Ask what the team needs you to produce
CIPD describes recruiting as a process that spans defining a role, attracting applicants, selection and appointment. Its people-analytics guidance describes using people data to investigate business questions. These are professional descriptions, not a promise that an intern will own hiring decisions or a predictive model.
A recruiting internship might center on scheduling, candidate communication, requisition records or process improvement. An analytical internship might center on data quality, reporting, workforce questions or evaluation. The actual posting should settle the assignment. An HR label alone does not tell you which work you will do.
Before applying, identify one deliverable you would like to learn: a reliable interview schedule, a clean stage-history table, a reporting definition or an analysis of a process change. This produces a more specific application than saying you are a people person or enjoy data.
Make each stage a defined event rather than a vague status
Build a fictional recruiting funnel with five stages: started application, completed application, interviewed, offered and accepted. For the exercise, each count represents distinct people in one completed cohort, and later stages are subsets of earlier stages. Real systems require checking those assumptions rather than inferring them from a dashboard.
A current status field cannot always reconstruct history. An accepted candidate may no longer be labelled interviewed, and a withdrawn candidate may still have completed an interview earlier. Store or request the event history needed for the metric. Keep cancellations and rescheduling separate from completed interviews.
Decide whether the unit is a person or an application. One person applying to three roles can be one individual and three applications. Neither unit is inherently wrong, but the denominator and question must agree.
Compare two cohorts with the same final hiring result
Cohort A has 100 application starts, 40 completions, ten interviews, five offers and four acceptances. Cohort B has 100 starts, 80 completions, ten interviews, five offers and four acceptances. Both have 80% offer acceptance and 4% starts-to-acceptance, yet the completion rate differs sharply.
| Measure | Cohort A | Cohort B |
|---|---|---|
| Application starts | 100 | 100 |
| Completed applications | 40 | 80 |
| Completed interviews | 10 | 10 |
| Offers | 5 | 5 |
| Accepted offers | 4 | 4 |
| Completion / starts | 40% | 80% |
| Interviews / completions | 25% | 12.5% |
| Accepted / offered | 80% | 80% |
A has 40% completion and B 80%. Among completed applications, the interview rates are 25% and 12.5%. These figures do not tell you which process is better. The additional completed applications might be relevant candidates, duplicates, a different role mix or something else. You need evidence about what changed and the purpose of the process.
The table makes the trade-off visible without turning an observed ratio into a verdict about candidate quality or recruiter performance.
Do not compare unfinished cohorts with finished ones
Suppose a new cohort has ten offers outstanding but none has yet reached its response deadline. Reporting zero acceptances divided by ten offers as a final acceptance rate would confuse pending outcomes with rejections. Show pending, accepted, declined and withdrawn separately as of a stated cutoff.
Use comparable observation windows. A cohort recruited last week has had less time to progress than one recruited two months ago. For a time-to-stage measure, define the starting event, ending event, calendar or business-day convention and treatment of records that have not reached the stage.
Averages also need a denominator. An average calculated only for hired applicants does not describe everyone who entered the process. State the included group and preserve unresolved outcomes instead of dropping them because they complicate the report.
Use the minimum data needed for the question
For a public work sample, use the fictional aggregate table. You do not need names, résumés, interview notes, compensation histories or demographic details to demonstrate the calculations. A real internship should define which data you can access, how they are handled and who approves the analysis.
CIPD’s people-analytics guidance distinguishes description from stronger analytical claims. In this exercise, the counts describe a process; they do not establish that a form change caused the different completion rates. A before-and-after comparison could also differ because of job mix, sourcing, timing or other factors.
Propose a bounded next step: validate duplicate records, inspect where applicants stopped, or ask the process owner whether a documented change occurred. Do not build an automated candidate ranking from this exercise or infer personal traits from sparse application activity.
Sources[2]
Show the process improvement you can actually explain
Deliver a stage dictionary, a cohort definition, the computed table and a short question memo. Include the checks that stage counts reconcile under the exercise’s assumptions. If a real system violates the subset relationship, investigate rather than forcing the numbers into a funnel shape.
For a recruiting application, emphasize accurate coordination and communication: what changed, who needed to know and how you prevented conflicting versions. For an analytics application, explain the denominator, the cutoff and why pending records remain visible.
Ask the employer who uses the intern’s output and how it is reviewed. A small analysis with a clear decision user is more valuable than an impressive dashboard whose measures no one can define.