An improvement does not end when a procedure changes. It becomes useful when we know what changed, what unintended effects occurred, and how to sustain the result.

The PDCA cycle—Plan, Do, Check, Act—organizes an intervention around hypotheses, observations, and decisions. The American Society for Quality (ASQ) describes PDCA as an iterative method for planning changes, testing them, reviewing results, and acting on what is learned [R01]. No certification or large organization is required: a café can use a sheet of paper, and a digital service can use a change log. Yet neither a recognized method nor a persuasive chart proves that a change caused an improvement.

1. Decide when PDCA is the right tool

Continuous improvement means repeatedly improving products, services, and processes, through incremental changes or more substantial breakthroughs [R02]. PDCA is useful when a condition can be observed, a feasible change can be introduced, and its consequences can be checked. It must not delay immediate containment of serious harm: when safety or an applicable legal requirement is at stake, protect people and address the urgent issue first.

These concepts operate at different levels. PDCA provides four phases for a change. TQM, Lean, and Six Sigma are broader management approaches or families of methods. They may incorporate cycles, but they are not synonyms for PDCA [R02]. Nor should every minor incident become a formal improvement project.

Begin by specifying what improvement means and the boundaries it must respect. If all we know is that “orders are delayed,” define the observable signal and investigate competing explanations first. PDCA does not replace diagnosis.

2. PLAN: define an observable problem and predict the effect

Record at least the problem, population, baseline, period, objective, proposed change, owner, risks, guardrail metrics, and decision to be supported. Baseline and follow-up must use the same counting rules. Define who prepares the data, who checks its quality, and who can stop the test. A “7% target” is a choice for this business—not a mandatory threshold in PDCA.

Before acting, write a conditional prediction: “If we adjust preparation batches by time slot without changing portion size or safety rules, we expect less food waste; we will not accept the change if unmet requests increase or contribution after waste falls per fulfilled order.” State what could weaken the hypothesis. Waste might fall because fewer orders are accepted, rather than because preparation improved.

PLAN field Proposed record for “Almuerzo Cerca”
Problem 60 lunches discarded from 500 prepared across four weeks: 12.00%
Change Adjust time-slot batches using confirmed orders and demand signals
Local objective Prepared-food waste ≤7% over a comparable period
Service safeguard Unmet requests / (fulfilled + unmet observable requests) ≤4.5%
Economic safeguard Contribution after waste per fulfilled order ≥S/10.64; do not apply the rounded figure as an exact accounting rule
Owner Operations lead, reviewing records with sales and cash owners
Restriction Never remove essential safety buffers or applicable food-safety checks

Use the KPI definition guide for reproducible metrics instead of inventing different definitions for every test.

3. DO: make the change small, controlled, and observable

Do does not mean rolling out to every location. Start with one shift, one product line, or a comparable set of days when this protects service. The Institute for Healthcare Improvement (IHI) describes small tests, explicit predictions, and learning across cycles within PDSA (Plan-Do-Study-Act), a related variant whose terminology is not identical to PDCA [R03]. Here we apply the practical principle of scoping reversible changes narrowly without claiming every PDSA convention is a formal PDCA requirement.

Record the date, batch, owner, planned production, actual production, fulfilled orders, discarded units, unmet requests, exceptions, incremental costs, and concurrent changes. Avoid simultaneously changing the menu, pricing, opening hours, and sales channel if you then need to understand why results moved. Log deviations from the plan and who authorized them.

Set stop conditions before starting: food safety concerns, a critical quality failure, loss of traceability, promises the business cannot fulfill, or unmet demand exceeding the agreed local safeguard. Never continue a harmful trial merely to “finish the sample.”

4. CHECK: compare results, denominators, and side effects

Calculate counts and rates using comparable populations and windows. For food waste, use discarded/prepared. For shortages, define observable requests consistently: unmet/(fulfilled + unmet). For economics, declare costs and assumptions; revenue is not interchangeable with cash.

Look at the direction of change but also the variability by day and shift, sample size, seasonality, missing records, and concurrent adjustments. A favorable average may hide one shift that worsened. IHI advises comparing outcomes with predictions, recording surprises, and planning further tests rather than announcing success after one comparison [R03].

Use the business performance dashboard to monitor results over time, without mistaking it for an explanation of why a metric moved. When attribution matters, use an appropriate design and evidence: see operational improvement experiments.

5. Worked example: waste, shortages, and contribution

All figures and prices are fictional. We reuse exactly the two four-week periods from P7-01 to illustrate how PDCA can be interpreted. The second period is treated here as a hypothetical trial scenario, not as a real experiment or proof that the periods were statistically equivalent. Assume a price of S/30 per fulfilled lunch, S/18 variable cost per unit sold, and S/10 already incurred per discarded unit, without double counting. Fixed expenses and taxes are excluded from the simplified contribution calculation.

Measure and calculation Before (4 weeks) Hypothetical after (4 weeks)
Prepared / discarded / fulfilled 500 / 60 / 440 520 / 31 / 489
Unmet requests 20 21
Waste rate 60/500 = 12.00% 31/520 = 5.96%
Unmet-request rate 20/(440+20) = 4.35% 21/(489+21) = 4.12%
Contribution after waste 440×(30−18)−60×10 = S/4,680 489×(30−18)−31×10 = S/5,558
Contribution per fulfilled order 4,680/440 = S/10.64 5,558/489 = S/11.37

PLAN: lower waste to ≤7% while keeping unmet requests ≤4.5% and per-order contribution at least as good as the unrounded baseline. DO: prepare batches by time slot, recording exceptions. CHECK: the three conditions appear satisfied in this example. Waste falls by 6.04 percentage points, and total simplified contribution rises by S/878—but fulfilled volume also increases by 49 units. Do not confuse observed improvement with incremental profit attributable to the intervention. ACT: provisionally retain controls, repeat under different conditions, and look for a sustained result before organization-wide standardization.

What could invalidate an overconfident conclusion? Greater demand, different product mix, easier routes, or promotions during the second period; unrecorded discarded units; or an unmet-request safeguard that worsens in one segment. Also assess additional labor hours created by batching: the simplified calculation does not include them. Those costs must be added before making a final profitability claim.

6. ACT: adopt, adapt, stop, or investigate again

Act produces a documented decision, not an unsupported “it worked.” Four reasonable outcomes are:

  • Adopt with monitoring: evidence is adequate, constraints were verified, and there is a way to sustain the practice.
  • Adapt and retest: overall direction is promising, but one shift, product, or expense requires adjustments.
  • Stop or roll back: acceptance conditions fail or harmful effects emerge.
  • Investigate further: records are incomplete or competing explanations remain plausible.

If adopting, revise the current preparation instruction, training, ownership, incident log, and review date. IHI distinguishes testing from implementation: permanent implementation embeds the new practice in normal workflows and requires adaptation and follow-up [R04]. For procedure versions and operational control, consult the operating-procedures guide.

7. Keep the improvement working and restart the cycle

Set a follow-up schedule appropriate to the business—for example, weekly during adaptation, then monthly once stable. Preserve the original baseline, the implementation period, any definition changes, and the subsequent measurements. If waste climbs again, investigate capacity, training, new products, and process variation before assigning blame.

Maintaining an intervention is not mandatory forever. Retiring a once-useful rule may also be a sound decision. When several possible improvements compete for time, document them and prioritize them in an improvement roadmap instead of running three conflicting trials with the same people.

Reusable template: one-page PDCA Canvas

Field Control question Evidence
ID and owner Which cycle is this and who decides? ID, date, accountable person
Problem and baseline What condition are we trying to change? Source, period, numerator/denominator
Prediction What should improve if our hypothesis holds? Expected shift and competing explanations
Guardrails What cannot get worse? Service, quality, cost, safety
Test What changes, where, and for how long? Log and stop rule
Results What happened versus the prediction? Before/after, costs, exceptions
Decision Adopt, adapt, reverse, or investigate? Rationale and approver
Sustainment Who maintains the current practice? Instruction, training, next review

The accompanying editable CSV supplies this template, with a matching Spanish version. A spreadsheet, paper card, or shared note is enough for an initial cycle. Additional software is not required.

Common mistakes and limitations

Defining “improvement” after seeing the result; comparing unequal time windows; using sales as a proxy for all demand; ignoring shortages or safety; treating lower waste as proven causation; overlooking implementation costs; standardizing a practice that worked for just one employee; and failing to review later. PDCA is a decision-and-learning practice, not a guarantee of success.

What you should have by the end

A documented PDCA cycle with a problem, prediction, safeguards, execution record, before/after evidence, reasoned decision, standard-work owner, and follow-up date. The point is safer, more understandable changes—not more paperwork.

References