Financial analysis for food and beverage operators, where the output is a number an operator runs the business on: how much to order, how many staff to roster, whether a menu item stays. The role was advertised as a senior engineer for the analysis itself. The risk sat underneath the analysis, in whether the figures tie back to the point-of-sale system, the inventory count and the payroll records. Restaurant finance has several traps that a capable generalist reproduces faithfully, and each one produces a number that looks entirely reasonable.
Three restaurant numbers each have a plausible wrong version, and the analysis inherits whichever one it is given.
Restaurant operators manage by a small set of ratios. Food cost is what the ingredients cost as a share of food sales. Labour cost is what staff cost as a share of sales. The two together are often called prime cost, because they are the largest costs a restaurant controls week to week. Each ratio is a fraction, and in each one the top or the bottom has a version that is easy to get and subtly wrong.
Sales first. The till records gross sales, everything rung in. Some of that never becomes revenue. A comp is food given away, to a regular or to make up for a mistake. A void is an item rung in and then cancelled. A discount is a price reduction. Net sales is what remains, the money the business actually earned, and it is the right bottom line for every ratio. Using gross makes every cost look smaller than it is.
Labour next. Wages are what employees are paid. The cost to the business is higher, because the employer also pays payroll taxes and contributes to benefits and insurance. The full figure is called burdened labour. A labour ratio built on wages alone always flatters the business, and it is the ratio operators use to decide how many people to put on a shift.
Cost of goods is the subtle one. Kitchens do not record each ingredient as it is used. They count stock periodically and infer usage: what you had at the start, plus what you bought, minus what you have left at the end. That works only if the closing count is complete and taken at the right moment. If the count is late, partial or skipped, stock still sitting in the walk-in is missing from the last term and gets treated as used. Food cost then rises and falls with when people counted, not with what the kitchen did.
Three numbers mean something different here than they do in most businesses, and each has a plausible wrong version. Sales reported gross, before comps, voids and discounts are taken off. Labour taken as wages, without the taxes and benefits that make up its true cost. And cost of goods recognised from a stock count instead of from when stock was actually consumed, so food still sitting in the walk-in is treated as sold.
Restaurant analysis usually goes wrong before the analysis starts.
The analytical work in hospitality finance is rarely the weak point. A competent analyst can build trends, flag outliers and write clear commentary. The trouble is what they are handed. Each of the three core figures has a version that is the easiest one to extract from the systems, falls in a believable range, and is wrong in a consistent direction. Nothing in the output signals the problem, because the wrong version and the right one look alike.
The second pattern is that nobody checks the figures against their sources before interpreting them. Sales from the analysis are not tied to the point-of-sale totals, labour is not tied to payroll, cost of goods is not tied to the inventory records. When they disagree, the analysis still runs, and it produces commentary as fluent as it would have been on correct numbers.
The reflex, and the fix.
Improve the analysis
It is the advertised role and the visible deliverable.
Better analysis of figures that do not reconcile produces a more confident wrong answer. An operator acting on a food-cost percentage distorted by a stock-timing error makes a real purchasing or staffing decision on it.
Reconcile first, then analyse
A reconciliation gate before any insight is produced, and three domain definitions pinned explicitly instead of assumed.
Analysis running only on figures that tie back to their sources, with the insight withheld whenever they do not.
Withhold the insight when the numbers do not reconcile, because an operator has no way to tell a reconciled number from an unreconciled one.
Net sales, not gross
Comps, voids and discounts are the difference between what the till rang and what the business earned. The gap is large enough to move every ratio built on top of it.
Burdened labour, not wages
Employer taxes and benefits are part of what labour costs. A labour percentage computed from wages alone understates it every time, which is the worst direction for a number used to decide staffing.
Consumption timing, not stock count
Cost recognised when stock is counted, instead of when it is consumed, treats inventory on hand as sold. The resulting food-cost figure moves with how often and when people count, and not with what the kitchen did.
A reconciliation gate before any narrative
If the figures do not tie back to the source systems, no insight is produced. Silence is the correct output, because a plausible narrative over unreconciled numbers reads exactly like a correct one.
What each choice cost, and what it bought.
APutting reconciliation ahead of the analysis
The engagement was framed around analysis, and analysis is what an operator can see and value. Placing a reconciliation gate in front of it meant spending early effort on matching figures to the point-of-sale, inventory and payroll systems, work an operator never sees. It bought analysis that only ever runs on numbers that tie back, which is the precondition for any of it being worth reading.
BPinning the three definitions in writing
Each of net sales, burdened labour and consumption-timed cost of goods could have been left as an assumption, on the reasoning that everyone in the trade knows them. The generalist version of each is the one the systems hand over most easily. Writing the definitions down and checking every figure against them cost little and removed the most likely source of a consistently wrong ratio.
CWithholding the insight when reconciliation fails
An operator who asks for analysis and receives nothing for a period is not pleased, and the temptation is to produce something with a caveat attached. Caveats do not survive being read on a phone between services. Withholding the insight, and reporting which figures failed to tie back, cost some goodwill in the short term. It meant every insight the operator did receive was one they could act on.
DDrawing the line at source-system correctness
Reconciliation can show that figures agree with the systems they came from. It cannot show those systems were set up correctly. Saying so explicitly narrowed what the work claimed. It also stopped anyone treating a reconciled figure as an audited one, which would have been an overstatement nobody could see.
A single headline number hides where a system fails. This work was scored on the dimensions that actually decide whether it holds in production, measured on real, held-out cases rather than the demo path.
No measured figures are published for this engagement. Outcomes are described qualitatively.
A confident analysis built on a food-cost percentage that counted the stock in the walk-in as sold, with an operator making staffing and purchasing decisions on it.
Figures reconciled against the point-of-sale, inventory and payroll systems, with sales net of adjustments, labour burdened, cost recognised on consumption, and the narrative withheld whenever the reconciliation fails.
What it owns, and what it hands to a person.
Reconciliation proves the figures agree with the source systems. It does not prove the source systems are correct. A point-of-sale setup that files items under the wrong category at the till produces figures that reconcile perfectly and are consistently wrong. That is a separate audit, and this work does not claim it. Operational advice on what to do with the numbers is also outside it.
Wherever a system produces a number somebody runs a business on, the reconciliation gate comes before the analysis. A domain's own definitions have to be pinned explicitly, because a generalist will reproduce the plausible version of each one and nothing in the output will look wrong.
Restaurant ratios should be built on net sales, after comps, voids and discounts, because gross sales make every cost look smaller.
The true cost of labour includes employer taxes and benefits. A ratio built on wages alone always flatters the business.
Cost of goods is inferred from opening stock plus purchases minus closing stock, so a late or partial count turns stock on the shelf into apparent usage.
Reconcile figures against their source systems before interpreting them. A fluent analysis of unreconciled numbers is indistinguishable from a correct one.
Reconciliation shows that figures match their sources, not that the sources are right. Say which of the two a piece of work claims.