Operations first, beneficiaries never automated
Charities hold useful operational data - service demand, volunteer availability, referral patterns, outcomes - and it is generally under-used because analytical capacity is scarce.
The line to hold is clear. Forecasting how many people will need a service, and resourcing accordingly, is straightforwardly beneficial. Deciding which individuals receive support is a different matter and should remain a human judgement within a governed process.
Forecasting service demand
- Seasonal patterns - many services peak in winter or around holidays
- Referral source patterns, since partners refer in waves
- Local events affecting demand - closures, benefit changes, weather
- Day of week and time of day for drop-in services
- Lead indicators from enquiry volumes before formal referrals
The last point is frequently the most useful and least used. Enquiry or helpline volume often rises before formal referrals, giving useful warning of a coming increase.
Volunteer scheduling
Volunteer capacity is uncertain in a way paid staffing is not. Volunteers cancel more, availability changes, and retention varies considerably.
| Prediction | Use |
|---|---|
| Likelihood of attending a shift | Over-recruit sensibly for critical sessions |
| Volunteer retention risk | Earlier support and recognition |
| Skills availability by session | Match to service requirements |
| Seasonal availability | Plan recruitment ahead of peaks |
Retention prediction deserves care in framing. Used to offer support and recognition to volunteers showing disengagement, it is helpful. Used to deprioritise them, it is self-fulfilling and contrary to the point.
Outcome measurement
Funders increasingly require outcome evidence, and charities frequently hold the data without the capacity to analyse it.
Careful analysis can identify which service combinations correlate with better outcomes - though correlation is not causation, and beneficiaries who engage with more services may differ systematically. Comparison groups and honest caveats matter here more than elsewhere, because the conclusions affect funding and service design.
The constraints to write down first
- Beneficiary data is often special category data and needs corresponding protection.
- Any model touching individuals needs an assessment before it is built.
- Bias testing matters more here, since the people served are often those least well represented in data.
- Transparency with beneficiaries about how their data is used is a trust issue as well as a legal one.
- Prioritisation decisions should remain with people who are accountable for them.
None of this prevents useful work. It shapes which projects are appropriate, and starting with operational forecasting rather than beneficiary-level prediction avoids most of it entirely.
Forecast how many people will need the service. Do not build something that decides which of them gets it.