— Operator-to-operator
Answers, not opinions.
Every answer sourced to the live system. No hot takes. No vendor pitches.
For the solo operator
How to find money leaks in your restaurant: the 7-signal checklist
I can feel my restaurant leaking money but I can't name where. What do I check?
Seven signals, in the order of how fast they pay back. This is the exact set our agents read; you can check every one from data you already have. 1. Voids by name, not by store. Store void rate vs its peers, then broken out by employee and shift. Concentration by name is the tell. (Void Hunter) 2. The 3P gap. Blended-effective take rate per platform vs your contract rate — commission, ads, promos, refund chargebacks, DashPass premium all-in. This is routinely points of margin, not basis points. (3P Fee Finder) 3. Labor drift. Scheduled hours vs clocked hours, plus unbudgeted overtime and ghost shifts — the gap between the schedule you approved and the payroll you ran. (Labor Leak) 4. Tip variance. Week-over-week tip movement per name. A server whose tips fall 30% while the store's hold steady is telling you something about sections, shifts, or service — sometimes about the till. (Tip Variance) 5. Catering reconciliation. Invoice-vs-POS gaps on catering orders. Big tickets, loose process — catering leaks are small in count and large in dollars. (Catering Leak) 6. Vendor price drift. Same SKU, same vendor, price up 7% in a month while a competitor held. Nobody re-quotes mozzarella weekly; the drift compounds until someone looks. (Vendor Drift) 7. Pour variance. Beverage poured vs beverage sold. The bar is the oldest leak in the industry for a reason. (Beverage Score) Most kitchens are carrying 2–4 points of recoverable cost across these seven. Every one of these agents runs free on a CSV export from Toast, Square, Clover, or PDQ at never86.ai/trial — 60 minutes, no card, and it names your stores and your line items, not hypotheticals.
Read it
For the multi-unit operator
How to audit your DoorDash, Uber Eats, and GrubHub statements
My 3P fees feel higher than my contract rate. How do I actually check?
One rule before the steps: your contract rate is your floor, not your ceiling. Every audit starts by ignoring the contract and computing what each platform actually took. Step 1 — pull 90 days of statements per platform, plus the same window's orders from your POS. Step 2 — compute the blended-effective rate: total deductions (commission + fees + ads + promos + refund-class adjustments) divided by gross 3P sales. Do it per platform, per month. Step 3 — compare blended-effective to contract. The gap is your finding. Where the gap comes from: DashPass orders bill at a higher rate than your standard contract (a 10% contract commonly blends to 11%+ once DashPass mix hits normal volume). Promotions and ad credits get netted out of payouts where you stop seeing them as line items. Refunds and disputes get charged back against classes of orders you thought were settled. None of this is hidden, exactly — it's just spread across statement formats designed to be read one week at a time. Step 4 — reconcile statement deposits against POS-recorded 3P sales. The delta between what the POS says you sold and what actually landed in the bank is the single most clarifying number in the audit. Step 5 — take the blended-effective rates to a renegotiation: platforms match each other's rates for operators who show up with the math. We run this as the 3P Fee Finder and Rate Card Audit agents — both free on a CSV at never86.ai/trial. Related reads: never86.ai/answers/doordash-blended-rate-dashpass and never86.ai/answers/renegotiate-ue-gh-to-dd.
Read it
For the solo operator
What should my prime cost be — and why are you seeing it weeks late?
What's a healthy prime cost for a restaurant, and how do I actually track it?
Target: at or under 60% of sales for most full-service concepts. Food plus labor is prime cost, and the practical bands are 55–60% healthy, 60–65% watch it, 65%+ you're working for your vendors and your payroll company. QSR runs lower; chef-led, ingredient-forward concepts run the top of the band on food and earn it back on ticket. But the number matters less than the latency. Most independents see prime cost when the accountant closes the month — three to six weeks after the leak opened. Food cost drifts four points in a week (a vendor reprice, a portioning slip, a theft pattern) and the P&L tells you about it in five weeks, after it's cost you real money. A 4-point drift on $100K/month of sales is $4,000 a month, every month, until someone notices. The fix isn't heroic bookkeeping — it's reading what your POS and invoices already know, daily. Yesterday's sales are in your POS by 6am. Your vendor prices are on this week's invoices. Daily prime cost is just those two sources reconciled every morning, with a flag when the trend breaks. That's literally what we built Pulse for — daily food cost, the 30/60/90-day prime cost trend, cross-vendor price comparison, and one coach card per leak telling you the move, at $199/mo (never86.ai/pricing). But even if you never touch our product: get your prime cost read to daily. The gap between "monthly" and "daily" is where most of the recoverable money lives.
Read it
For the solo operator
What's a normal void rate for a restaurant?
What void rate should I be seeing on my POS, and when is it a problem?
The number people quote is 1–2% of gross sales. It's not wrong, but it's the wrong question — because a "normal" void rate is a store-level average, and leaks don't live at the store level. They live at the name level. The better test is the one we run in production: build a peer band. Take stores doing comparable volume with a comparable service model, find the median void rate, and flag anything running above roughly 1.5x that median. In the networks we reconcile, healthy full-service stores commonly sit anywhere from 0.3% to 0.7% — and a store at 1.2% against a 0.5% peer median is a finding even though it would pass the "under 2%" folk rule with room to spare. Then break the flagged store out by employee. A store's excess void rate is almost always concentrated: one or two names carrying void counts far above their shift-mates on the same stations. That's not automatically theft — in the 87 days of data that started this company, 38% of voids were process noise, 27% were comps rung as voids, and 19% were untracked employee meals. But concentration by name is where the conversation starts. So: don't benchmark against a magazine number. Benchmark each store against its own peers, flag the band-breakers, and attribute by name. That's the whole method, and it's what Void Hunter runs on a Toast, Square, Clover, or PDQ export in about 30 seconds at never86.ai/trial — free, no card.
Read it
For the multi-unit operator
What's the difference between Void Hunter and Leak Detector
You have two products that both look at voids. Which one do I need?
Different windows on the same operation. Both are free to try; you probably want both, but most operators start with one based on what export they can get out of their POS today. Void Hunter operates on aggregated employee-performance data — the row-per-employee-per-period export. The columns are Location, Employee, Net Sales, Void Amount. Toast calls this "Employee Performance Report"; Square calls it "Team Sales Report"; Clover has "Employee Reports." This export is small (one row per employee per period) and downloadable in 30 seconds from any POS that supports it. What Void Hunter tells you: per-store void rates ranked against each store's own peer median, top names by void $, the annualized excess-vs-peer-band recovery surface. It's the 30-second canary check — "is anyone obviously above the band?" Leak Detector operates on ticket-level data — one row per ticket — with columns for tender, void status, comp $, discount $, optional timestamps. Toast calls this "Sales Detail"; Square has the full Transaction History export; Clover has Reports → Transactions. This export is bigger and more granular. What Leak Detector tells you: six separate signals from the same data — void-after-payment (the cash skim), cash-only voiders, comp abuse, promo stacking, discount-after-close, and day-of-week patterns. Each name gets a composite risk score 0-100, and the same name showing up on three or four signals at once is your strongest theft signal. How to pick: if all you have is the summary export, run Void Hunter and you'll catch the gross outliers. If you can get the ticket-level export, run Leak Detector — it'll catch the same names plus surface schemes Void Hunter can't see (because aggregated voids don't show the tender or the timestamp). The trial at never86.ai/trial supports both with a toggle. Drop your CSV, pick the agent, see the read.
Read it
For the multi-unit operator
What's the real EBITDA hit of 1 percentage point of void rate at scale
Quick math — if I run a 10-unit group at $1.2M AUV per store, what does 1 percentage point of excess void rate actually cost me?
Quick math, then the explanation. Stack: 10 units × $1.2M AUV = $12M net sales. 1 percentage point of excess void rate = 0.01 × $12M = $120,000 in voided revenue per year. Most operators stop there. Wrong number. Voided revenue at zero-COGS recovery would be $120K of pure margin, but voids carry the food cost AND the labor cost already spent on that ticket. So the actual P&L hit is: - Recovered revenue if you fix the pattern: ~$120K - Less the COGS you spent that you can't get back on the voided tickets: ~$38K (assuming 32% food cost) - Less the labor you spent already: ~$36K (30% labor cost) - Net P&L recovery: ~$46K of pure EBITDA per percentage point of void rate at this scale Now annualize against your equity. At a 5x EBITDA multiple, $46K of annual EBITDA is worth ~$230K in enterprise value. Per percentage point. Per year. For a 10-unit chain. A 16-unit chef-led group running 3pp above the peer band on void rate is sitting on roughly $700K of enterprise value just from this one signal. That's why the Void Hunter exists — and why the next move after the report is a 5-minute conversation with one name, not a CFO project. Source-tag for this answer: Estimated. Food cost and labor cost are reasonable industry midpoints; your actual numbers may differ. The framework holds across the range. For your specific stack, drop a Toast employee-performance CSV at never86.ai/trial and we'll compute the exact lever in 30 seconds.
Read it
For the multi-unit operator
What a healthy comp rate looks like for a chef-led restaurant
What's a normal comp rate for a chef-led concept? My EBITDA is getting eaten and I don't know if it's the floor or the food.
For a chef-led, ingredient-forward concept, comp rate as a percentage of net sales typically lands between 1.5% and 3.5%. Below 1.5% you're likely under-comping (denying retention opportunities); above 4% you have a leak. The bucket matters more than the headline number. Healthy comps are concentrated in three categories: 1. Service recovery (a guest had a real problem) — typically 60-70% of total comps for a chef-led group 2. Manager hospitality (table got recognized, regular got an amuse) — 20-25% 3. Pre-shift / training tastes (line cooks need to know the menu) — 5-10% Unhealthy comps concentrate elsewhere: - One server owns >20% of the comp dollars across all tables → coverage / favoritism review - 30%+ of comps are "manager comp" with no service-issue note → process gap (or theft proxy) - Beverage comps as a percentage of total comps run higher than beverage's share of net sales → liquor program drift A specific lever for chef-led groups: split your comp report by category (food / beer / liquor / wine) and compare each category's comp rate to its share of net sales. If wine is 12% of net but 28% of comps, you have a wine training problem, a wine theft problem, or a wine quality problem — pick the right conversation. The Never 86'd Leak Detector surfaces the comp-abuse signal using two rules: above 1.5× the network's peer median, OR above 10% of an individual employee's own revenue with $200+ in absolute dollars. The second rule catches the case where only one person is comping — the peer median is zero, so the multiplier rule misses, but the absolute floor still flags.
Read it
For the multi-unit operator
Day-of-week patterns in employee voids — what they tell you
My void rates look fine in aggregate but I think one server has a pattern. How do I check?
Aggregate void rate is the wrong window. A 1.8% network void rate is healthy; a single server with 1.8% might still be running a scheme — if all of his voids happen on closing shifts when the manager's gone. The diagnostic is concentration. For each employee with five or more voids in your window, build a histogram of their void timestamps by day-of-week. Now look at the max bucket. If 40% or more of that server's voids cluster on a single weekday, you have a pattern worth a review. Why 40%? A perfectly distributed schedule would put 14% on each weekday (1/7). A schedule weighted toward weekends would put 25-30% on Fri/Sat each. Anything above 40% on a single day signals deliberate timing, not natural distribution. What to do next: - Cross-reference the day-of-week with the closing manager that night. If the same name appears as closing manager on every void day, you have a coverage gap. - Look at the tender split. Day-of-week + cash-only is the strongest combined signal — that's the classic post-payment skim. - Pull the actual tickets for those voids and read them. Most schemes leave a paper trail. The Never 86'd Leak Detector runs this check automatically against ticket-level CSV exports. The card shows up in the result whenever an employee crosses the 40%-concentration / 5-voids-minimum threshold, with the day name and the ratio (e.g. "Tue · 6/6 · 100% concentration"). Pattern detector, not verdict detector. Same rule as Void Hunter.
Read it
For the multi-unit operator
How to spot a void-after-payment skim in your POS data
How do I tell if an employee is voiding tickets after they were already paid (the classic cash skim)?
The classic cash skim works in three steps: customer pays in cash, server pockets the cash, then voids the ticket so it never appears on the close-out report. The pattern shows up in your POS data in a specific way — once you know what to look for, it's defensible to the penny. What to look for: 1. Tickets where the void timestamp is AFTER the payment timestamp. In Toast Sales Detail this shows as a non-null `void_amount` on a ticket that also has a `tender_amount > 0`. In Square transaction history, look for refunded transactions on a previously-settled payment. 2. Concentration on a single tender. A "regular" void (item put in by mistake, voided before payment) skews credit because credit is more common. A skim concentrates on CASH because that's the whole point — you can pocket cash, you can't pocket a Visa. 3. Concentration on a single employee. The skim relies on the same person being able to pay AND void the ticket. So one name owns 60-100% of the post-payment voids in a window. 4. Concentration on a single day-of-week or shift. Operators who run the scam tend to do it on their closing shifts when fewer eyes are on the drawer. Tuesday-Thursday closes are the classic pattern. The Never 86'd Leak Detector runs all four of these tests in one pass against a ticket-level Toast / Square / Clover CSV and surfaces every name that matches more than one signal, sorted by composite risk score 0-100. Names that hit all four typically score 80+. A working example from the design partner's data: one name showed up with 6 post-payment voids in a six-week window, all of them on cash tickets, all on Tuesday closes. Risk score 83. The conversation with that employee was short. We never call it "theft" — we call it a pattern. The pattern triggers a 5-minute review with the operator. The pattern is the receipt; the verdict is theirs.
Read it
For the multi-unit operator
How do I know if my 3P contract rate is competitive
How do I know if my DoorDash / Uber Eats / Grubhub contract rate is competitive for my unit count?
The contracted rate is almost never the actual blended-effective rate. The peer band at multi-unit scale (5-50 units) is roughly 10% at the floor, 18-20% typical, 25-30% at the ceiling. Where you sit depends on chain volume, market concentration, and which channel mix you're running. The Rate Card Audit at never86.ai/demo/rate-card-audit takes your contracted percentages and tells you where they land on the band — without needing any of your store data. The real lever isn't the contract anyway; it's the gap between contract and effective rate once promotions, premium-tier orders (DashPass / Uber One), and platform marketing fees are stripped in. That gap is typically 1.2 to 2.8 percentage points per partner. Multiply by your annual 3P revenue. That's the renegotiation conversation.
Read it
For the multi-unit operator
How to renegotiate Uber Eats and GrubHub to the DoorDash rate
My DoorDash contract is 10% — can I take Uber Eats and GrubHub there too?
For the chef-led 16-unit group, yes — and the math says it is the cleanest dollar move on the table. Uber Eats and GrubHub both contract at 18% delivery; DoorDash already contracts at 10%. UE costs ~$89.8K / 4 weeks; GH costs ~$11.4K / 4 weeks; the gap between 18% and 10% on combined volume is ~$45K / 4 weeks ≈ $585K / year. The precedent is already inside the house — DoorDash already gave it to you at chain volume. One conversation per partner. Walk in with the DD contract, your trailing four-week 3P revenue numbers, and the question: why is your rate two-thirds higher than DD on the same chain?
Read it
For the multi-unit operator
Why your DoorDash 10% blends to 11.2% — the DashPass gotcha
My DoorDash contract says 10% — why does the platform show 11.2%?
A 10% delivery contract is the rate DoorDash charges on standard orders. DashPass orders are billed at 14%. If DashPass is the typical ~30% share of your DoorDash volume, the blended effective rate sits at 10% × 70% + 14% × 30% = 11.2%, not 10%. For a chain doing $1.8M / 4 weeks through DoorDash, that 1.2pp blended drift is ~$73K / year. The chef-led 16-unit group caught this on May 8 reading their own rate card; the fix is pulling DashPass share from the DD merchant portal, not changing the contract.
Read it
For the multi-unit operator
Why the labor screen managers actually want looks different
What is the labor metric that managers actually act on?
Not labor % alone. Managers act on drift — the gap between scheduled hours and clocked hours, plus the unbudgeted overtime that piles up when scheduled coverage doesn't match actual demand. The Labor Leak quick win shows network labor % vs budget, OT $ YTD, ghost shifts (clocked-in windows with no sales attached), and the top schedule-vs-clocked offenders by employee. Pull the timesheet on the top-drift row and you've found the leak.
Read it
For the multi-unit operator
How catering economics break for multi-unit operators
How can a 10-unit operator see where catering is leaking?
Catering tickets average 10x in-store tickets — so the channel is high-leverage. But three things go wrong at scale: (1) third-party catering platforms charge 15-18% in effective fees while Toast Catering charges 2-3%, (2) phone orders quoted to a regular get prepared but never invoiced, and (3) store-level deposit policy is inconsistent. The Catering Leak quick win renders per-store reconciliation gap and channel mix so the leak is obvious in one screen.
Read it
For the multi-unit operator
What does "people-native AI" mean for restaurants
What is people-native AI and how is it different from AI for the back office?
People-native AI means the people on the floor — line cooks, servers, bartenders, dishwashers, managers — see the same numbers the back office sees. Shift Pulse renders tonight's covers vs forecast, your station median, your shift goal, and your streak. The crew doesn't need to be an analyst to know if they're on pace. The back office doesn't need to wait for end-of-night to see drift. One platform, both audiences, same source of truth.
Read it
For the multi-unit operator
How we walked our own number back — $8.3M to $1.81M
Why should I trust the recovery number a vendor shows me?
You shouldn't — until they've shown you they'll correct it down. Here's the time we did. Our first reconciliation for a 16-unit chef-led group put the recovery surface at $8.3M a year. The number was on the screen, the math was internally consistent, and the design partner had already seen it. Then one signal broke the story: a rollup view reported chain net at $72M for a four-month period — in a group that has never done $72M in a year. The rollup was double-counting, and the model was confidently extrapolating from a doubled number. We pulled the model down, re-pulled net sales from the leaf-channel view, and de-duplicated. The honest network number came out at $15.72M reconciled across 545,677 orders. The honest recovery surface came out at $1.81M — about 22% of what we'd originally reported. We walked it back in writing, to the partner who'd already seen the original figure. That's the rule the company runs on: every figure is source-tagged Verified, Estimated, or Unverified, and when the model is wrong, we publish the correction. The discipline of correcting your own number down is the product. It's why the design partner stayed — and why the next number we shipped landed without anyone needing to verify it twice. The full write-up, with the timeline, is at never86.ai/case/walked-the-number-back.
Read it
For the multi-unit operator
The void rate question every CEO should be asking
What does a healthy void rate look like by store?
There is no single number. A void rate is healthy when it's consistent across stores doing the same volume, and explainable when it spikes. Void Hunter ranks employees against your own peer median per store — not against an industry benchmark you can't verify. The pattern is the signal, not the absolute number.
Read it
For the multi-unit operator
Why your DoorDash effective fee is higher than your contract
Why is my DoorDash effective fee higher than the contracted rate?
Your contract is your floor, not your ceiling. Promotions, ad credits, dispute chargebacks, and refund-class voids quietly stack on top. We sum every line of the DoorDash MFS settlement and divide by gross sales — that's your real effective rate. Operators routinely see 3-7 points of leak above contract.
Read it