Short answer
Employee retention signals should start a bounded inquiry into a working condition or process, not a prediction about who will leave. Record what was observed, who was represented, the period and denominator, then look for counterexamples and competing explanations. A named reviewer can decide whether a limited change is worth trying and when to reconsider it.
The distinction matters. A resignation count tells an organisation that people left during a defined period. An employee account of an unclear handover may help identify a question worth checking. Neither fact, on its own, shows why a person will leave or proves that a local practice caused retention.
Start with an observation, not a label
Avoid labels such as “at-risk employee,” “best team,” or “protective practice.” They collapse an uncertain observation into a conclusion before the evidence has been reviewed. Instead, write a plain observation that another person could check:
During the April to June onboarding window, several participating store starters described not knowing who could answer a late-shift handover question.
That sentence is useful only when its scope is visible. The CIPD's turnover and retention factsheet distinguishes turnover measurement from the broader work of understanding and responding to retention. Keep those jobs separate: use a turnover measure for its stated period and denominator, and use qualitative accounts as context for a specific inquiry.
The observation-to-inquiry-to-action worksheet
Use this one-page record before treating repeated accounts as a pattern.
| Field | What to write | Why it matters |
|---|---|---|
| Observation | The wording, count or operational event that prompted review | Stops a general impression becoming a claim |
| Population and period | Who could have contributed, where, and the start and end dates | Makes the scope explicit |
| Numerator and denominator | For example, 9 of 28 participating starters described the issue; 28 of 40 eligible starters participated | Separates the finding from coverage |
| Source and missing voices | Conversation, check-in, service record or exit feedback; name important groups not represented | Limits over-reading one channel |
| Counterexamples | Accounts or records that do not fit the observation | Tests whether the explanation is too broad |
| Alternative explanations | At least two plausible explanations, including a measurement or timing issue | Avoids mistaking correlation for cause |
| Inquiry | The narrow question a reviewer can check | Turns a concern into a testable next step |
| Action, owner and date | One proportionate change, one accountable person, and a review date | Creates a decision trail rather than a vague promise |
This original worksheet follows basic evaluation discipline: define the question and evidence before interpreting results, then design review into delivery. HM Treasury's Magenta Book is a useful public reference for that approach. It does not validate a retention intervention; it explains why a transparent evidence trail and an appropriate evaluation design matter.
Worked example: a fictional store onboarding inquiry
This example is fictional. It does not show that the proposed action changes retention.
Observation. From 1 April to 30 June, four stores hired 40 starters into the same role. Twenty-eight took part in a structured onboarding conversation. Nine of those 28 described uncertainty about who could resolve a late-shift handover problem. The finding is 9/28 participating starters, while participation coverage is 28/40 eligible starters. It is not a rate for all starters and it is not a prediction about any person.
Counterexamples and alternatives. Seven participants said the handover was clear; the other twelve gave accounts that did not establish either position. Keep those twelve unclassified for this question, rather than counting them as satisfied. The reviewer also finds that two of the nine accounts came from the same busy weekend and that the written guide had been moved in one store. Plausible explanations include different local induction routines, the guide's location, unusually high workload on that weekend, or the way the question was understood. None is established yet.
Inquiry. The onboarding lead asks: “For starters on a late shift, can a named handover contact and the guide location be confirmed before the first independent shift?” The site lead checks this with a small next cohort and records whether the contact and guide were available. The HR partner reviews the record on 31 July with the onboarding lead, including accounts that do not fit the initial observation. The decision is whether to keep, amend or stop the local routine, not whether the nine people are likely to leave.
This form of review can be paired with regular manager conversations. Acas advises employers to keep and share a written record of performance discussions, while also using regular informal feedback and check-ins where appropriate. See Acas: reviews and appraisals. The retention worksheet is not a performance record and should not be repurposed as one.
Decide what evidence would change the action
Before an owner acts, agree what would make the team reconsider. Examples include:
- the next eligible cohort has materially different access to the handover contact or guide;
- the same uncertainty is raised across another shift or location;
- records show the reported obstacle was already resolved before the conversation took place;
- participants who had clear handovers describe a different routine that can be examined; or
- the proposed change creates a practical burden that outweighs its usefulness.
Do not convert these checks into an individual score. A qualitative account can be relevant evidence without being a measure of commitment, intent or future behaviour. Similarly, a change in a local account after an action does not prove the action caused a retention outcome. Record the result, its limits and the next human decision.
Where Lontra fits
Lontra can support a disciplined listening workflow when an organisation chooses to run employee conversations. Employee input can inform an action-oriented manager brief; managers do not receive raw conversations. A manager and the relevant people partner still decide what to check, whether an action is proportionate and how to follow up.
Where it is permitted for the organisation, Living Memory can retain dated context from prior campaigns so a later reviewer can see what was previously agreed. Ask Lontra can help an authorised user explore accessible campaign context and memory, but its output needs verification and does not make the decision. These capabilities do not establish a cause of turnover or a forecast of resignation.
For a view of the employee-conversation workflow and its boundaries, see Lontra's product overview. For broader action planning, the employee retention strategies guide and stay interview guide cover related but different reader tasks.
For a first conversation test, choose one work question. The platform trial covers one campaign with up to 30 invitations over 60 days, with no card required. Use the worksheet to define what the manager will review before sending invitations.
Sources
Frequently asked questions
What is an employee retention signal?
An employee retention signal is an observation that merits a bounded inquiry about a working condition or process. It is not a prediction that a person will resign.
What should a retention review record?
Record the source, population, period, denominator, counterexamples, alternative explanations, reviewer, proposed action and review date.
Can employee conversations prove why people leave?
No. Conversations add context to review alongside other evidence. They do not establish a cause or a forecast for an individual.
How can Lontra support a human retention review?
Lontra can support employee conversations and action-oriented manager briefs without giving managers raw conversations. People remain responsible for the decision and any follow-up.