A budget miss doesn’t automatically mean you spent too much. It can mean you’re cutting the stage that turns work into something a customer can receive and pay for.
Call it the variance-to-constraint chain: budget line, operating stage, then saleable output, queue, late order, credit, or cash.
Across-the-board cuts are expensive because they treat every dollar as equally detachable. Dispatch, final inspection, competent setup, and an unused subscription don’t have the same relationship to throughput.
The general ledger is not invited to the handoff meeting.
The Starting Point
Build your budget from sales drivers first: units or hours by rep, territory, service line, or product family, priced and summed from the bottom up.
In an invented four-crew landscaping business, each crew lead forecasts contracted accounts, renewal probability, and new-work volume before the roll-up. Then size labour, material, fuel, crew-hours, inventory, and overhead against that volume.
A top-down 15% growth target is an ambition. It isn’t an operating load.
Once the period closes, calculate variance as actual − budget in dollars and (actual − budget) / budget as a percentage. Keep your sign convention honest: higher revenue than plan is favourable, higher cost than plan is unfavourable.
A $500 overspend on a $50,000 payroll line is 1%. Repeated for twelve pay periods, it’s $6,000. That’s why the signs and both views belong in the report before a hurried meeting relabels them.
Use a stated dual threshold: a dollar floor and a percentage threshold. The corpus carries a working percentage range of 5–10% and an illustrative policy of 10% or $25,000, whichever fires first.
Those aren’t universal cutoffs, and your house rule may instead require both tests. What matters is that you decide the rule before a bad month arrives, rather than rediscovering materiality while you’re looking for a reason to cut.
Review exceptions line by line monthly, then step back quarterly to ask whether the plan still describes the business you actually have.
A repeated favourable variance can be slack. A repeated unfavourable one can be a changed supplier price, demand condition, or operating assumption. Neither means “tell everyone to spend less.”
The Map
Give your largest unfavourable variance an operating address.
“Overtime” might be final test, dispatch, installation, correction, or incomplete work packets. “Materials” might be normal demand, scrap, rework, a quality failure, or price leakage. Name the stage before choosing the remedy.
Walk one order, job, or service case from request to cash receipt. Record queues, handoffs, decisions, corrections, and return trips, including the spreadsheet and email work.
In one 60-minute observation, a technician might work for 31 minutes, wait 18 for authorization, walk for 7, and correct paperwork for 4.
Put commitments you can’t disrupt and informal expedites on the current-state map. A conference-room account usually omits both.
At your suspected restriction, keep the same operational unit for load and effective capacity: orders per hour, invoices per clerk-day, machine minutes per shift, or qualified leads per salesperson-week. Include changeovers, breaks, maintenance, staffing, and expected rework.
As a locally tested starting control, plan only 80% of demonstrated constraint capacity and hold 20% for variation and recovery. State the buffer in time — perhaps 4–6 constraint-hours, pausing release above 8 and reviewing expedites below 2.
Nameplate speed is not saleable-output capacity.
Then hold the evidence in three layers:
- Process: constraint completions per staffed constraint-hour, queue age, availability, and first-pass accepted units.
- Customer: complete orders shipped on the original promise, callbacks, returns, and corrected invoices.
- Financial: shipped net revenue, gross profit, transaction-level realized price, and cash collection.
One layer moving isn’t enough, and you need all three. More jobs released can be unfinished work, lower rework can be a narrower defect definition, and a higher average selling price can be a richer mix or departing low-price customers.
Publish 3 of 37, not just 8%, beside a defect or service rate, and keep the original promise rather than a rescheduled date. The chain is built to make those little miracles answer questions.
Find the Restriction
Use the symptom to find the stage that broke before you authorize a cut.
Twelve jobs waiting before final test for a 42-minute median, while upstream completes one every 9 minutes, is a test-station question. It isn’t a reason to buy faster upstream tools.
Revenue misses while a resource is visibly busy → the named resource may be a false constraint
Test whether giving it better input or more available work raises completed system output. If shipments, accepted service work, or customer delivery don’t move, follow the oldest queue and inspect the next restriction instead.
Local cycle time improves but total output does not → the measure ends before the customer-facing end
The change may have pushed work into a downstream queue or created rework. Measure complete orders, original-promise service, returns, and corrected invoices alongside the local cycle.
Budgeted overtime falls while overdue orders or queue age rise → the cost cut has starved the actual constraint
Separate productive run time from setup, starvation, blockage, planned downtime, and unplanned downtime before celebrating the lower payroll line.
A cost variance looks favourable after a volume change → the comparison is volume-blind
Recalculate the budget at actual activity volume with a flexible budget, then separate the volume effect from the price or unit-cost effect. A static comparison bundles them together.
Output rises after a cut, but rework, returns, credits, or complaints rise too → the gain is being measured before quality and service
Track units requiring any correction, rework events, correction hours, and the original customer promise, rather than an average that can sand the splinters off.
List price rises but gross profit or cash does not → price realization is broken
Trace discounts, rebates, credits, allowances, included scope, overrides, and stale quote or invoice rules through to the settled transaction.
The rule isn’t that every busy resource gets protected. Test capacity at the actual constraint against customer-facing output before you remove it.
A machine can be busy running the wrong jobs, and a non-constraint can be idle without constraining flow. Utilization alone is a bad witness.
The Proof
| Budget line | Process measure | Customer measure | Owner | Intervention | Proof of durable gain |
|---|---|---|---|---|---|
| Overtime or temporary labour | Constraint completions per staffed hour; productive time, starvation, and blockage | Complete orders on original promise; overdue orders | Operating owner | Improve work readiness or availability at the suspected constraint in a bounded trial | Total system output rises without a worse rework, return, safety, or service record after overtime is disclosed |
| Rework, scrap, or correction labour | First-pass accepted units; reworked-unit incidence, events, and hours | Returns, callbacks, corrected invoices | Quality and operating owners | Test the missing work condition or decision point, then update standard work only if accepted | Same quality boundary; fewer corrections; original-promise service and shipped output remain comparable |
| Discounts, credits, or freight concessions | Invoice-line realized price; posted increase, discount, credit, mix, and volume bridge | Retention, quote conversion, cancellations, complaints | Commercial owner | Correct quote, approval, and invoice execution for one defined cohort | Matched transactions retain price after concessions without unexplained customer loss or mix substitution |
| Maintenance, inspection, or dispatch spend | Availability, queue age, interruption type, and accepted completions | Shipped or accepted work; delivery against original promise | Named stage owner | Test the smallest reversible change that protects the flow | Improvement persists after backlog release, deferred work, and unusual staffing are marked |
Every metric in your table needs a data contract: the event counted, unit, timestamp, denominator, exclusions, owner, and system of record. Timestamp arrival, start, interruption, completion, and acceptance at the suspected constraint.
“On time” has to state whether the clock ends at completion, shipment, delivery, or customer acceptance — and it uses the original promise, not a date moved after the order was already late.
Preserve the pre-change extract and original calculation. A later correction creates a restated series beside it. It doesn’t quietly improve the baseline.
Stratify elapsed time by shift, operator, product family, order type, and setup family, and report its median and tail rather than just an average. One stranded customer can disappear inside a run of routine jobs.
The Test
Run one bounded correction through Plan-Do-Study-Act. State what changes, where, for whom, how long it runs, who may stop it, and how the old method is restored.
At a test station, reserve the first 15 minutes of one shift for calibration, require complete test records from upstream, and run it for 10 working days against a comparable shift.
If you change staffing, routing, incentives, software, and price together, you may get lucky, but you won’t know why.
Your first 90 days are a useful control interval, not a licence to declare the operation fixed when the interval closes. Keep the daily or weekly sequence.
Mark intervention dates, shutdowns, promotions, backlog release, supplier failures, staffing changes, and accounting cutovers. Adjust or disclose changes in business days, demand, product mix, staffed hours, opening backlog, material availability, and seasonality.
Don’t decorate a thin series with a capability score. NIST guidance describes about 50 independent observations as a meaningful starting point for several capability indices, once process stability and the other assumptions are checked.
Control limits describe observed variation. Customer specifications and management targets describe required performance. Meeting one doesn’t prove the other.
For the budget loop itself, carry the original forecast, latest estimate, and actual result. Score receipts and payments separately, because a small net miss can hide two large errors that happened to cancel.
A rolling forecast is not an alibi for the budget. It’s the place where the operating facts finally get admitted.
Keep your owner report short: 5 to 8 elements, readable in under two minutes.
Cash position, receivables aging, revenue booked, the constraint measure, rework, and original-promise service can earn weekly attention. Full P&L variance, gross margin, and net margin belong in the monthly review, when the period is complete.
A lower expense line is not an operating improvement. A customer-visible gain that survives the counter-metrics is.
Find the restriction before you cut the resource. Prove the change at the customer end. Keep the cash, rework, and backlog in the same frame.