← Research library

Practical research guide · desk research and BespokeData methodology · 18 Nov 2026

Operational analytics opportunity assessment

Find a measurable operational problem, test the explanation and assign an improvement your team can review.

Intended reader: Owners and operations managers with recurring delays, rework or unclear performance measures.

THE EXECUTIVE TAKEAWAY

Define the process and measures consistently, establish a baseline, and prioritise an evidence-backed pilot with a named owner. Treat bottleneck explanations as hypotheses, not proven causes.

Executive answer: make the process and denominator visible

Start with one process and one decision you can influence. Agree the metric definition, establish a credible baseline and identify where evidence is missing. Then compare a small set of improvement hypotheses by measurable impact, evidence strength and effort. Choose a bounded test with an owner and review date. A dashboard, a ranked list of problems or a persuasive benefit estimate does not by itself establish why a process underperforms or whether an intervention will work.

ASQ describes flowcharts as process representations and Pareto charts as a way to organise categories by their frequency or magnitude [A1, A2]. We use those ideas as starting tools, not causal tests. The method, worksheet and order-fulfilment example here are original BespokeData applications based on desk research. They are not findings about an actual customer, an independently reviewed study or a promise of savings.

1. Map the work before choosing a measure

Draw the process from request to completion, including queues, decisions, hand-offs and rework. For orders, a useful first map might be received, validated, allocated, picked, packed, dispatched and delivered. Confirm the map with people doing the work, not only the process document. Record which steps happen in different systems and which timestamps are captured manually. Where an event is not recorded, label the gap rather than drawing an apparently precise timeline.

Choose the process boundary explicitly. Order-to-dispatch and order-to-delivery are different lead times. Define an order, an order line and a shipment so split deliveries cannot quietly double the population. Identify who owns the process and who owns each input. If the question concerns waiting between teams, an aggregate daily total may hide the relevant hand-off. Request only the authorised fields needed for the decision and preserve a traceable extraction date.

2. Define denominator, grain, timezone and exclusions

Write the metric as a reproducible rule: eligible population, event dates, comparison, aggregation and missing-data treatment. For on-time dispatch, decide whether eligibility follows the promised dispatch date within the reporting month or orders received within it. Specify whether a revised promise replaces the original, and retain revision history if the distinction matters. Agree treatment of cancellations, partial shipments and orders not yet due. Different valid definitions answer different questions; do not combine them under one label.

Use an explicit business timezone and say whether elapsed time uses calendar or working hours. Keep source timestamps and conversion rules so daylight-saving changes do not produce unexplained shifts. Lead time is elapsed time across the agreed start and end; active cycle time is time spent doing the defined work. A system timestamp may show elapsed stage time including waiting, not hands-on effort. Do not claim labour savings from elapsed time alone.

Check whether the records are fit for the proposed measure before interpreting it [Q1]. Report unclassifiable records separately. A baseline should include its period, population size and known operational context, such as promotions or system changes. Retain the metric version so the next month's result is not compared with a silently revised definition. If you change the rule, recalculate both periods where possible and show the change.

3. Worked hypothetical fulfilment baseline

Assume a hypothetical month has 500 distinct eligible orders with an original promised dispatch date in the period. All 500 have usable promise and dispatch timestamps; cancellations are excluded before forming this population. Under the agreed local-time rule, 400 were dispatched on or before the promise and 100 were late. On-time dispatch is 400 ÷ 500 × 100 = 80%. This says nothing about delivery punctuality, profitability or the quality of customer communication.

Suppose separately that the total order-to-dispatch elapsed time across these 500 orders is 36,000 hours. Mean lead time is 36,000 ÷ 500 = 72 hours. A recorded active pick-and-pack total of 3,000 hours gives mean active cycle time of 6 hours. The 66-hour difference is not automatically removable waste: it may include overnight periods, intentional scheduling and unavoidable waits. Examine the distribution and stage records; averages alone can conceal long-tail exceptions.

For the 100 late orders, a reviewer records one provisional primary category per order: awaiting stock 50, approval queue 30, address clarification 15 and other 5. Those counts sum to 100, giving shares of 50%, 30%, 15% and 5%; cumulative shares are 50%, 80%, 95% and 100%. The top two categories contain 80 recorded cases. That is a prioritisation clue, not evidence that fixing those categories will remove 80 delays.

4. Turn bottleneck observations into testable hypotheses

A recorded category may describe where an order was noticed, not the reason it became late. Stock shortages might reflect purchasing decisions, inaccurate availability records or an unusual demand mix. An approval queue may contain orders already delayed upstream. Sample the event history, compare category definitions and ask the process owner what alternative explanations fit. Record contradictory evidence alongside supporting examples rather than presenting one convenient story as established cause.

Express each opportunity as an intervention and a measurable expectation. For example: routing a defined low-risk order class to a daily approval window may shorten approval waiting without increasing rework. Specify eligibility, expected effect, counter-metric and evidence needed. Counter-metrics might include errors, complaints or staff workload. Avoid selecting a pilot simply because its metric is easy to improve through exclusions or revised promises. Keep the original customer commitment visible where it is relevant.

5. Prioritise impact, evidence and effort separately

Create a short opportunity register rather than an opaque universal score. For each candidate, record the affected volume, plausible impact range, evidence strength, required effort, dependency and owner. Use concrete descriptions: observed timestamps support a waiting-time hypothesis; interviews alone suggest another. An intervention with a large theoretical benefit but unknown feasibility should not automatically outrank a small, well-supported test. Identify hard constraints such as safety, authority and capacity before ranking preferences.

In the example, the approval category has 30 late orders. If a bounded test prevented ten of those delays while nothing else changed, the scenario would have 410 on-time orders out of 500: 82%, or a two-percentage-point improvement from 80%. That is a scenario, not a forecast or promised ROI. It assumes the ten cases are preventable and no new delays are introduced. A pilot must establish whether those assumptions hold.

6. Separate a benefit scenario from realised value

Suppose an illustrative intervention saves ten minutes of manual chasing on each of 30 orders. That is 300 minutes, or five hours, for the hypothetical month. Applying an explicitly assumed AUD 40 per hour gives AUD 200 of gross time value. It is not cash saved, net benefit or return on investment. Staff may use the freed time for other work, and implementation, review and maintenance consume time too. State every assumption before comparing the scenario with costs.

Avoid counting the same benefit twice, such as valuing reduced lead time and the same staff minutes as separate labour savings. Distinguish customer outcomes, capacity, cost and risk rather than converting everything into a confident money total. Where the evidence is weak, retain a range or an unquantified benefit. A modest, auditable estimate is more useful for a decision than a large number whose denominator and mechanism cannot be explained.

7. Test, review and assign the next action

Agree the pilot population, period, operating owner and stop conditions before starting. Preserve the baseline definition and record other changes during the test. A before-and-after improvement can reflect seasonality, staffing, order mix or a system repair rather than the intervention. Where feasible, compare a similar unchanged group and inspect whether the groups were genuinely comparable. Without that evidence, describe an association and keep the causal limitation visible.

Use a short review cadence tied to the decision, not a reporting ritual. The operations owner reviews exceptions; the data owner checks definition and extraction issues; the decision sponsor chooses continue, revise or stop. Record each action, due date and read-back result. If a test fails, keep the learning and close or change the opportunity. Repeating an unsupported explanation in a monthly report does not make it stronger evidence.

Worksheet, limitations and next steps

The CSV includes metric definitions and opportunity rows with hypothetical values, owner placeholders and next-review fields. Replace the examples with authorised evidence, keeping timezone, grain and denominator in the definition. This guide is desk research and methodology, not a study of your operations or an independent expert review. Start by mapping one hand-off and defining one metric. A scoped commissioned assessment can turn a scattered evidence base into a written register, findings and priorities without guaranteeing an outcome.

record_typemetric_or_opportunitydefinition_or_hypothesispopulation_period_timezone_grainbaseline_or_scenarioevidence_and_limitowneractionreview_date
MetricOn-time dispatchDispatched by original promise; cancellations excludedHypothetical month; 500 distinct orders; agreed business timezone400/500 = 80%All classifiable; not delivery punctualityOperations owner (example)Confirm extraction and promise historyTo agree
MetricMean lead timeOrder received to dispatch; elapsed hoursSame 500 orders; source timestamps retained36000/500 = 72 hoursMean hides distribution; not active labourData owner (example)Inspect stage waits and long-tail casesTo agree
OpportunityApproval window pilotTest whether low-risk daily routing reduces waitsDefined eligible subset; bounded pilot10 prevented delays would give 410/500 = 82%Scenario only; confounding and new errors possibleApproval owner (example)Agree comparison and rework counter-metricTo agree
Benefit scenarioReduced chasing30 orders x 10 minutesHypothetical month; AUD40/hour assumption5 hours; AUD200 gross time valueNot cash savings or ROI; costs excludedSponsor (example)Measure actual time and implementation costTo agree
—————————

Dated method & source register

Read and reviewed 18 Nov 2026. Desk research and original BespokeData recommendations and hypothetical examples; no primary study, independent expert review or Atora generation.

Apply this to your business

Agree a scope around your decision and authorised evidence.

Discuss this goal →
Read the fictional sample report →

Related Straight Fact articles