Mother’s
Day
Yield

The answer is split. Given a binding constraint on Mother’s Day — tables, seats, or kitchen — different party sizes paid different rates for the constraint they consumed. An eight-top yields almost three times more per table-hour than a deuce; a deuce yields nearly twice as much per seat-hour as an eight-top. Which constraint binds the floor decides which party wins the day.

Mother’s Day did not change the yield curves. Revenue-per-seat-hour and revenue-per-table-hour by party size held within $1–$4 of a typical Sunday. What moved was the mix: deuces fell from 48.5% to 33.5% of checks, parties of five-plus nearly doubled. Eighteen percent more turns through the same floor; forty-five percent more guests filling the same number of chairs.

The portfolio split into two cohorts that mattered. Two stores ran a turbo-charged regular Sunday — large-party share under fourteen percent, thirty-percent-plus turn lift, +7.7% YoY each. Four stores ran a different restaurant — twenty-five to thirty-five percent large-party share, forty-two to fifty-two percent of dine-in revenue from groups of five-plus. Two of the four held; two slipped (−4.8% and −20.6% YoY). Having the family-day demand did not, on its own, convert.

Three threads carry the brief: chairs and the kitchen bound the floor, not tables; four stores ran a meaningfully different day from the average; and the deuce-anchored winners earned their day with a demand mix that fit their floor, not with a different playbook. The operational lever is pre-staging combinations the night before, not refusing the demand.

Two numbers govern the rest of the brief: $142.43 and $27.50. The first is what an eight-top earns the restaurant per hour for the single physical table it sits at. The second is what a deuce earns per hour per chair its guests fill. Both are real revenue, measured on the same day, against the same kitchen and the same labor. The reason the two numbers seem to argue with each other is that they describe different constraints.

If the binding constraint is tables — the floor has fewer surfaces than parties willing to wait — then revenue-per-table-hour is the right yield. The restaurant chooses which party to seat at each table. An eight-top at $142.43 per hour does more for the day than two deuces at $55 each (an aggregate $110 across two tables). A six-top at $108 beats a four-top at $81 on the same logic. Big parties dominate.

If the binding constraint is seats — every chair is full and the kitchen or guest experience caps the throughput — then revenue-per-seat-hour is the right yield. The deuce extracts $27.50 per chair per hour against the eight-top’s $14.76. Two deuces filling four chairs earn $55 per hour against an eight-top filling eight chairs at $118. Per chair, the eight-top is worth roughly half. Small parties dominate.

The operator's floor on Mother’s Day almost certainly bound on seats and on kitchen at the 11 AM–1 PM peak, even if it didn’t bind on tables: point-in-time table count ran at typical-Sunday P90, but every chair within those tables was usually full and the kitchen line was operating well past its peak throughput. The labor data confirms this same pinch — same-store peak FOH down 11%, peak BOH down 8%, against demand off 5%. The line bound; the seats bound; the tables had room but the kitchen and the chairs didn’t. The worst case in this matrix is therefore the only configuration the operator can avoid: a deuce-at-a-four-top. Two seats consumed at deuce check size, but a full four-top occupied — the worst $/table-hour rate and the worst $/seat-hour rate at the same time. A four-top with four people seated at it earns $81 per table-hour and $20 per seat-hour. A four-top with two people earns $55 per table-hour and $27 per seat-hour. Either way the four-top is committed; the question is only how many seats you sold to fill it.

Two facts dominate this table. The first is the collapse of the deuce. Fifteen points of share moved away from parties of two on Mother’s Day — the largest single mix movement in the data. The second is that the yield rates per bucket held within a few percent. Check size per party moved up modestly (deuce +$3, quad +$3, eight-plus +$6 versus baseline), and turn times ran systematically longer by three to five minutes per bucket — consistent with a busier room slowing service slightly. But the two effects roughly offset: revenue-per-table-hour and revenue-per-seat-hour by bucket sit within $1 to $4 of a typical Sunday. The operator did not learn a new restaurant on Mother’s Day; it ran the restaurant it has, slightly more slowly, with the customers who showed up.

The mix shift translated to a 45% lift in chairs filled, against an 18% lift in turns. That ratio — chairs growing 2.5x faster than turns — is the operational truth of the holiday. The floor turned slightly harder than usual, but each turn carried meaningfully more guests. The two together produced the day’s revenue lift; the labor cut against same-store made the result fragile on the experience side. Deuces did not vanish — 743 checks still ran across the portfolio at thirty-four percent of dine-in — but the pressure of the day didn’t fall on deuce supply. It fell on the chairs inside the larger parties that took their place.

The shape of the day at the portfolio level differs from the planned 9–11 AM peak labor window in a worth-knowing way: the dine-in turns peak is two hours later, and the three peak-equivalent hours sit entirely outside the planned labor window. Top-of-hour concurrency holds within 2.5% of its 11 AM high from 11 AM through 1 PM — three consecutive peak-equivalent hours, not a single spike. The same-store peak-window FOH cut of eleven percent is therefore mis-timed as much as it is under-stated: the labor cut hit a window that turned out to lead the actual peak by a full two hours.

Second, average party rose across the morning and peaked late in service. Early diners (8 AM) ran 2.90 average party — close to the typical-Sunday corporate average of 2.75. The pre-church wave at 11 AM lifted that to 3.71. Average party then dipped to 3.51 at noon — the lull between the early-church and post-church waves — before climbing again to 3.92 at 1 PM, the day’s peak. At the top of the 1 PM hour, large-party share reached 30% (95 of 316 active turns) and average party was well over a full guest larger than baseline. The day did not just get busier; it got progressively more family-shaped by the back half of service. The implication for floor planning is direct: stores that finish the day on a four-top-and-six-top crowd should staff and stage for that load, not for the morning’s deuce flow.

The third reading reframes how Mother’s Day pressed the floor. At roughly twenty active turns per store at peak, against a portfolio-wide typical-Sunday P90 of about twenty-two active tables, the portfolio was not running more tables at once than a busy regular Sunday. The lift didn’t show up as more concurrent tables; it showed up as +45% chairs filled at those same tables, plus +18% more total turns spread across a wider three-hour peak window. The kitchen was working 45% harder against a same-as-typical table count — that is the operational shape that broke the Ovation curve, not a tables-overflow problem. Reservation discipline is the highest-leverage operational fix because reservation pacing redistributes the chairs-filled load across the peak window; it isn’t fixing a table-overflow problem (there wasn’t one).

The cross-reference with the sales-YoY bands does not produce a clean correlation, which is the point. Among the four family-magnet stores, one was an A Winner, one a B Holder, one a C Slipper, and one a D Big Slip. Among the deuce-anchored stores, both were A Winners. Having a higher large-party share did not protect a store from slippage. Having a lower large-party share did not, on its own, predict a winner — but every store that ran low large-party share and high throughput won the day.

One more observation: the two top-yield stores both grew turns more than thirty percent against a typical Sunday while keeping average party within thirty cents of baseline. The arithmetic of that operation — same kind of customers, many more of them — produces the highest revenue-per-seat-hour day available in the data. They did not earn Mother’s Day by becoming family destinations. They earned it by being good at being themselves, with the demand turned up.

What this means going into Section V and the implications: the demand mix a store receives is necessary but not sufficient. The family-magnet stores still had to convert it, and three of the four didn’t. The deuce-anchored stores were given a mix that fit their floor, and they pushed it hard. The next section uses three case-study stores to show what conversion looks like under each demand shape.

The top-yield store ran Mother’s Day’s cleanest operating profile, and the floor-plan data shows why. Demand on the day (12.3% large-party share, twenty 5+ parties) fit cleanly within the store’s four verified standalone big-tops — tables 53, 71, 51, and 21 each capable of hosting 7-10 people without combining adjacent tables (verifiable from timing-overlap data: their neighbors stayed in independent service throughout). With ~5 turns per standalone big-top across the six-hour peak, that absorbed roughly twenty large-party seatings — almost exactly the day’s twenty 5+ parties. No combination was needed; the demand fit, just. The store kept its average party within a half-guest of typical, turned its floor thirty-eight percent harder than a regular Sunday — the strongest throughput lift across the sixteen stores — and posted +7.7% YoY.

Easter 2026 shows what the top-yield store looks like under heavier large-party demand, and it’s instructive. That Sunday, the top-yield store’s large-party share ran 17.2% (higher than Mother’s Day), and the floor hosted a 24-person party — recorded as parallel 12-checks on combined tables 50 and 51 — plus a 12-top via combined 31+32, a 10-top via combined 34+35, and an 8-top via combined 61+62. The same dormant-neighbor pattern that flags on-the-fly combination elsewhere in the operator shows up at the top-yield store under stress. The top-yield store combines tables when demand exceeds standalone capacity, like every other store in the portfolio. Mother’s Day at the top-yield store ran clean because the demand mix didn’t require combination, not because the top-yield store’s floor refuses to combine. The corollary matters for the family-magnet stores: at 35% large-party share, no portfolio floor’s standalone capacity is sufficient on its own. Combination is mathematically required; the only question is whether it’s pre-staged or improvised.

The structural-exception store ran the day’s hardest combinatorial math. Twenty-eight dining tables — the smallest dining floor of the three case-study stores — absorbing 35.1% large-party share. Only one table on the floor (table 17, the corner round) is drawn as a real big-top; every other 7+ party on Mother’s Day was hosted by combining four-tops. The worst-case version of this played out at table 24: a single 10-top occupied the table for 94 minutes and produced exactly one turn for the entire day. While ten people ate from a single firing wave and the FOH concentrated its attention on the one table, the rest of the floor absorbed the portfolio-wide thirty-five percent large-party tide on a footprint forty percent smaller than the overflow store’s. The store grew turns 31% and posted +3.6% YoY — it got the revenue.

From a guest-experience perspective, it didn’t. The structural-exception store’s Ovation came in at 3.36 with a 50% issue rate — the worst guest signal on the day across the sixteen stores. The yield framework explains why. Six to seven concurrent large-party turns ran at top-of-hour peak on a 28-table floor — a quarter of the dining room tied up in 60-to-90-minute combined firings while the kitchen line absorbed the chair load of those parties. The same large-party density on the overflow store’s 32-table floor or the top-yield store’s 31 would leave meaningfully more standalone four-top inventory in independent rotation.

The operational reading is that the structural-exception store is structurally exposed in a way the overflow store isn’t. The overflow store can absorb its 35% large-party share by pre-staging the combinations it already improvises (overflow goes from chaotic to choreographed; the math works). The structural-exception store’s gap is that even with maximum combination, the demand exceeds the combinatorial capacity its physical footprint can absorb cleanly. Reservation pacing helps (capping eight-tops at two per peak hour would mathematically smooth the load) but the deeper question, eventually, is whether the family demand the structural-exception store receives is structurally too thick for the building.

The overflow store was the day’s extreme on family-party density: 52% of dine-in revenue from five-plus groups, more than any other store in the portfolio. It had the demand. It did not convert. Where the structural-exception store’s identical large-party share produced +3.6% YoY (with a 50% issue rate), the overflow store produced −4.8%. The throughput tells the rest: the overflow store grew turns only fourteen percent against typical, against the top-yield store’s thirty-eight percent and the structural-exception store’s thirty-one. Same physical footprint as the top-yield store, similar demand to the structural-exception store, fewer turns served than either.

The arithmetic of the gap points to demand exceeding standalone capacity. It has roughly five verified standalone big-tops on the floor — tables 44, 20, 12, 40, and 31, each capable of hosting 7-11 people without combining (verifiable from timing-overlap data with neighbors active). At ~5 turns per table over the peak, that standalone inventory can absorb roughly 25 large parties. The overflow store received 58 large parties on Mother’s Day. The remaining 30+ had to land somewhere — and they landed in combination zones improvised on the fly: 62+61+63 (the heaviest, all-day banquette behavior), 31+32, 43+42, plus several smaller ad-hoc pairings. Tables 32, 42, 61 ran zero independent turns because their chairs were absorbed into neighboring lead-table combinations. The cost is not empty chairs (the chairs were full); it is lost floor flexibility — a 94-minute combined eight-top replaces what could have been two independent four-top turns in the same physical space.

The same pattern surfaces at the deuce-anchored stores when their demand presses. At Easter 2026, with the top-yield store’s large-party share running 17.2%, the top-yield store combined four-tops on the fly across multiple zones — the same dormant-neighbor signature the overflow store ran on Mother’s Day. The Mother’s Day difference is that the top-yield store’s 12.3% share fit inside standalone capacity while the overflow store’s 35% share required combination as the overflow valve. The corrective at the overflow store isn’t to refuse demand; it’s to pre-stage the combinations in the morning pre-shift rather than improvise them in service. The data already tells us which zones the overflow store instinctively uses: 62+61+63, 31+32, 43+42. Designate those three pre-pushed combinations the night before family-day service, assign them to specific servers, and reservation-pace the queue to the resulting 8-position grid (5 standalone big-tops + 3 pre-staged combinations). At ~2 turns per position over the 2-hour peak that yields ~16 designated peak slots; the remainder of the day’s 58 large parties spreads across the pre-peak and post-peak windows via reservation pacing, with walk-in 5+ deferred to the next available 2-turn window rather than absorbed by improvised combination.

Read at the operator. Even on the year’s biggest day, deuces remained the single largest party-size bucket across the sixteen stores at thirty-four percent of dine-in checks, and parties of four or fewer were seventy-nine percent of turns. Steady-state deuce demand at peak (roughly seven per store) sat well under the typical-Sunday P90 of twelve to thirteen — Mother’s Day actually relieved deuce concurrency. The portfolio ran point-in-time table concurrency at typical-Sunday P90 (roughly twenty active tables per store) and pushed +45% more guests through that same table count — floors built to typical-Sunday demand had usable table headroom but no chair headroom. The case-study floor plans confirm what the aggregate suggested: portfolio floors are mostly identical four-top grids portfolio-wide, with one or two structurally larger tops per store and no dedicated deuce inventory. Floor design is not the operator’s differentiator on the holiday. Nothing in the day suggests the operator’s small-party-anchored model is structurally wrong — the holiday is a mix excursion, not an inversion.

Read at the store. Four stores in the portfolio ran a meaningfully different day from the portfolio average — the family-magnet four. Each ran 25%+ large-party share with 40%+ of dine-in revenue from five-plus groups. The case-study evidence (the overflow store vs the top-yield store on near-identical floors) shows the gap at most of these four is demand exceeding standalone capacity rather than a difference in operational competence — every store combines tables when forced. The structural-exception store carries the structural exception: its 28-table dining room is the smallest in the case-study set, and the combinatorial math becomes adversarial even with maximum combination — the building can’t cleanly absorb the family demand it receives. The remaining twelve stores look closer to the portfolio average.

Read at the operational layer. Two levers, in priority order. Pre-staged combination zones first. Every store in the portfolio has similar physical capacity (4-5 verifiable standalone big-tops). Every store combines tables when demand exceeds that capacity — the top-yield store proves it on Easter (a 24-top hosted via combined tables 50+51, plus 12-top, 10-top, and 8-top combinations); the overflow store on Mother’s Day (six combination zones); the structural-exception store under the worst combinatorial pressure of any case-study store. The actual host-stand lever isn’t whether combinations happen but whether they are pre-staged the night before (combinations physically pre-pushed at the morning pre-shift, server stations assigned, reservation grid built around them) or improvised in the hottest moment of service. Pre-staging moves the decision-making from peak service to the calmest moment of the day. Reservation pacing second. Reservation discipline is Mother’s Day’s highest-leverage fix; this brief sharpens how it should be implemented. An eight-top is the single longest turn in the data (sixty-eight average minutes, ninety-four at the ninetieth percentile). The reservation pacing that matters is “reserve the eight-top for two seatings at known designated positions” — which only works if the positions are pre-designated (lever one). Reservation pacing without pre-staging leaves the combination scramble in the hottest part of service.

A concrete pre-shift playbook, using the overflow store as the worked example. Two weeks out: pull the prior year’s table-level activity for the holiday — the data reveals each store’s instinctive combination zones (the overflow store Mother’s Day 2026: 62+61+63, 31+32, 43+42). Confirm Toast Waitlist & Reservations is configured for two-turn windows on those positions plus the standalone big-tops. Night before or morning pre-shift: physically pre-push the designated combinations so they stay set all day. Assign one server per combination zone plus servers for the five standalone big-tops (44, 20, 12, 40, 31). Load the reservation grid for two seatings per position at peak — an 8-position grid yields ~16 designated peak slots across the 11 AM–1 PM window, with pre-peak (8–10 AM) and post-peak (2 PM) absorbing the remainder of the day’s ~58 large-party arrivals. Brief the kitchen line on the predicted large-party count by hour, with the heaviest load at 12–1 PM. During service: enforce exit pacing on combined zones (dessert menu pre-dropped at 45 min, check at 60 min, table cleared by 90 min — matching the p90 turn time observed in the data). Walk-in 5+ parties during peak (11 AM–1 PM) go to the waitlist with a deferred reservation for the next 2-turn window; no improvised combinations once service begins. The grid is built from the start, not assembled in the moment.

Wait-list spillover. Toast records only seated parties. Groups that left without sitting because no table fit are invisible to the data. On Mother’s Day — with +45% more guests pushed through a same-as-typical-Sunday peak table count — the share of walked-off parties is almost certainly higher than baseline. The 322-active-turn peak is a floor, not a ceiling, on demand.

Indoor vs patio split. Table assignments don’t carry inside/outside metadata in toast_order_turns. Stores with patios (the second top-yield store among them) may have run a meaningfully different mix outdoors than indoors; the aggregate yield numbers smooth across that. A patio-tagged future cut would sharpen the case study on the two top-yield stores.

The deuce-at-a-four-top case. Table-size metadata is not in the schema. We argue the deuce-at-a-four-top problem from theory and from the yield arithmetic, not from observed table assignments. A future enrichment of toast_order_turns with table-capacity from the Toast configuration API would make this case observable rather than inferred.

Off-premises. This brief is dine-in only. Mother’s Day off-premises (takeout + delivery) contributed roughly thirty-four percent of the corporate day’s revenue, with its own party-size and yield questions. The Toast-side takeout data is in; the DoorDash side does not finalize until May 13 and beyond.

Kitchen capacity. We argue from line-pinch evidence (the labor cut, Ovation issue rates concentrated at 10 AM and 12 PM peaks) that the kitchen was a binding constraint. We have no direct measure of expo time or ticket-to-fire latency on the day. A future toast_kitchen_timing data source would let us test the “seats bound first” thesis more directly.

Reservation-channel attribution. Toast does not currently distinguish dine-in turns that came via Waitlist & Reservations from walk-ins. The reservation-pacing recommendation is informed by the turn-time distribution observed here; tightening it further requires reservation-channel data.