Humans spot-check only.
Your team of three reviewers manually checks every invoice your automated system generates - 2,000 per week. They're buried, overtime is killing your Cost Per Unit, and they still miss errors when fatigued. Your CFO asks why you have humans doing what your system already does. You realize: the humans shouldn't be doing all the work. They should be spot-checking it.
Spot-Check means humans review only a random sample of outputs - not every one. You get most of the Error Cost protection of full inspection at a fraction of the Labor cost, because Quality Control already proved that sampling catches systemic problems.
Spot-checking is an operational pattern where humans review a random subset of work product instead of every unit. The system (automated or otherwise) handles 100% of the volume. Humans inspect maybe 2-5% of it, looking for patterns that signal something has gone wrong.
This is the next step beyond Quality Control. QC taught you that sampling works because Variance of an average shrinks with sample size. Spot-Check applies that insight to your most expensive resource: human attention. Instead of putting a human in the critical path of every transaction, you pull them out of the main automated flow and have them audit a sample.
The word only matters. Spot-Check is a commitment: humans do not touch the main automated process. They review samples, investigate exceptions, and intervene when the defect rate drifts. Everything else flows through without human hands.
Human review is the most expensive line item you can put in a process. Every unit that requires a person to look at it carries their full Labor cost - salary plus benefits, taxes, and overhead. When you run the Unit Economics, full human review often makes an operation unprofitable at scale.
Consider the P&L impact:
Spot-checking breaks this tradeoff. You get quality signal at a cost that stays flat as volume scales.
1. Automate the primary path. Before spot-checking makes sense, you need a process that runs without human intervention. If humans are doing the core work, you don't have a spot-check - you have a workforce.
2. Define what you're sampling for. Spot-checks need criteria. You're looking for specific failure modes: did the system miscalculate a total? Did it apply the wrong Pricing? Did it misclassify an item? These connect directly to your Quality Gates.
3. Pull random samples at a fixed rate. A common starting point is 2-5% of volume, selected randomly. The randomness matters - if reviewers cherry-pick what looks suspicious, you lose the statistical validity that Quality Control depends on.
4. Track the defect rate. Every spot-check is a data point. If you review 50 items and find 1 error, your observed defect rate is 2%. Track this over time. The trend is more valuable than any single check.
5. Set escalation thresholds. Decide in advance: if the defect rate exceeds X%, what happens? Maybe you increase the sampling rate. Maybe you pause the automated system. Maybe you trigger a full Exception Review of recent output. This is your decision rule, defined before you need it.
6. Adjust the rate based on evidence. If six weeks pass with zero defects found, you can reduce the sampling rate. If a system change deploys and defects spike, you temporarily increase it. This is Graduated Autonomy applied to process oversight - earn trust through demonstrated performance, tighten controls when trust erodes.
Spot-checking is the right pattern when:
Do not use spot-checking when:
Your accounts payable team processes 8,000 vendor invoices per month. Three reviewers manually check every invoice before payment. Each reviewer costs $65,000/year including salary, benefits, taxes, and overhead ($31.25/hr). They review ~20 invoices/hour. Your automated matching system already flags mismatches on line items, purchase order references, and totals.
Current cost: 8,000 invoices / 20 per hour = 400 review hours/month. At $31.25/hr = $12,500/month in review Labor. Cost Per Unit for review alone: $1.56/invoice.
You run a Quality Control analysis on the last 3 months of reviewer corrections. Out of 24,000 invoices, reviewers caught 96 errors the system missed - a 0.4% defect rate. That's 32 errors per month (96 total / 3 months). Average Error Cost per missed invoice error: $340 (overpayments, duplicate payments).
Switch to spot-check at 3% sampling rate: 240 invoices/month reviewed instead of 8,000. That's 12 hours/month of review Labor = $375/month. One reviewer handles this in about 1.5 days.
Expected defect detection: at 0.4% defect rate, your 240-sample should surface roughly 1 error per month (240 x 0.004 = 0.96). This is enough to detect if the rate is drifting upward - a spike to 2% would show as ~5 errors in your sample, a clear signal.
Cost comparison: Old model = $12,500/month. Spot-check model = $375/month. You freed 388 hours of human capacity. Two of three reviewers can be redeployed to Exception Review of flagged items, vendor disputes, or other higher-value work.
Residual risk: Your reviewers used to catch about 32 errors per month during full review. Under spot-check, roughly 31 of those now flow through undetected - about $10,540/month in potential Error Cost (31 x $340). Notice this is close to the $12,125/month you saved in Labor. The economics only work if you use the spot-check data to reduce the defect rate itself. In month one, your spot-check reveals that the system miscalculates totals when a vendor invoice lists the same item at different quantities across multiple lines - the matching logic sums them as one line instead of two. You add a line-item disambiguation rule to the matching system. Month two's spot-check shows the defect rate dropped from 0.4% to 0.2%, cutting residual errors to ~16/month and residual Error Cost to ~$5,440/month. Now your net position is clear: $375 in spot-check Labor plus $5,440 in residual Error Cost versus $12,500 in the old model. The Feedback Loop between spot-checking and system improvement is what makes the transition profitable, not the Labor savings alone.
Insight: Spot-checking didn't eliminate errors - it changed the economics. You traded $12,500/month in guaranteed Labor cost for a monitoring system that costs $375/month plus residual Error Cost that you actively drive down through the Feedback Loop. Without using spot-check findings to improve the automation, you would be trading $12,500 in Labor for $10,540 in Error Cost - nearly a wash. The key was knowing the defect rate first through Quality Control, then using spot-check patterns to shrink it.
Your platform generates 50,000 pieces of user-submitted product descriptions per month through a self-service tool. An automated screening system flags 4% as potentially non-compliant (2,000/month). A human moderator reviews the flagged items. But your VP asks: how do you know the screening system isn't missing things in the other 48,000?
You implement a spot-check: pull 500 random descriptions per month from the 48,000 that passed the automated screen (roughly 1% sample rate). One moderator spends about 25 hours reviewing them.
Month 1: 500 reviewed, 3 actual violations found that the screening system missed. Observed defect rate in the 'passed' population: 0.6%.
Extrapolating: 0.6% of 48,000 = ~288 violations slipping through monthly. Each violation carries Compliance Risk - you estimate the Error Cost at $50 per incident on average (customer complaints, potential regulatory flags).
Expected monthly cost of missed violations: 288 x $50 = $14,400. The spot-check costs ~$780/month in Labor (25 hours at $31.25/hr). The spot-check doesn't catch all 288 - it catches ~3. But it tells you the rate exists, so you can improve the screening rules.
You feed the 3 missed patterns back into the automated system. Next month's spot-check finds 1 violation in 500 - defect rate dropped to 0.2%. Extrapolated cost dropped to $4,800/month. The Feedback Loop between spot-checking and system improvement is where the real value lives.
Insight: Spot-checking the passing population is where Operators often miss. You're not just checking what the system flagged - you're verifying what it let through. The sample from the 'good' pile is what tells you if your automation has blind spots.
Spot-checking means humans review a random sample, not every unit. The word only is a commitment - take humans out of the main automated flow and use their judgment on samples and exceptions.
The prerequisite is knowing your defect rate. Without a baseline from Quality Control, you can't set a meaningful sampling rate or detect when things drift.
The value isn't in catching individual errors - it's in detecting systemic drift early enough to fix the root cause before Error Cost compounds across thousands of units.
Sampling what looks suspicious instead of sampling randomly. If reviewers pick the items that 'seem off,' you've destroyed the statistical foundation. Your defect rate estimate becomes meaningless because you're not measuring the population - you're measuring your reviewer's intuition. Random selection is the entire mechanism that makes Quality Control math work.
Never adjusting the sampling rate. A 3% rate that made sense at 5,000 units/month might be wasteful at 200,000 units/month (6,000 reviews). Conversely, after a major system change, holding at 1% when you have no data on the new code is reckless. Tie sampling rate to your confidence in the system, not to a number you picked once and forgot.
Your fulfillment center ships 12,000 orders per week. A packing accuracy audit last quarter found a 1.2% error rate (wrong item, missing item, damaged packaging). Each packing error costs an average of $18 in returns processing and replacement shipping. You currently have two auditors checking every order before it leaves. Each auditor costs $45,000/year (salary, benefits, taxes, and overhead). Design a spot-check program: what sampling rate would you start with, what would it cost, and what's your escalation threshold?
Hint: Calculate your current full-inspection cost first, then figure out how many samples you need per week to reliably detect a meaningful change in the 1.2% defect rate. Think about what defect rate would make you want to intervene.
Current cost: 12,000 orders/week at ~15 checks/hour = 800 hours/week. Two auditors at $45K/year = $90K/year = $1,730/week. That's $0.14/order in audit cost.
Start with a 3% sample: 360 orders/week. At 15 checks/hour = 24 hours/week. One auditor handles this comfortably. Cost: $865/week (one auditor). Savings: $865/week = ~$45K/year.
At 1.2% defect rate, you'd expect ~4.3 errors in 360 samples per week. If you see 8+ errors in a week (roughly double the expected rate), that's your escalation threshold - it suggests the true defect rate may have jumped above 2%. At that point, increase sampling to 10% for a week to confirm, and investigate root cause.
Residual risk: ~144 errors/week x $18 = $2,592/week now flow through instead of being caught. But your auditors weren't catching all of them at full inspection either (fatigue). The real play is using the spot-check data to fix the packing process - reduce the 1.2% rate itself.
You manage an automated pricing engine that updates 40,000 product prices daily based on competitor data and margin rules. A pricing error (setting a price below cost or absurdly above market) has an average Error Cost of $200 per product per day before someone notices. Your current process has a product manager manually approve every batch. She spends 6 hours/day on this and is your Bottleneck for launching new Pricing rules. What do you propose?
Hint: The Bottleneck framing matters here - the problem isn't just cost, it's that human review is constraining your Throughput. Think about what the spot-check buys you beyond Labor savings.
Current state: Product manager at ~$130K/year spends 75% of her time on price approvals (effective cost: $97.5K/year on review). She's the Bottleneck - no pricing rule ships without her approval, creating a queue.
Proposal: Implement spot-check at 1% = 400 products/day. She reviews 400 prices in ~1 hour instead of 6. Freed 5 hours/day for strategic Pricing work and rule development.
Set automated guardrails as the primary quality gate: any price below cost or more than 30% above prior day triggers automatic hold (these are your Exception Review items). Spot-check the 400 from the population that passed guardrails.
Escalation: If spot-check finds more than 2 pricing errors in a day's 400-sample (0.5% rate), freeze the batch and investigate. Expected errors at a healthy 0.1% base rate: ~0.4 per day, so 2 is a strong signal.
The bigger win: removing her from the critical path increases Throughput for pricing experiments. You can now test new rules without waiting for her 6-hour review cycle. The spot-check actually makes you faster, not just cheaper.
Spot-Check is a direct application of the sampling principles from Quality Control. Where QC taught you that sampling gives reliable estimates of defect rate, Spot-Check operationalizes that insight by committing to only using human review on samples. This connects forward to Graduated Autonomy - the idea that systems earn less oversight through demonstrated performance, and spot-check sampling rates are the mechanism for granting or revoking that autonomy. It also feeds into Exception Review, which handles the items that fail automated checks. Together, these form a layered Operations pattern: automation handles the volume, spot-checks monitor the automation, and Exception Review handles the outliers. The economic logic ties back to Unit Economics and Throughput - every human you remove from the main automated flow is capacity you can redeploy to higher marginal value work.
Disclaimer: This content is for educational and informational purposes only and does not constitute financial, investment, tax, or legal advice. It is not a recommendation to buy, sell, or hold any security or financial product. You should consult a qualified financial advisor, tax professional, or attorney before making financial decisions. Past performance is not indicative of future results. The author is not a registered investment advisor, broker-dealer, or financial planner.