People Analytics ROI: A Practical Finance Worksheet

People analytics ROI compares the financial benefit attributable to an analytics program with its full cost over a defined period. The arithmetic is simple; attribution is not. Establish the baseline, name the decision and intervention, include internal work, show benefit ranges, test other explanations, and keep unmonetized outcomes visible.

Use the formula only after defining the case

The common formula is:

ROI (%) = (attributed financial benefit - fully loaded program cost) / fully loaded program cost x 100

Write these definitions beside the calculation:

• Program: the tool, data work, operating process, and decisions being evaluated.
• Period: the implementation and measurement window.
• Baseline: what is expected if the current approach continues.
• Intervention: the action taken because of the analysis.
• Financial benefit: a measured change converted with an approved unit value.
• Attributed benefit: the defensible share of that benefit linked to the program.
• Fully loaded cost: external spending plus internal time and associated change work.

HM Treasury's Green Book 2026 separates appraisal before a decision from evaluation after implementation. It also emphasizes business as usual, options, assumptions, risk, uncertainty, and early evaluation planning. A company does not need to copy a government method, but those disciplines prevent an ROI percentage from becoming a sales claim with no counterfactual.

Step 1: Name the decision and mechanism

Start with one sentence:

The program will help regional operations and HR identify recurring schedule-notice problems, choose a specific process change, and review whether the affected sites improve against comparable sites and their own prior periods.

Then map the expected chain:

If the chain jumps from "dashboard launched" to "turnover fell," the business case is missing the work that could plausibly connect them.

Step 2: Build the full cost table

Use finance-approved rates and label cash cost separately from allocated internal time.

Avoid counting salaries twice. If a fully loaded analyst rate already includes overhead, do not add the same overhead as another line. Record whether internal time displaces other work or uses spare capacity.

Step 3: Define benefit units before observing results

Choose measures the business already understands. Examples include:

Do not monetize every positive outcome. Better source traceability, broader coverage, faster decisions, or clearer employee communication may justify a decision without a reliable cash value. Keep these as measured non-financial benefits until finance accepts a conversion method.

The UK government's Digital and Data Benefits Framework is designed to help teams identify, quantify, plan, and track benefits from digital and data work. Its structure is useful here: define the benefit, owner, baseline, measure, timing, dependencies, and evidence rather than attaching a percentage after launch.

Step 4: Choose an attribution design

Use the strongest practical design available and state its limits.

1. Randomized or phased assignment: appropriate units receive the intervention at different times under a pre-agreed design.
2. Matched comparison: compare similar teams or locations while documenting material differences.
3. Interrupted time series: examine a long pre- and post-intervention trend and other events near the change.
4. Difference in differences: compare change over time between intervention and comparison groups when assumptions are credible.
5. Contribution analysis: document the mechanism, implementation evidence, other explanations, and the program's plausible contribution.
6. Before and after only: useful for observation, weak for causal attribution.

The Magenta Book analytical methods describes experimental, quasi-experimental, theory-based, and economic approaches. Select the design before results are known when possible. Have a qualified analyst review causal claims.

Step 5: Calculate a range, not a single future

Create low, base, and high cases for uncertain inputs:

• adoption and eligible-population coverage;
• implementation delay;
• benefit unit value;
• observed outcome change;
• attributed share;
• recurring internal effort;
• duration of the benefit;
• one-off and exit costs.

Show the assumptions that move the result most. A sensitivity table is more useful than a decimal-heavy answer that hides uncertainty.

Work a fictional example

The figures below demonstrate the worksheet and are not Lontra results or market benchmarks.

A services company evaluates a people analytics program over 12 months. Finance approves a fully loaded cost of £48,000: £22,000 for software and implementation, £14,000 for data, governance, and evaluation, and £12,000 for administrator, analyst, and manager time.

The target group records six fewer regretted departures than its agreed baseline. Finance values the replacement and vacancy cost at £18,000 per departure, so the observed gross difference is £108,000. A review finds that pay changes, local hiring conditions, and a manager reorganization also plausibly affected the result. The team therefore tests three attributed shares rather than calling the entire change ROI.

The program also removes 300 hours from a recurring preparation task. At an approved loaded rate of £45 per hour, the maximum time value is £13,500. Managers report that about half of the time is reassigned to other necessary work in one scenario, so its economic capacity value is £6,750. This is not a cash saving or demonstrated margin unless spending actually falls or the released capacity produces a separately measured financial result.

The 25%, 50%, and 75% attributed shares below are arbitrary teaching scenarios, not attribution estimates or confidence bounds. No attributable effect, meaning 0%, remains possible and should stay in the decision range unless the evaluation provides evidence to narrow it.

The table does not prove which case is true. It shows decision makers what must be believed for the investment to pay back. If the program cost excludes ongoing review or if the baseline is unstable, revise the model.

Keep a benefits register

For every claimed benefit, record:

• name and precise definition;
• baseline period and population;
• owner and data source;
• expected timing and duration;
• unit value and approving finance contact;
• intervention and decision record;
• attribution method;
• other plausible explanations;
• low, base, and high assumptions;
• actual result and review date;
• double-counting links to other benefits;
• decision to continue, change, or stop.

Update the register when the operating process changes. A forecast is not an achieved benefit, and an achieved operational change is not automatically attributable financial value.

Present the case so a CFO can challenge it

Lead with the decision, current cost, and range. Then show:

1. business-as-usual baseline;
2. options considered, including a smaller process change;
3. full cost by year and owner;
4. benefit mechanism and time to realization;
5. low, base, and high outcomes;
6. attribution design and alternative explanations;
7. non-financial benefits and risks;
8. implementation, evaluation, and stop conditions.

The U.S. Office of Personnel Management's Data Strategy emphasizes high-quality data products, standards, governance, advanced analytics, and evidence-based decisions. A credible ROI case funds those operating capabilities, not just a dashboard license.

Where Lontra may fit

Lontra can be evaluated when an ROI hypothesis depends on gathering and reviewing employee context. Review the current platform capabilities, then use one focused campaign, up to 30 invitations over 60 days with no credit card, to test the evidence path and measure the actual internal work required.

Use the result to populate costs, coverage, review effort, and decision records. Do not treat trial participation as financial return. Verify required exports or integrations separately, and do not infer individual performance, intent, or future behavior from conversation data.

Sources

• HM Treasury: The Green Book 2026
• GOV.UK: Digital and Data Benefits Framework
• HM Treasury: Magenta Book analytical methods
• U.S. Office of Personnel Management: Data Strategy 2023–2026