Human by choice · Worked example · A Policy Window advocacy programme
Sharing the time saved by AI
Prepared Published Edition 2
In this reading
A fictional allocation statement
Five minutes saved on administration could mean more attention for a service user, less work for staff, or lower spending. Here, three invented requests compete for the same time. The example fills in the eight questions in Human by choice's allocation statement; it shows how to record a disagreement without pretending it has been resolved.
Every person, preference, power and number below is invented. The arithmetic uses the appointment model, rather than observations of a service.
1 Decision and responsibility
The proposal concerns administration around 1,000 community appointments a year. In this fictional organisation, the governing body owns the overall decision and authorises spending. The service manager sets schedules and has delegated authority to pause the comparison or affected change for review. The employer must approve changes to paid working hours; further permissions remain to be checked. Workers and service users can submit objections; neither group has a veto here.
2 Starting point
Each appointment uses 20 minutes of conversation and ten of administration: 500 paid hours annually. At an assigned £40 per hour, that is £20,000. The example has no baseline observations of errors, waiting, workload or access. These would need recording before comparing a changed service. An unchanged appointment count does not guarantee that the same people receive it.
3 Time genuinely available
Assume that checking and administration fall to five minutes, including corrections. Keeping the original conversation time would then release 83⅓ hours annually. This assumes the whole task becomes faster; work passed to colleagues or users must also be counted. The service must establish whether scattered minutes can become usable schedule changes. Until then, 83⅓ hours is a calculation, not time anyone can claim on a rota.
4 Cash costs and payer
Software costs the service £2 per appointment, plus £2,000 setup: £4,000 in year one and £2,000 annually thereafter. Training and evaluation remain unpriced. Unchanged pay yields no salary cash saving. The budget holder must identify payments that could stop; £40 per hour is only an assigned value. Funding remains unidentified, so no purchase is authorised.
5 Competing allocations
The invented requests are deliberately incompatible:
| Request | Where all five minutes would go | What remains to establish |
|---|---|---|
| Service users want longer conversations | Conversation rises from 20 to 25 minutes | Whether people want this and whether access and quality improve |
| Staff want shorter working time on unchanged pay | 83⅓ fewer hours worked annually across the service | Whose hours, what coverage and agreement |
| The budget holder wants lower expenditure | Buy 83⅓ fewer paid hours annually | Whether payments can stop and whose income falls |
An illustrative split gives conversation three minutes and shorter working time two per appointment: 50 extra conversation hours and 33⅓ fewer work hours annually, allocating 83⅓ once. Pay stays unchanged. Work takes 28 minutes: 23 conversation and five checking and administration. Shorter working time depends on schedules and coverage; no group has accepted the split. The service still has to fund the technology.
6 Worker and user input
No real participants were asked. In a real decision, the three requests would need replacing with actual views, including those of people who struggle to access the service or cannot attend a meeting. Staff could disagree with one another; service users could prefer shorter waits to longer conversations.
A proposed procedure is to circulate the options before deciding, record objections, and require the decision owner to answer them. An unresolved objection could be sent to the governing body for reconsideration. Under the fictional powers above, that body could still approve an option over objections. A right to refuse or negotiate an allocation would need separate agreement; this statement does not supply it.
7 Best feasible alternative
Compare the mixed allocation with improving the manual process using the same available resources. Better scheduling or simpler records might free time without this software, but that has not been tested either. Estimate the work, cost and effects of each option before choosing. Recording a distribution of gains cannot establish that buying the tool is worthwhile.
8 Review and decision status
Status: proposed. The next decision is whether to fund a comparison, not whether to book a saving or change anyone's hours. The comparison's own cost and permissions need approval.
Before starting, the governing body would name recorders and set a review point, for example after the first 100 appointments. Record whole-task time, corrections, actual payments, hours worked, conversation time, access and workload.
Measure the administration and checking gain before allocating it. Five minutes freed funds three extra conversation minutes and two minutes less work. Reconsider the split if that initial gain, after corrections and work shifted elsewhere, falls below five minutes per appointment, or coverage prevents shorter working time. The manager would pause the affected change for investigation if quality or access materially deteriorates; define those measures and thresholds before starting. Record disagreement again at review. A faster process leaves the question of who decides open.
Human by choice is Policy Window's advocacy programme on AI, work and who decides. This example is analysis and advocacy, distinct from the neutral Policy Window wiki.