Small business grants measurement: baselines, costs and honest limits

Good grant measurement begins before delivery. Define the result, baseline, indicator, evidence source and reporting period while the project can still collect usable data. A late dashboard cannot repair a missing baseline or prove that the grant caused a change.

This guide was researched on 4 September 2026 for small businesses delivering projects in England. It uses current UK government evaluation guidance as a methodological reference. A recipient must still follow the definitions, templates and dates in its own funding agreement.

What to take away

• Define the result, baseline, indicator, evidence source and reporting period before delivery starts.
• A late dashboard cannot repair a missing baseline or prove that the grant caused a change.
• Monitoring shows change but does not by itself establish that an intervention caused it.
• Freeze indicator definitions before reviewing results and record any change with its reason and date.
• Report planned, committed, paid and forecast expenditure separately rather than mixing them.

Start with the decision the evidence must support

List the people who will use the report and the decisions they face. A funder may need assurance that public money was used for the approved purpose. A director may need to decide whether to keep funding the activity. A delivery lead may need to fix a process that misses its target group.

Write one question for each decision. Remove measures that do not help answer any question or meet a binding reporting duty. More data can create cost and privacy risk without improving the decision.

The 2026 HM Treasury Magenta Book describes evaluation as a systematic assessment of an intervention's design, implementation and outcomes. It says evaluation should be proportionate and fit for purpose. The guidance is for government interventions, so a business should adapt the principles rather than claim formal compliance without evidence.

Build a results chain

Write the project as a chain:

1. Inputs: money, people, equipment and time.
2. Activities: the work performed.
3. Outputs: products or services delivered.
4. Outcomes: changes experienced by the business, customers, staff or other intended group.
5. Impacts: wider or longer-term effects to which the project may contribute.

Add the assumptions connecting each stage. New machinery is an input, installation is an activity and trained operators may be an output. Fewer defects or higher production capacity could be outcomes if measured. The purchase itself is not proof of those changes.

The chain should match the funded purpose and approved milestones. If the agreement uses a term differently, preserve its definition in the formal report and explain any analytical distinction separately.

Define every indicator

Give each measure a specification:

• name and purpose
• exact numerator and denominator where relevant
• unit of analysis
• inclusion and exclusion rules
• geography and population
• baseline date and value
• target and target date
• source system or instrument
• collection frequency
• owner and reviewer
• known limitations

For a job measure, state whether it counts employees, full-time equivalents, vacancies or contracts. Say whether roles must be new, based in England, sustained for a period and directly employed. A label such as "jobs created" is not reproducible without these rules.

Freeze definitions before reviewing results. If a necessary definition changes, record the reason, date and effect on comparability.

Establish a baseline

The baseline is the measured starting position against which change is assessed. Choose a period representative of normal operation where possible. Record unusual events, seasonality and data gaps.

For new activities without historical records, use a defensible alternative such as a pre-project survey, existing comparable process or zero where zero is logically correct. Label an estimate and state its method. Do not reconstruct the starting value after seeing the result unless the limitation is explicit.

Preserve the source extract and calculation. A number copied into a presentation without its underlying period, filters and version cannot be independently checked.

Distinguish monitoring from evaluation

Monitoring tracks whether delivery is proceeding as planned. It may cover spend, participants, tasks, outputs, timing and operational outcomes. Evaluation asks deeper questions about implementation, effect, causes and value.

The Magenta Book identifies process, impact and value-for-money evaluation. Process work asks how delivery operated and for whom. Impact work examines what difference the intervention made. Value-for-money work compares costs and benefits or cost-effectiveness.

A small grant may justify a light, well-designed review rather than an elaborate study. Proportionate does not mean vague. State what the evidence can and cannot establish.

Plan collection before activity begins

Map each indicator to a source. Prefer records already produced through delivery when they are accurate and suitable. Add surveys, interviews or observation only where they answer a defined question.

Set the collection moment. A participant contact list may be available during delivery but unusable for follow-up if consent, privacy information and secure retention were not planned. Consult the ICO data-protection guidance before collecting personal data.

Test forms and extracts with realistic cases. Check missing values, duplicates, date formats, unit conversions and accessibility. Train the people entering data and give them a correction route.

Maintain an evidence register

Record each reported figure with its indicator version, source file, extraction date, calculation, owner, reviewer and report use. Keep raw evidence protected from edits and store working calculations separately.

Use consistent IDs for participants, purchases or outputs while applying data-minimisation and security controls. Avoid copying personal identifiers into a dashboard merely to support a count.

The Cabinet Office Model Grant Funding Agreement illustrates monitoring, reporting, audit and access provisions that may appear in central government funding. The recipient's actual agreement controls what must be retained and supplied.

Measure cost on the same scope

Report planned, committed, paid and forecast expenditure without mixing them. Reconcile financial periods, activities and populations with the outcome measures. A cost per participant is misleading if the cost covers the whole project but the participant count covers one quarter.

Show full project cost, grant contribution and business contribution separately where the decision requires them. Define whether VAT and in-kind resources are included. Explain allocations of shared overheads.

For unit cost, divide a relevant cost by a clearly defined output or outcome. Do not compare it with another programme until scope, period, geography, quality and accounting basis are sufficiently similar.

Handle attribution honestly

Observed improvement after a project does not prove that the project caused it. Sales, employment or productivity can change because of demand, competitors, staffing, prices or other investment.

The government's Quality in Policy Impact Evaluation guidance states that monitoring outcomes can show change but does not by itself establish that an intervention caused it. Robust causal attribution generally needs a suitable counterfactual and an appropriate design.

Small businesses will not always have a viable comparison group. In that case, use careful contribution language, test alternative explanations and triangulate operational data with interviews or other evidence. Write "the project was associated with" rather than "the grant caused" unless the design supports causality.

Report uncertainty and data quality

For each key finding, state coverage, missing data, measurement error and material assumptions. Use ranges or sensitivity analysis where a single value suggests false precision.

Investigate missingness rather than deleting it. If customers with poor results are less likely to answer a survey, the responses may overstate the benefit. Show response numbers and rates, not percentages alone.

Correct errors through a logged process. Preserve the submitted report, correction, date and reason. Never adjust a target retrospectively to make performance appear favourable.

Design a useful dashboard

Keep the top view small. Show grant and project spend, cash position, milestone status, core outputs, outcome trend, evidence quality and open exceptions. Every tile needs a definition and route to the supporting detail.

Use a consistent reporting date and status rule. A green milestone should mean that evidence shows it is complete or on track within stated tolerance. Avoid manual colours without recorded criteria.

Restrict access according to sensitivity. A board may need aggregated performance without participant names, bank information or payroll detail.

Write the narrative around evidence

Begin with the approved purpose and reporting period. Summarise what was delivered, what changed, the cost and whether the project remains on plan. Then explain material variance, limitations and action.

Separate facts, interpretation and forecast. Link each material factual claim to a source in the internal evidence register. Use concise tables where comparison matters and prose where context or limitation needs explanation.

Report adverse findings. An honest account of a missed group or delayed benefit helps a funder and business make a better decision. Selective success stories weaken assurance.

Review and release

Have the measure owner check source accuracy, finance reconcile cost figures and an independent reviewer test calculations and claims. Obtain the required authorised signature. Check that sharing complies with the agreement, confidentiality and data-protection duties.

Retain the approved report, source extracts, calculations and review record. Schedule later outcome measurement only where it has an owner, lawful data route and budget.

Set an update timetable

Give each indicator a refresh cycle that matches the decision. Cash and material delivery risks may need weekly attention, while a sustained employment outcome may be meaningful only after several months. Do not refresh a slow-moving measure merely to fill a monthly chart.

Record the next collection date and the trigger for an unscheduled review. A changed project scope, failed data feed or revised funder definition should prompt an assessment of every affected figure. Keep earlier issued values visible when a correction changes the trend.

This draft contains no live internal links. It remains on hold pending an identifiable author, fact-checker, qualified methodological review and publication-day verification of every current source.

Before you act

• Write one question for each decision the report must support.
• Specify numerator, denominator, inclusion rules and baseline for every indicator.
• Freeze definitions before reviewing results.
• Plan collection and consent before activity begins.
• Log each reported figure with source, date, owner and reviewer.
• State coverage, missing data and assumptions for each finding.

Common questions

What is a baseline and why does it matter?

The baseline is the measured starting position against which change is assessed. Choose a period representative of normal operation where possible, and record unusual events, seasonality and data gaps. For new activities without historical records, use a defensible alternative such as a pre-project survey or zero where zero is logically correct.

How should a small business handle attribution?

Observed improvement after a project does not prove that the project caused it. Sales, employment or productivity can change because of demand, competitors, staffing, prices or other investment. Without a viable comparison group, use contribution language, test alternative explanations and triangulate operational data with interviews or other evidence.

What is the difference between monitoring and evaluation?

Monitoring tracks whether delivery is proceeding as planned, covering spend, participants, tasks, outputs, timing and operational outcomes. Evaluation asks deeper questions about implementation, effect, causes and value. The Magenta Book identifies process, impact and value-for-money evaluation, and a small grant may justify a light, well-designed review rather than an elaborate study.