Lower sales do not prove that “we need more advertising.” Higher sales do not prove that the business is healthy, either. A useful review checks demand, economics, customers, quality, and capacity before declaring what the problem is.

The purpose is not to identify a definitive “root cause” in one meeting. It is to establish what changed, where, what the available evidence says, and what needs investigation or immediate protection. The UK Government Digital Service recommends reviewing trends and segments alongside performance records and user research rather than trusting an isolated number [R01][R02]. If the records are incomplete, first reconcile the data sources. Once KPIs are defined, display them with context.

1. Review business health across connected dimensions

Health cannot be assessed through sales alone. Examine the following areas without counting the same event twice:

Dimension Health question Minimum evidence Common mistake
Demand and sales Are the right customers requesting and buying the offering? Inquiries, net orders, cancellations, and unmet requests Treating inquiries as sales
Customers and repeat business Do expectations get met, and do eligible customers come back? Complaints, refunds, relevant repeat purchases Substituting likes for satisfaction
Economics Does each transaction generate enough contribution? Revenue, variable costs, product mix, fixed expenses Treating sales as profit
Liquidity Can obligations be paid when due? Reconciled cash, receivables, payables, due dates Assuming an invoice is cash
Quality Are goods and services delivered correctly and safely? Defects, omissions, rework, incidents Hiding serious failures behind a favorable average
Capacity and service Can resources meet the promises being made? Slots, lateness, load by time window, queues Using daily capacity to assess an hourly peak
People and processes Can the work be completed consistently? Absences, handoffs, training, deviations Blaming an individual before examining the system
Risk and external conditions Has a supplier, constraint, or relevant outside condition changed? Operational alerts, applicable rules, market or weather changes if relevant Treating general news as a proven cause

A solo operator can review all eight areas in one spreadsheet. Larger businesses may have different owners, but the definitions must remain compatible. For cash evidence, use daily cash, receivables, and payables controls; for capacity, use the operating capacity calculation, not the sales total.

2. Separate symptoms, anomalies, deviations, risks, problems, and opportunities

A symptom is a visible signal: “fewer orders than last month.” A deviation is a difference from an agreed target or condition: “88% on-time performance against an internal 92% threshold.” An anomaly is unusual relative to a reasonably established pattern, not merely an uncomfortable number. A confirmed problem requires evidence of an undesirable outcome, but its cause may still be unknown. A risk is an uncertain future event with potential consequences. An opportunity is a potentially beneficial change that still needs evaluation.

Ten complaints about late deliveries establish ten recorded incidents; they do not demonstrate whether the source was cooking, dispatch, an unrealistic commitment, traffic, or incorrect timestamps. Avoid writing “cause: not enough drivers” when the only confirmed observation is “late delivery.” Testing competing explanations belongs in cause investigation.

When there is a credible safety hazard or critical compliance issue, take appropriate protective action immediately rather than waiting for weeks of reporting. Diagnosis must not delay customer protection.

3. Set a reasonable review cadence and evidence-based alerts

You do not need to summon the whole company every morning. For each dimension, specify who reviews it, when, from which source, and which decisions that person can make. A microbusiness may check cash and live commitments daily, quality and demand weekly, and margins and sustainability monthly. The right cadence follows risk and the speed at which conditions change; it is not universal.

Use explicit written statuses:

  • Stable / monitor: no material deviation, or movement consistent with what the available complete records could reasonably show.
  • Investigate: repeatable anomaly, significant impact, or data gaps that prevent reliable interpretation.
  • Contain / act: severe safety concern, critical noncompliance, near-term inability to pay essential obligations, or another non-negotiable condition.

If you use traffic-light colors, add the words and the reason. Thresholds should reflect objectives, evidence, capacity, or applicable requirements. A 5% boundary chosen for visual convenience is not an industry standard. Time plots can reveal patterns, but statistical control limits require their own methods and data. ASQ distinguishes common from special-cause variation using properly constructed control charts [R04]. A day below target is not automatically “out of control.”

4. Triangulate evidence without requiring a sophisticated system

Evidence becomes more informative when you check different perspectives. If orders fall, compare sales records, turned-away requests, menu availability, actual opening hours, returns, support conversations, and known external changes. Simultaneous events are not necessarily cause and effect: a competitor might launch a promotion on the same week that an essential product was unavailable.

Customer research and observations from frontline workers can reveal questions missing from the spreadsheet. GOV.UK guidance recommends combining quantitative monitoring with qualitative research to guide improvement [R01][R02]. Record the source, number of cases, and alternative explanation for each observation. One dramatic review is not a representative sample.

If something was never recorded, write “unavailable,” not “zero.” Without a log of unmet requests, lower sales could mean less demand, stockouts, or constrained capacity. Sales history alone cannot distinguish those explanations.

5. Break down averages by segment and operating context

ASQ calls stratification the separation of pooled observations into meaningful groups to reveal patterns by shift, day, team, supplier, item, or other relevant condition [R03]. For a business, examine time slot, location, order type, product category, channel, and comparable period as needed. For digital products, device, journey stage, and consistently defined user groups may matter.

For example, if one location fulfills 90% on time and another 60%, the combined rate depends on their respective order counts. A change in which location handles more orders can move the overall average even if neither site improves. Keep numerators and denominators before aggregating. Avoid unnecessarily identifying individual customers or drawing strong inferences from tiny groups.

Cohorts can follow comparable customer groups over time, especially for repeat business, but a full cohort model is not required for every weekly decline. Choose the grouping that could actually change the decision.

6. Worked case: sales are down, but the cause is still unknown

This is a separate, entirely hypothetical “Almuerzo Cerca” scenario; do not combine it with the B01 four-week example. Compare two nominally similar four-week periods, with an unchanged assumed S/30 price per lunch. Suppose 480 orders were delivered in the reference period and 420 in the later period. Illustrative gross sales of delivered lunches fell from 480×30=S/14,400 to 420×30=S/12,600: S/1,800, or 12.5%, lower. Discounts, total costs, tax treatment, and actual collections remain unknown. These are not profit or cash comparisons.

Observation Reference Current period Supported conclusion
Delivered orders 480 420 60 fewer deliveries (−12.5%)
Walk-in channel 400 300 100 fewer; inspect hours and availability
Delivery platform 80 120 40 more; channel mix has changed
Days with a documented stockout of a key item 2 10 More affected days; lost demand remains unquantified
Customer complaints recorded 11 15 Illustrative rates: 11/480 = 2.29%, 15/420 = 3.57%; reasons not yet classified
Reconciled cash and obligations Unavailable Unavailable Cannot conclude liquidity is stable or worsening

Complaint rates in this table mean recorded complaints per delivered order, assuming the system counts each complaint once. They do not establish the percentage of distinct orders with at least one complaint; that would require deduplication by order ID. The channel figures sum exactly to the period totals and are not duplicate source records. More stockout days do not prove that stockouts caused the sixty missing deliveries: we do not know the number of affected requests, the product mix, or whether the menu changed. Likewise, an increased complaint rate does not automatically explain the sales drop.

Before launching advertising, the owner tests five explanations: (a) genuinely lower demand; (b) an unavailable product; (c) reduced opening hours; (d) a changed effective price or channel commission; and (e) competitors or seasonality. Each requires evidence that could also refute it. Time-stamped unmet inquiries help test the stockout hypothesis; comparing advertised versus actual hours tests availability; channel-level contribution helps determine whether rising platform orders are economically attractive.

7. Prioritize what to investigate and what to contain now

Consider impact, urgency, reversibility, uncertainty, and the cost of obtaining evidence. Do not hide judgments behind arbitrary scoring. In this example, three current investigation priorities may be manageable, but three is not a universal quota:

  1. Stock and availability: ten affected days versus two. Audit unmet inquiries and inventory records, and contain any current shortages without declaring a root cause.
  2. The walk-in decline: orders fell from 400 to 300. Verify opening hours, customer communication, comparable foot traffic, and changes in the experience.
  3. Quality and complaints: recorded complaints rose from 11 to 15 while volume fell. Classify severity, type, and time window; act earlier if a serious hazard is found.

Liquidity is not measured in the example, so assign someone to reconcile it promptly rather than asserting that it is fine. A serious legal or safety alert would override these commercial priorities regardless of its score.

Next decide which problem needs causal investigation, not which solution to deploy. If lateness is concentrated in a repeatable step, use structured cause analysis. Testing interventions and continuous improvement will be covered in later Pillar 7 articles.

8. Make a one-page business health record

Each entry should include dimension, defined metric, value and denominator, period, valid comparison, evidence source, data completeness, justified status, open hypotheses, urgency, owner, and next check. One page can cover eight areas and three open priorities without revealing individual customer data.

Example: “Availability | ten stockout days in four weeks | investigate | inventory closing records | possible forecasting, supplier, or preparation issue | next check: match stockouts with unmet requests by time slot | owner: operations.” Keep containment (stop offering unavailable products) distinct from investigation (find out why the stockouts happen).

Save a dated version and a short decision history. If the metric later changes, verify whether performance changed or the way it was measured changed. A commercial target cannot justify suspending safety controls.

Common mistakes, limitations, and outcome

Avoid diagnosing from one metric, treating income as cash, combining unequal time horizons, labeling a coincidence as a cause, comparing sites without denominators, interpreting falling sales without checking availability, displaying status colors with no rationale, collecting intrusive customer details, or delaying critical action until a polished slide deck is complete.

By the end, you should have a defensible inventory of signals and risks, justified priorities, testable hypotheses, and a short evidence-gathering plan. A business health review does not prove why a problem happened; that is the next methodological task.

References