DoorDash / Wolt Market Israel Divestiture — AI-Assisted Data Analysis
Each data source measures a different population at a different point in time. The gap is not surprising once you understand what each one actually counts.
These three sources are not reconcilable to a single number — they measure fundamentally different things. ADP measures cash paid to all 1,437 people who cycled through the year. The census measures 631 people alive on one specific day in March 2026. Trying to force a match is the wrong frame. The right question is: which methodology produces the most defensible forward-looking standalone cost?
Twelve monthly Worker Census extracts (Jan–Dec 2025) from FY25 Snapshots for PWC were analyzed to test whether ADP’s 1,437 distinct FY25 IDs imply volatile staffing. Annualized base compensation uses: Monthly — Total Base Pay × 12; Hourly — FTE × 34 × 52 × hourly rate (Total Base Pay Amount). This is base pay only (not employer load, premiums, or benefits) — intended to show staffing shape, not to match ADP ₪77M cash cost.
Headcount 569–654 (coefficient of variation 4.36%) · FTE 442–505 (CV 4.13%) · Annualized base comp ₪39.78–46.35M (CV 6.04%). By job family and (for Fulfillment) compensation grade, run rates stay in a tight band month to month — consistent with churn replacing roles rather than a wildly shifting operating headcount.
| Job family / segment | 01/2025 | 02/2025 | 03/2025 | 04/2025 | 05/2025 | 06/2025 | 07/2025 | 08/2025 | 09/2025 | 10/2025 | 11/2025 | 12/2025 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Fulfillment · D0 | 431 | 418 | 413 | 414 | 391 | 429 | 466 | 446 | 435 | 445 | 445 | 456 |
| Fulfillment · D1 | 114 | 113 | 115 | 118 | 117 | 120 | 124 | 123 | 124 | 128 | 128 | 123 |
| Fulfillment · D5 | 29 | 30 | 29 | 28 | 27 | 26 | 29 | 29 | 27 | 27 | 30 | 30 |
| Audience Marketing | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 0 |
| EA/Administrative | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| FP&A | 0 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Facilities Team | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 4 | 4 |
| Growth Marketing | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 4 |
| Purchasing | 5 | 5 | 6 | 7 | 7 | 7 | 7 | 7 | 7 | 7 | 7 | 7 |
| Strategy & Operations | 9 | 9 | 9 | 10 | 10 | 10 | 11 | 11 | 12 | 12 | 12 | 12 |
| Supply Chain Management | 7 | 7 | 7 | 7 | 7 | 7 | 7 | 8 | 8 | 8 | 8 | 8 |
| Month | Headcount | FTE | Ann. base ₪M |
|---|---|---|---|
| 01/2025 | 604 | 467.19 | 39.909 |
| 02/2025 | 592 | 459.95 | 39.869 |
| 03/2025 | 589 | 455.76 | 39.777 |
| 04/2025 | 594 | 459.83 | 42.787 |
| 05/2025 | 569 | 442.47 | 41.520 |
| 06/2025 | 609 | 468.14 | 43.152 |
| 07/2025 | 654 | 505.32 | 46.259 |
| 08/2025 | 634 | 489.3 | 45.355 |
| 09/2025 | 623 | 479.13 | 44.661 |
| 10/2025 | 637 | 491.01 | 45.506 |
| 11/2025 | 640 | 493.84 | 46.083 |
| 12/2025 | 646 | 497.74 | 46.354 |
ADP counts every distinct employee ID with FY25 payroll activity — including short-tenure and churned roles. Monthly census snapshots count who is on roster each month at ~600–650 people with stable job-family / grade mix. The two measures answer different questions; snapshot stability does not contradict ADP churn — it shows the operating run rate did not swing proportionally to the 1,437 cumulative ID count.
These three hypotheses are not additive — they are separate analytical lenses, each measuring the gap from a different angle against a different baseline. They are not findings. Some amounts are directly verifiable from the data; others are back-calculated estimates that depend on assumptions or record-matching between ADP and census. Each card below notes its data basis. This analysis is intended to identify the questions that need to be resolved, not to serve as a definitive reconciliation.
The ADP file has 1,437 employees. The March 19, 2026 census has 631. The difference — 924 people — worked during FY25 but were gone before that snapshot. Their costs appear in ADP and TB but are invisible to a single-point-in-time census model. High churn among the 1,437 IDs can suggest the “missing” mass is turnover — but that framing implies the operating workforce was unstable month to month.
Twelve monthly Worker Census files (Jan–Dec 2025) show stable run-rate headcount and FTE (approx. 569–654 employees, CV 4.36%; FTE CV 4.13%) and steady mix by job family — for Fulfillment, also by compensation grade (D0/D1/D5). Annualized base pay roll-up (per the snapshot methodology in the FY25 Snapshots section) moves in a ~₪39.8–46.4M band (CV 6.04%), not a sawtooth pattern. So: churn is real, but it largely replenishes seats within a relatively constant operating footprint — the ADP-vs-single-census gap is still about cohort definition (cumulative IDs vs. point-in-time roster), not proof that steady-state staffing was whipsawing throughout FY25.
Dark stores operate 7 days/week. Saturday pay at 150–200% premium rates is a legal requirement in Israel and a structural operating cost, not an exception. Overtime and holiday pay compound this. A census annualization of base rate × hours misses all of it.
Verified FY25 premium pay in ADP (Saturday, OT, extra paid hours, holiday — the components above) totals ₪17.2M. Total gross pay in the FY25 ADP payroll register is approximately ₪66M. Expressing the premium bundle against that denominator: ~26.1% of ADP gross (₪17.2M ÷ ₪66M). That anchors the “missing” premium layer to the same ADP gross universe as the roll-forward, rather than to census-modeled base.
123 employees hired January–March 2026 appear in the census and their full annualized costs (~₪8.24M) are included in the model's BG column. But they contributed zero to FY25 ADP. This partially offsets the gap when comparing to FY25 actuals — but inflates the model for forward-looking purposes relative to what was actually paid.
These observations don't produce a clean waterfall that sums to the gap — they overlap and measure different things. What they collectively demonstrate is that the gap is structural and expected, not a data error. A census-based model will always understate actual payroll for this type of business given high-turnover hourly workforce, 7-day operations, and Israeli premium-pay requirements. The next step is resolving which of these factors should be addressed in the standalone assumption — and how.
The model contains two columns that compute hourly worker annual cost differently. Each reflects a distinct methodological choice. Understanding what each includes — and what neither captures — is the basis for an informed standalone cost decision.
Deliberately weights cost by each employee's contracted FTE% — the correct approach for reflecting that not all 631 census employees are full-time. Accurately captures base pay for part-time workers on a contracted-hours basis. What it cannot capture: the premium pay multiplier (Saturday 150–200%, overtime) that applies to those scheduled shifts once the store is open 7 days.
Applies 2,080 hrs/year to all hourly workers, ignoring FTE%. Happens to land close to ADP actuals numerically — but for a different reason: it over-annualizes part-time workers' base hours, which roughly offsets the premium pay and transient employee costs that ADP captures but neither census column includes. The near-match is coincidental, not a methodological improvement.
The business ran with 1,437 people in FY25. The census sees only 631 of them. Understanding who the 924 "missing" employees are — and why they matter — is the foundation of a defensible standalone headcount assumption.
719 employees were active at Dec 31, 2025 — a better standalone anchor than 631 (March 2026 census). Benchmarks support 700–750: 25–30 dark stores × 15–25 employees = 375–750 range.
The data supports a defensible path to ₪77–80M as the forward-looking standalone headcount cost. Here is how to structure the argument.
The FTE%-adjusted Adj FLC is a deliberate and valid choice for base pay. The remaining gap to ADP actuals is primarily the ₪17.2M in Saturday premiums, overtime, and holiday pay — structural costs of running dark stores 7 days/week under Israeli labor law. The standalone model needs an explicit view on whether these costs belong in the forward-looking baseline.
The census 631 is a March 2026 point-in-time snapshot. ADP shows 719 active employees at Dec 31, 2025. Benchmarks (25–30 stores × 15–25 staff) support 450–750. The running average workforce is materially higher than the census implies — and the business is growing.
The ADP file currently has no hire or termination dates — only binary Starter/Leaver flags. Start and end dates would allow calculation of exact person-months worked, a true running-average FTE count for FY25, and a precise census-to-ADP bridge.
The census captures 631 employees as of a single day in March 2026, but 1,437 employees were on payroll at some point in FY25. The ₪77M reflects actual cash paid to all of them, including premium rates for Saturday and overtime operations that are structural to this business model. The correct forward-looking cost for a business of this size is ~₪78M — not the ₪50M implied by annualizing only the current census headcount at base rates."
The conclusions above are directional, not definitive. Several key numbers rest on assumptions or back-calculations that better data would either confirm or revise. These limitations should be disclosed when presenting to DoorDash or the client.
The claim that 207 employees were active at Dec 31, 2025 but absent from the March 2026 census is derived as: (719 ADP year-end actives) minus (census employees who were also in FY25 ≈ 631 − 123 Q1 hires = 508), yielding ~211. The ADP file has no departure dates — we cannot confirm when those ~207 people left or what they individually cost. The ₪9.8M attributed to this group is estimated proportionally, not computed from individual records.
Update (this revised analysis): Twelve monthly Worker Census snapshots (Jan–Dec 2025) now provide month-by-month active headcount and FTE by job family and Fulfillment grade — see the FY25 Snapshots section. What remains incomplete is matching those roster rows to ADP employee IDs without hire/termination dates in ADP. The 719 year-end ADP actives and March 2026 census still do not automatically reconcile to each monthly file at the person-ID level.
Saturday premiums, overtime, and holiday pay totalling ₪17.2M are verified FY25 actuals from the ADP file. However, the amount in a standalone scenario depends on store count, staffing patterns, and any contractual changes post-separation. If the business opens additional stores, premium pay will increase. If operating hours or scheduling practices change, it could decrease. The ₪17.2M is the right reference point, but it should be presented as a historical baseline, not a confirmed forward-looking cost.
The directional conclusion — that the right standalone cost is ~₪77–80M, not ~₪59M — is well-supported by the data and unlikely to change materially with better information. The specific mechanisms explaining the gap are partially derived and should not be presented as a definitive reconciliation. Until start/end date data is received from DoorDash, the analysis should be framed as: "the gap is structural and explainable; here is the hypothesis; here is what we need to confirm it."
Prioritized actions for the engagement team. Data requests to DoorDash are the critical path to converting this analytical hypothesis into fully validated proof.
The ₪17.2M in Saturday premiums, overtime, and holiday pay in FY25 ADP reflects the statutory cost of 7-day operations in Israel — not a one-time anomaly. If the standalone business runs the same operating model, some level of premium pay is a recurring cost. The current model baseline does not carry this explicitly.
These employees are in BD but not in FY25 ADP. Separately disclose: "Census includes 123 employees hired post-FY25 whose costs are not yet reflected in historical payroll." This prevents a false reconciliation attempt.
Replace the 631 census headcount anchor with 719 (Dec 31, 2025 active count) as the FY25 run-rate anchor, trending to 700–750 as the stable standalone assumption. Supported by benchmark data and ADP history.
The single highest-value data request. Would allow exact calculation of person-months worked per employee, a validated running-average FTE for FY25, and a precise numerical bridge from census to ADP — converting this analysis from directional to definitive.
Alternative to start/end dates. Monthly headcount as of Jan–Dec 2025 would allow direct computation of the running-average employee count and validate whether the business ran at ~700–800 employees throughout the year.
The ADP file shows ₪17.2M in premium pay but the Cost Center column is 99.6% null. Function-level premium pay allocation would allow a more precise standalone cost per functional tab, rather than estimating it as a percentage of base pay.
These employees appear in the current census and their annualized costs (~₪8.24M) are included in the model, but they contributed zero to FY25 payroll. Worth flagging explicitly when comparing the model's forward-looking baseline against FY25 actuals to avoid a false reconciliation attempt.