Designing a 5‑Signal Retention Score: A Worked Example for UK SMEs

(What this article actually does)

We focus on one thing only: how to design a simple, 5‑signal retention score for a 30–80 person UK SME using data you already hold.
You get a worked example: one illustrative employee, their five signals over three months, exact points, and how the score changed after action.
You leave with a practical scorecard template and thresholds you can adapt, not a full HR automation blueprint (we already covered that in our HR automation 2026 guide). If you want to quickly sanity‑check the time impact, you can plug your own HR workflows into our free Workflow Time Savings Calculator.

Most writing on “AI employee retention” jumps from buzzwords to vague advice. What UK SME owners and People leads usually need is narrower: how do I turn a handful of HR data points into a fair, explainable risk score that warns us before people leave?

This is the only article on our site that answers that exact question.

We ignore the wider HR automation picture (onboarding, workflows, chatbots) – that sits elsewhere. Here we stay at one level: pick 5–7 signals, define thresholds, assign points, and track how one employee’s score moves over time.

The 5‑Signal Retention Score: What it is and what it is not

A 5‑signal retention score for a UK SME is a deliberately simple scoring model that turns familiar HR data into a 0–10 risk number per person. It is explainable, lightweight, and built for teams without a data science function.

Definition: Retention score — a single number (for example 0–10) that summarises several HR indicators linked to staff turnover risk, used as a triage tool for early conversations, not as proof that someone will leave.

We design this score with three hard constraints for 20–100 person UK SMEs:

No surveillance data. Only planned HR data: pay, holiday, absence, workload, 1:1s, role changes.
Human‑explainable. A manager must be able to tell an employee, in plain English, why their score is what it is.
Low maintenance. Update monthly from exports (Xero Payroll, Sage, BambooHR, Hibob, etc.), not a full‑time analytics team.

If you want the full people‑ops automation picture – from recruitment to exit – that is in our AI HR automation blueprint. Here we assume you have already decided you want a people‑risk early warning and now need to design the actual score.

Definition: People risk early warning — a light‑touch way of surfacing roles or individuals where losing them in the next few months would create outsized pain, based on objective patterns in HR data rather than rumours or gut feel.

Which 5 signals should a 50‑person UK SME start with?

For most 40–60 person firms we work with, five core signals give you most of the predictive value without drifting into creepiness or complexity. They are all available in typical SME HR stacks (payroll + HRIS + basic scheduling).

Definition: HR data signals (retention) — specific, measurable values in your HR data (for example, overtime trend, unused holiday, time since pay review) that tend to move before a resignation.

A strong, non‑creepy starting set is:

Overtime trend (workload)
- Measure: average weekly hours over contract in the last 8 weeks.
- Source: timesheets, calendar analytics, or simple “hours worked” export where available.
- Why it matters: sustained overwork is a classic burnout precursor.
Unused holiday (rest)
- Measure: % of annual leave allowance still unused relative to where you are in the leave year.
- Source: HRIS/holiday tracker.
- Why it matters: people not taking leave often carry unsustainable load or feel unable to disconnect.
Manager contact (1:1 cadence)
- Measure: weeks since last proper 1:1 logged in your HR/people system.
- Source: HRIS notes, simple 1:1 tracking sheet.
- Why it matters: weak manager relationship is one of the most consistent reasons for leaving.
Time since last pay review (reward)
- Measure: months since last salary review or promotion.
- Source: payroll/HRIS.
- Why it matters: in London and the South East, static pay against rising costs is a major push factor.
Short‑term sickness pattern (wellbeing)
- Measure: number of 1–2 day sickness instances in the last 12 weeks.
- Source: absence records.
- Why it matters: a run of short sickness spells can signal stress or disengagement.

Definition: AI employee retention (UK SME) — the use of simple algorithms or models on existing HR data in a small or mid‑sized UK business to prioritise where to focus retention efforts, not to make automated firing decisions.

These five cut across workload, rest, relationship, reward, and wellbeing – which is the balance we look for in our own projects. You can add more later (for example, recent manager change or declined training), but five keeps the model practical and explainable.

The SIMARA 5‑signal scorecard template (points and thresholds)

To move from “signals” to a consistent score, you need concrete thresholds and points. Below is the 5‑signal template we commonly start with in a 40–60 person London SME.

Definition: Scoring model (HR) — a structured way of assigning points to different data ranges (for example, overtime hours) so that similar patterns receive similar scores across the business.

Each signal is scored 0, 1, or 2 points. Total score = sum of all five (0–10 range).

Signal	0 points (low concern)	1 point (watch)	2 points (red flag)
Overtime trend (last 8 weeks)	≤ 3 hours/week above contract	3–7 hours/week above contract	≥ 8 hours/week above contract
Unused holiday (pro‑rated in leave year)	Within 10% of where they should be	10–25% behind expected usage	>25% behind expected usage
Weeks since last 1:1 with manager	≤ 3 weeks	4–6 weeks	≥ 7 weeks
Months since last pay review	Within normal cycle (for example ≤ 14 months)	3–6 months beyond normal cycle	>6 months beyond normal cycle
Short‑term sickness instances (last 12 weeks)	0–1 instances	2 instances	≥ 3 instances

Once scored, we typically use this banding:

0–3 points → Low risk (no extra action, trust normal management).
4–6 points → Medium risk (manager to review and discuss in next scheduled 1:1).
7–10 points → High risk (manager + HR to discuss within 2 weeks and agree actions).

This is not a statistical model. It is a transparent rule‑set you can pilot in Excel or Google Sheets before you even think about automation. When we later add an AI layer, we mainly use it to refine these thresholds and weights, not replace them with a black box.

Calculate this for your business: If you are asking “How many hours would automating this actually give back?”, plug your existing HR admin steps into the free Workflow Time Savings Calculator. It converts a repetitive workflow into estimated hours and pounds saved per year so you can see whether a retention radar is worth building now or later.

A worked example: one employee’s score over three months

To make this concrete, here is an illustrative scenario. This is not a real person, but it reflects patterns we see repeatedly in 30–80 person UK SMEs.

Illustrative scenario:
A 45‑person digital agency in London uses the 5‑signal scorecard. Jamie is a mid‑level account manager on a 37.5‑hour contract.

Month 1 – subtle early drift

In Month 1, the data for Jamie looks like this:

Overtime trend: Averaging 42.5 hours/week (5 hours above contract) over the last 8 weeks.
→ Score: 1 point (3–7 hours/week above contract)
Unused holiday: It is 6 months into their leave year; typical expected usage is around 50%. Jamie has used 40%.
→ Score: 1 point (10–25% behind expected usage)
Weeks since last 1:1: 4 weeks.
→ Score: 1 point (4–6 weeks)
Months since last pay review: 13 months. The company’s “normal” cycle is every 12–14 months.
→ Score: 0 points (within normal cycle)
Short‑term sickness (last 12 weeks): 1 single‑day absence.
→ Score: 0 points (0–1 instances)

Total Month‑1 score for Jamie: 1 + 1 + 1 + 0 + 0 = 3/10 (low band).
Jamie does not hit any escalation threshold – but is sitting at the top of “low”. The radar records this but does not trigger a special intervention.

Month 2 – score crosses into medium risk

By Month 2, workloads have stayed high and some management routines have slipped:

Overtime trend: Now averaging 46 hours/week (8.5 hours above contract).
→ Score: 2 points (≥ 8 hours/week above)
Unused holiday: 8 months into the leave year; expected usage ~65%. Jamie has used 45%.
→ Score: 2 points (>25% behind expected usage)
Weeks since last 1:1: 7 weeks.
→ Score: 2 points (≥ 7 weeks)
Months since last pay review: 14 months (still just within cycle).
→ Score: 0 points
Short‑term sickness (last 12 weeks): Now 2 single‑day absences.
→ Score: 1 point (2 instances)

Total Month‑2 score for Jamie: 2 + 2 + 2 + 0 + 1 = 7/10 (high band).

Under the banding above, this would trigger:

An alert to Jamie’s manager and HR.
A requirement for a joint check‑in within 2 weeks.

The key is what the score means in practice. Not “Jamie is leaving”, but: “Jamie shows sustained overwork, skipped leave, weak recent contact with their manager, and a bump in short sickness. This pattern deserves attention.”

Month 3 – after targeted intervention

Assume the manager and HR follow a simple playbook:

Workload: Reallocate one large client account to another manager, bringing Jamie closer to contracted hours.
Holiday: Encourage and approve a solid week of leave within the next month.
1:1s: Reinstate fortnightly 1:1s and log them properly.
Reward: Book in a formal development and pay review conversation for the following quarter.

One month later, the data looks like this:

Overtime trend: Averaging 40.5 hours/week (3 hours above contract).
→ Score: 1 point (3–7 hours/week above)
Unused holiday: After a week off, usage now sits ~60% vs an expected ~70% – still slightly behind, but not extreme.
→ Score: 1 point (10–25% behind expected)
Weeks since last 1:1: 2 weeks (since the new cadence started).
→ Score: 0 points (≤ 3 weeks)
Months since last pay review: 15 months (now 1 month beyond the upper end of the normal 14‑month cycle, but a review is scheduled).
→ Score: 1 point (3–6 months beyond normal cycle)
Short‑term sickness (last 12 weeks): No new sickness; 2 instances still within the window.
→ Score: 1 point

Total Month‑3 score for Jamie: 1 + 1 + 0 + 1 + 1 = 4/10 (medium band).

Jamie’s situation is far from perfect, but the score has shifted from 7 to 4. In practice, that means:

The crisis‑risk has reduced.
Jamie still appears on the radar, but the weekly “urgent” list shrinks.
Management attention can be shared more fairly across the team.

This is the essence of a retention radar for SMEs: a low‑friction way to see where targeted changes move scores in the right direction over a few weeks, long before someone submits a resignation letter.

How we calibrate signals: what usually predicts exits first?

The 5‑signal template above is our default starting point. In practice, we refine it based on real patterns in your own data using a light version of the approach we use in our broader AI Readiness Scorecard and ROI work.

In several SME pilots (40–70 staff, London and South East), we have seen the following relative pattern when reviewing 12–18 months of leavers vs stayers:

Sustained overtime and overdue 1:1s often move months before someone leaves.
Short‑term sickness patterns sometimes spike closer to the exit date.
Time‑since‑pay‑review and unused holiday vary more by culture and sector.

We translate that into a simple signal‑tuning process:

Export 12–18 months of history from your HR/payroll system with the five signals per month, plus a flag for “left within next 90 days”.
Compare average scores on each signal for leavers vs stayers (for example, did leavers average 1.7 points on overtime vs 0.6 for stayers?).
Adjust weights where there is a clear gap. If overtime and missed 1:1s show the largest differences, you might weight them more heavily.

Definition: Signal weighting — increasing or decreasing the number of points a particular indicator can contribute to the total score when it shows strong correlation with the outcome you care about.

Here is a practical view we often end up with after calibration for a 50‑person London SME:

Signal	Base max points	After calibration	When we up‑weight it
Overtime trend	2	3	When leavers show consistently higher overtime scores than stayers
Unused holiday	2	2	When culture normalises rolling leave into next year
Weeks since last 1:1	2	3	When leavers almost always have longer gaps than stayers
Months since last pay review	2	2	When pay cycles are already well‑managed
Short‑term sickness	2	2	When sickness patterns are noisy or seasonal

In this adjusted model, the total score range becomes 0–12 instead of 0–10, and the banding shifts slightly:

0–4 → Low,
5–8 → Medium,
9–12 → High.

Even with these tweaks, the system stays transparent. A manager can still see that a high score is mostly overtime + missed 1:1s, not some opaque “AI insight”.

For many SMEs that is as far as you need to go. Only when you have a couple of years of decent data and 70–100+ staff does it become worth testing a lightweight machine‑learning model against this tuned scorecard, as we outline in our AI ROI framework for deciding when extra complexity pays off.

Trade‑offs, risks and limitations of a 5‑signal model

A 5‑signal scorecard is a deliberate compromise. It trades some predictive power for clarity, speed, and cultural safety. That comes with specific limitations SMEs should understand before rolling it out.

First, it will generate false positives and false negatives. Some people will score high and never leave; others will resign with low scores because of external life events or a competing offer you could not see. The score is a triage tool, not a crystal ball.

Second, there is a risk of self‑fulfilling labels if managers treat scores as verdicts. If someone is labelled “high risk” and then sidelined rather than supported, the system can actually increase churn. That is why, in parallel posts such as our guide to where AI should and should not answer HR questions, we stress clear human boundaries.

Third, the model can bake in past inequities. If, historically, certain demographic groups have had less access to progression or carried more overtime, a naive score will simply reflect that. You need HR oversight to check whether particular groups are over‑represented in high scores and, if so, whether that points to structural issues rather than individual risk.

Finally, there is maintenance overhead. Someone needs to:

sanity‑check monthly exports;
review any obvious data errors (for example, mis‑logged holiday);
update thresholds annually as your policies or working patterns change.

This is why we design these systems so that a People lead or Operations director can maintain them in 2–3 hours per month, not a full analytics team.

When this approach backfires (and when not to use it)

There are clear situations where a 5‑signal retention score does more harm than good, or is simply not worth the effort.

You should probably not build this – yet – if:

You are under ~20 people. In a very small team, direct weekly conversations usually surface issues faster than any model. At that scale, invest in management hygiene rather than analytics.
Your HR data is messy or scattered. If holiday records live in email, 1:1s are never logged, and overtime is guesswork, your first project should be cleaning up the basics, not scoring flawed inputs.
You cannot commit to action. If HR or managers are too stretched to respond to high scores, the radar will simply highlight problems you are not ready to fix – which erodes trust.
Your culture is deeply sceptical of data. If previous attempts at measurement have been weaponised, you may need a period of rebuilding confidence before introducing any kind of individual‑level scoring.

There are also edges to where this model makes sense:

In highly seasonal businesses (for example, e‑commerce around Q4), overtime signals need to be compared to seasonal norms, not flat thresholds.
In very flexible environments where time is not tracked at all, you may swap the overtime signal for another, such as “number of concurrent projects” or “client load index”.

If any of these apply, a lighter approach – such as team‑level dashboards on holiday usage and 1:1 cadence only – may be safer as a first step. We talk more about sequencing automation and avoiding over‑reach in our workflow automation field guide.

What to explore next

If you want to turn this 5‑signal model into something live in your business:

See how we look at automation economics in our AI ROI analysis framework – useful for deciding how much to invest beyond an Excel pilot.
Compare building this in‑house vs wider HR automation in our guide to fixing People Ops bottlenecks.
For the end‑to‑end HR automation picture (hiring, onboarding, retention, exit), read the 2026 AI HR blueprint.

Ready to go deeper:

Sources & further reading

Information Commissioner's Office (ICO) – Employment practices and data protection – Guidance on using employee data under UK GDPR.
FSB – UK Small Business Statistics – Overview of the UK SME landscape and context for people decisions.
CIPD – People Analytics Factsheet – Introductory guidance on using HR analytics responsibly.
ACAS – Discipline and Grievances at Work – Context on fair treatment and consultation that should inform any use of employee data.

Be explicit that the score is there to offer support earlier, not to police performance. Share the five signals openly, explain the thresholds in plain English, and give examples of the kinds of interventions you will take (redistributing workload, prioritising leave, fixing 1:1s) rather than consequences. Involve employee reps or a small pilot group in testing the score before you scale it. Transparency, clear boundaries, and visible positive actions are what keep this from feeling like surveillance.

Do we need machine learning to predict staff turnover in a 50‑person SME?

In most 30–80 person SMEs, a transparent 5‑signal scorecard gets you most of the benefit without the overhead of a machine‑learning model. You only start to see a strong case for more advanced modelling once you have a couple of years of clean data, 70–100+ employees, and a clear owner for people analytics. Until then, you are better off refining your signals, tightening data quality, and making sure managers actually act on the early warnings. If you are unsure whether the extra investment pays off, use the free Workflow Time Savings Calculator to compare the hours and cost of building a basic scorecard versus a more complex AI project.

How often should we recalculate and review the retention scores?

Monthly works well for most UK SMEs. It balances freshness with practicality: HR can run exports, update the scorecard, and then managers review high and medium‑risk cases in their regular cycles. Weekly scoring is usually unnecessary noise; quarterly is too slow to catch issues early. The key is to pair each monthly refresh with a short, structured review so scores consistently lead to action.

What HR tools do we need in place before building this?

You do not need a full HR analytics platform. You need three basics: a reliable holiday and absence tracker, a way to log 1:1s (even a simple shared spreadsheet), and payroll or HRIS data that tells you contract hours and last pay review date. Most SMEs running on tools like Xero Payroll plus a lightweight HR system (BambooHR, CharlieHR, Hibob, etc.) already have enough data. The scorecard can run in Excel or Google Sheets initially, with automation added later once you have proven its value.

How many hours would automating this actually give back?

For a 40–60 person SME, building and maintaining the score manually might take an HR lead half a day each month, plus scattered manager time spent hunting through systems. A basic automation (scheduled data pulls and a simple dashboard) can cut that to well under an hour of HR time and give managers a single, consistent view. To put real numbers against this for your own set‑up, feed your current HR admin steps into the Workflow Time Savings Calculator. It will translate those repetitive steps into an estimate of hours and £ saved per year so you can decide if automation is justified now.

Ready to design a retention score that fits your culture as well as your numbers? Contact SIMARA AI and we will help you calibrate a people‑first, GDPR‑aligned model for your SME.