Frame the workforce decision and its evidence · Lesson 1

From business question to testable decision

13 min read

Lesson objectives

  • Translate a service-level threat into a decision statement, horizon, alternatives, and a falsifiable success rule.
Reading tools

Imagine a technology service company saying, 'We need more analysts.' That is a proposed remedy, not yet a workforce-planning problem. Delays might reflect too few people, knowledge concentrated in one shift, a change in the mix of requests, or a slow approval step. Start with a decision question: which business result is current capacity failing to deliver, over what horizon, and which alternatives can the accountable leader actually choose? CIPD frames workforce planning as balancing the supply of people and skills against demand. A headcount snapshot therefore cannot settle the matter. People analytics brings people data and business context to a problem. The two practices meet when an analysis informs an executable choice, instead of producing a dashboard and then looking for a use for it.

Make the vague request specific. The outcome is the share of complex enterprise customer requests resolved within two days; the population is enterprise requests, not all tickets. The horizon is six months following a product launch. Additional spending must stay below 480,000 currency units, and customer satisfaction must not fall below its current level. Plausible alternatives are hiring analysts, training current support staff, redistributing shifts, and changing triage. None is privileged before demand and capacity are understood. The operations director owns the operating decision with HR and finance; a people analyst supplies evidence and a recommendation but does not alone authorize hiring spend. A decision statement might read: 'Which mix of options raises two-day resolution from 76% to 88% by the end of the next two quarters, within the spending cap and without a material decline in customer satisfaction?' A known measurement can confirm or refute this result. 'Improve customer experience' and 'hire ten people' cannot serve the same role: one is too vague and the other presupposes the answer.

Fix four boundaries before requesting data. First, the outcome's unit is a completed request, not a person or an hour. Employee-level records may later explain capacity, but their denominators cannot be mixed with ticket counts. Second, state the baseline period, population definition, and volume behind 76%; a small, unstable sample can make an apparent improvement meaningless. Third, specify a counterfactual: what happens to service if demand grows and the company does not change its workforce plan? Comparing the proposed plan with itself says nothing about its value. Fourth, guard against harm. If teams close tickets without resolving them, the speed metric improves while service worsens. Pair it with reopened tickets, customer satisfaction, and workload. Descriptive metrics are a start, but explaining delay requires later examination of request mix, skills, and process. An association between analyst count and speed does not by itself establish that hiring caused faster service.

A decision question is different from a subordinate analysis question. The former asks, 'Which option should we approve?' The latter might ask, 'How many requests did each team complete on its shift?' Several analyses may feed one decision, but collecting them without an order consumes time and produces numbers that never alter the choice. Order investigations by the value of information. If a change in request severity could reverse the ranking of training and hiring, inspect severity first. If both options fail even under optimistic assumptions, the director does not need a sophisticated forecast to know that a third option is required. For every proposed data request, state what answer would change the recommendation. This makes the inquiry bounded and lets the team stop when more detail is unlikely to alter the decision before its deadline.

In the brief, separate observations, assumptions, and unknowns. The observed baseline count is only sound after confirming the definition of 'complex' and the completeness of ticket records. The assumption is that volume will reach 1,300 and the mix will be stable; that belongs in a scenario, not in a box marked fact. The unknown is how triage or training will affect capacity and quality. Suppose the product director expects 1,300 post-launch requests, while sales expects 1,600. The analyst must not pick the figure that makes the budget look comfortable. Record the source and date of each estimate and show what changes under the higher scenario. The recommendation remains provisional until productivity by request type and the capacity alternatives have been assessed. Communicating uncertainty in this form is part of professional judgment, not an apology for the analysis.

Is the information worth delaying the decision?

An urgent message arrives from the service manager. She wants to sign a six-person contract this evening, while analysts expect case-complexity data in two days. Saying that data are incomplete leaves the choice unresolved. Instead, name the information that could change the option and estimate what waiting would cost. A sample might show that triage is the main bottleneck, making a large contract unnecessary. Alternatively, it could reveal a need for expertise unavailable internally. Information has value when it distinguishes those routes, rather than merely adding report columns. Prepare a temporary recommendation the manager can implement today, with a review date and the evidence that would change it.

Distinguish the decision deadline from the date its effect begins. A contract may need approval this week although staff start next month. That intervening month could support record checks or a change to the initial scope. Sketch a timeline showing the last date at which the commitment can still change. Waiting costs more if that date is today. Where an adjustment window remains, a small initial commitment followed by expansion after checking may be feasible. This turns uncertainty into decision design instead of an indefinite reason to hesitate. Ask the authorized owner to confirm cancellation terms; the analyst explains how those terms affect the comparison.

Suppose a highly reliable check costs four thousand and waiting for it adds three thousand in expected operating cost. Treating it temporarily as perfect information leaves a three-thousand net benefit: ten thousand minus four thousand minus three thousand. Calling a test highly reliable does not justify assuming perfection. Request its likely errors and their consequences for the contracting choice. If expected error costs exceed three thousand, waiting loses its advantage here. A mandatory service obligation could also rule out waiting despite an attractive average cost. The calculation gives the investigator a concrete question: can this sample distinguish the states well enough to justify delay?

Test the committee's probability assumption as well. If complex demand has only a one-quarter probability in the Jeddah example, expected triage cost is thirty-five thousand. At a three-quarter probability it rises to sixty-five thousand. Full contracting remains at sixty thousand in either case. Disagreement about probability can therefore reverse the recommendation and deserves an explicit record. The crossover follows from the triage costs: twenty thousand plus sixty thousand times the probability of the complex state. This equals sixty thousand at roughly two thirds. Show the crossover with the evidence behind the probability estimate, then ask operations how plausible crossing that point is.

Keep a compact snapshot of assumptions known at the decision date, naming each estimate's owner and period of validity. When the team returns a month later, assess the decision against available information before examining the realized outcome. A well-reasoned option can fail because a low-probability state occurs; a weak choice can succeed by chance. Separating decision quality from outcome supports better learning. Tomorrow, choose one actual analysis request and write the answer that would make you act differently. If no changed action can be named, revisit the investigation's purpose before adding more data to the report.

Lesson summary

Imagine a technology service company saying, 'We need more analysts.' That is a proposed remedy, not yet a workforce-planning problem.