with both a clock-in and a POS login	Complete, employer-generated, no self-report
Stratification	By store (46) and year (3)	Guarantees every store is represented; supports the commonality question directly
Sample size	400 shifts (≈ 3 per stratum, proportional)	Margin of error ± 4.5 points on a proportion; ± 0.8 minutes on the mean gap
Selection	Simple random within stratum, seeded; selection log retained	Reproducible by the other side

4. Results

Measure	Estimate	95% CI	Note
Shifts with POS login before clock-in	71%	66% – 75%	284 of 400 sampled shifts
Mean pre-clock-in time, affected shifts	9.4 min	8.6 – 10.2	Median 8 min; 90th percentile 19 min
Mean pre-clock-in time, all shifts	6.7 min	6.0 – 7.4	Includes zeros
Store-level affected share, range	58% – 82%	—	No store below 50%
Between-store share of variance (ICC)	0.04	—	96% of variation is within stores, i.e. shift-to-shift, not store policy

Figure 1. Every store shows the pattern; the spread across stores is narrow relative to the class-wide rate.

Aggregate damages. 5,100 employee-years × 4.1 shifts/week × 52 weeks × 6.7 minutes per shift ≈ 170,000 unpaid hours. At a blended rate of $23.30 (regular $19.40 with 20% of hours at the 1.5× overtime premium), $3.96 million (95% CI $3.4M – $4.6M). Sampling error accounts for the interval; the shift count and pay rates are from payroll and carry no sampling uncertainty.

5. Where the case is strong and where it is exposed

Strong on commonality: the violation is recorded in the employer's systems for every shift; the sample shows the same pattern in every store with low between-store variance. This is the situation Tyson Foods v. Bouaphakeo addressed: representative evidence of a common practice that each member could have used in an individual suit.
Strong on measurability: a full match of the two systems (recommended before the merits) replaces the sample estimate with a census and removes the sampling interval entirely.
Exposed, de minimis: if time under 5 minutes per shift is excluded, unpaid hours fall by about 22% to 133,000 and damages to $3.1M. If under 10 minutes is excluded (an aggressive reading), damages fall to $1.6M. The distribution matters: 41% of affected shifts exceed 10 minutes.
Exposed, "not work": whether a POS login before clock-in constitutes compensable work is a factual and legal question. The data show only that the login occurred and when.
Exposed, opt-in rate: for the FLSA collective, damages scale with the opt-in rate; the figure above assumes the full Rule 23 class.

6. Limits of this analysis

The sample estimates a class-wide pattern; it does not establish any individual's damages, which a full record match would.
Liquidated damages, penalties, interest, attorney's fees and the limitations period are for counsel to apply.
Records were produced by the employer and not independently verified; clock and POS system clocks were checked for offset (none found beyond 30 seconds).
Prepared for settlement and strategy; not an expert report under FRCP 26(a)(2)(B).

Karim SouidiM.Sc. Econometrics · DASCA Senior Data Scientist · NorthLaw.ai · karim@northlaw.ai

Appendix A — Technical results

A1. Sampling. Stratified simple random sample, 138 strata (46 stores × 3 years), allocation proportional to stratum shift counts with a minimum of 2; n = 400. Estimates use stratum weights; variances by the stratified formula with finite-population correction (negligible). Seed and selection log retained.

Table A2. Distribution of pre-clock-in minutes, affected shifts (n = 284).

Percentile	10th	25th	50th	75th	90th	Mean
Minutes	2	4	8	13	19	9.4

A3. Commonality diagnostics. Random-intercept logistic model of "affected" on store and year: store variance component 0.14 (ICC 0.04); likelihood-ratio test for store effects p = 0.09; year effects negligible. Interpretation: the practice is not concentrated in particular stores or periods.

A4. Damages sensitivity.

Assumption	Unpaid hours	Damages
Base case (all pre-login time)	170,000	$3.96M
Exclude shifts under 5 minutes	133,000	$3.10M
Exclude shifts under 10 minutes	69,000	$1.61M
Overtime share 10% instead of 20%	170,000	$3.79M
Two-year limitations period	113,000	$2.64M

A5. Method. Shift-level match of timekeeping and POS records on employee ID and date; pre-clock-in time = clock-in minus first POS login where positive. Stratified estimation of the affected share and mean gap. Aggregate hours = employee-years × shifts per week (payroll) × 52 × mean gap over all shifts. Blended rate from payroll distribution of regular and overtime hours. Intervals from stratified sampling variance propagated by the delta method.

Appendix B — Data

Timekeeping export (clock-in/out by employee and date, 2023 – 2025); POS authentication log (first and last login per employee per day); payroll (hourly rates, regular and overtime hours by pay period). Class: 3,200 hourly sales associates, 46 stores, ≈ 5,100 employee-years, 1.06M matched shifts. Shifts without a POS login (non-selling roles, 6% of shifts) excluded from the frame. System clocks compared on 200 shifts; maximum offset 30 seconds. Data produced by the employer; not independently verified.