Could Home-Care Visits Be Planned Like Delivery Routes? It Depends Who Pays for Travel

The problem

Starting point. The anchor paper found that machine learning inside delivery route planning is documented in detail at only two very large operators, and that the large headline savings usually credited to AI route optimisation belong mostly to older optimisation methods¹. Where one learning component was isolated in an A/B test, which splits traffic randomly between the old and new versions, it cut travel distance by 1.20% and raised the on-time rate by 0.27 percentage points (operator-run A/B test across nine cities, published 2024)². Couriers also departed from the plan: in Shanghai they chose the computed next stop 47% of the time (observational study, independent, February 2025)³.

This paper asks what the same kind of planning could offer where the stops are care visits rather than parcels, and who would keep the gain.

The setting. In England, an estimated 740,000 filled posts, 46% of all adult social care filled posts, were in domiciliary care (home care) in 2024/25 (Skills for Care estimate)⁴. Among domiciliary care workers, 35% were on zero-hours contracts, and among care workers in domiciliary services the share was 42% (Skills for Care estimates, 2024/25)⁴.

Travel as working time. Official guidance states that time spent travelling from one work assignment to another counts as working time for minimum wage purposes. Its care worker examples show pay having to cover at least the minimum wage for the time at appointments plus the travel between them, while travel between home and work does not count (official guidance, updated 8 November 2024)⁵. Enforcement history shows the rule has not always been followed: in 2013/14 HMRC told one social care provider to repay more than £600,000 to almost 3,000 workers who had not been paid for travelling time and other hours worked (official news release, search extract, older than three years)⁶. In June 2026 a parliamentary question concerned the non-payment of travel time by homecare employers. The department's answer restated that travelling between appointments counts as working time, and that HMRC enforcement continues until the Fair Work Agency takes over minimum wage functions in April 2027 (parliamentary answer, 16 June 2026)⁷.

No current measure of how many home-care workers go unpaid for travel was found. The 2023 to 2024 enforcement report gives no social care breakdown and does not report travel time as a cause of underpayment, although the Low Pay Commission's 2024 enforcement report focused on social care (government report, June 2025)⁹. In the consultation on fair pay agreements, pay for the time and cost of travelling between visits was chosen as a priority by just over half of respondents to one multi-select question, which the department says had technical problems in data collection (consultation response, 16 July 2026)⁸.

Late and missed visits. CQC's 2013 themed inspections of 250 home care agencies, covering around 26,500 people, found most providers delivering a good service but a minority of people affected by late or missed calls, and numerous examples of rotas that did not allow time for staff to travel between visits. Thirty-eight of the 250 providers were not meeting the standard on care and welfare (regulator report, 13 February 2013, older than three years)¹⁰. A CQC assessment of one provider in July 2024 found significant delays, visits shorter than planned, and rosters not always planned to give staff travel time and breaks (regulator assessment, July 2024, search extract)¹¹. No current national count of missed or late visits was found.

What already works

Planning software for home visits is not new. The sources fall into three groups: older optimisation systems, research, and vendor claims about newer products described as AI.

Optimisation in routine use. Laps Care, a Swedish system developed in 2002, uses operations research modelling to plan staff scheduling and routing. By 2009 more than 200 municipal units used it daily for 4,000 home care workers, and its developers report a 10 to 15% gain in operational efficiency, equivalent to €20 to 30 million a year (developer-reported, peer-reviewed abstract, 2009, older than three years; method not stated)¹².

Research on the trade-off. A simulation of community midwives in the English NHS found that continuity of care, meaning the same midwife seeing the same mother, could be achieved with relatively small increases in travel time, although shift patterns made perfect continuity impractical (simulation, peer-reviewed abstract, 2015, older than three years)¹³. The setting is midwifery rather than personal care, and the evidence is a simulation.

Newer products, vendor claims. Cera, a home healthcare provider, reported in February 2024 that an AI-led scheduling platform had delivered 30% more care visits in parts of England by cutting carers' average travel time between patients by 50% (vendor-reported, trade press, 23 February 2024; method not stated)¹⁴. A software company reports that its scheduling tool, deployed at an NHS foundation trust, cut average travel time by 12.7% and travel distance by 12.9%, and raised utilisation by 33% in one team (vendor-reported, date not shown; method not stated)¹⁵. Neither source says which part of the product uses machine learning.

What does not exist. None of the sources read gives an independent, controlled measurement of routing software, with or without machine learning, on routes actually driven in home care, or of its effect on missed visits or workers' paid hours. The delivery evidence for machine learning comes from China and Germany and is described in the anchor paper¹.

The possibility

One way this could work: a home-care provider or council team would plan each day's visits the way a delivery firm plans a van's stops, as a single optimisation across all workers and visits. The planner would carry the constraints that distinguish care from parcels: visit time windows tied to medication or meals, workers' skills, visits needing two workers, and a preference for the same worker each time. The plan would be revised during the day when a visit overruns or a worker is absent, and the system would record the travel it creates so that it can be paid and checked.

The role of AI. Most of this would be ordinary software. The search for good assignments of visits to workers is operations research, mathematical optimisation, which Laps Care shows in home care: developed in 2002 and in daily municipal use by 2009¹². Machine learning, meaning software that learns patterns from historical data, could add three narrower things: predicting how long each visit and journey really takes instead of assuming a standard slot; learning which pairings of workers and clients tend to work; and suggesting re-plans as the day changes. The anchor paper found these jobs documented at two delivery operators, with small isolated gains where measured¹². The rest of the possibility is not AI at all: contracts, the rate councils pay per visit, pay rules for travel, call-monitoring systems and clean address data. On the evidence read, the plausible contribution of machine learning is narrower than that of the planning engine and, above all, of pay and commissioning rules.

Who would gain. If shorter journeys let a provider fit more visits into the same paid hours, the gain would accrue mainly to the provider unless contracts or rules direct it elsewhere. Cera's reported use of its travel saving was more visits¹⁴. If travel time were instead paid as a separate item, for example under a fair pay agreement, every minute of travel would cost the provider directly, and a planner that shortens journeys would cut wage costs as well as raise capacity. Whether a fair pay agreement would cover travel pay is open: the department's response places pay and terms and conditions in the negotiating body's remit but leaves the choice of issues to the body, within a remit letter still to be developed (government response, 16 July 2026)⁸.

Missed visits. If part of the late and missed visits comes from rotas that assume no travel, a planner that models travel realistically would remove that cause. CQC found rotas without travel time in 2013 and in a July 2024 assessment¹⁰¹¹, but none of the sources read measures how much of the wider problem they explain.

What would have to be true

1. Realistic time estimates. Plans would need measured visit and journey times rather than standard slots, because plans that ignore practice are not followed: the Shanghai courier study found the computed next stop chosen 47% of the time (observational study, independent, February 2025)³.
2. Continuity as a constraint. Continuity of care would have to be built in as an explicit requirement, with some flexibility in visit times. The midwifery simulation points to a modest travel cost for reasonable continuity, but not for perfect continuity¹³.
3. Travel counted and paid. The planner would need to record the travel it creates so that it can be paid and audited as working time, which the guidance already treats as working time⁵. Any separately itemised travel pay would depend on enforcement of that rule or on a negotiated agreement, the first of which the department projects to take effect in April 2028 (projected by the department, 16 July 2026)⁸.
4. Notice and fair re-planning. Re-planning during the day would have to sit within the Employment Rights Act 2025, which provides for rights to guaranteed hours and to reasonable notice of shifts, with payment for shifts cancelled, moved or curtailed at short notice, delivered in phases over two years (government factsheet, date not shown; government report, June 2025)¹⁶⁹. The commencement dates for those measures were not confirmed in the sources read.
5. Pay and commissioning with room in them. Consultation responses stressed that funding might not reach frontline staff and that commissioning reform would be needed to deliver a fair pay agreement (government response, 16 July 2026)⁸. A planner could shorten paid travel only if the contracts it works under pay for the time it creates.
6. Independent evidence. A controlled evaluation of planning software on routes actually driven, reporting late and missed visits and workers' paid hours, would be needed before the vendor figures could be treated as more than claims. None was found.

What could stop it

What the evidence shows

• Headline gains are self-reported. The home-care efficiency figures come from a system's developers¹² and from vendors¹⁴¹⁵. None of the pages read states how the figures were measured, and none was independently evaluated in the sources read. In delivery, the isolated machine-learning gain was about 1% in distance where measured (operator-run A/B test, published 2024)¹².
• Plans are not always followed. Couriers chose the computed next stop 47% of the time in Shanghai (observational study, independent, February 2025)³.
• Contract structure. 35% of domiciliary care workers were on zero-hours contracts in 2024/25 (Skills for Care estimate)⁴. The Employment Rights Act 2025 provides for rights to guaranteed hours and reasonable notice of shifts¹⁶.
• Slow rule change. The department's stated timetable has regulations laid in 2026, negotiations beginning around April 2027 and the first fair pay agreement taking effect in April 2028, backed by £500 million in the first year, which is not ring-fenced (projected by the department, 16 July 2026)⁸. The negotiating body chooses which issues to pursue within the remit letter⁸.
• No measured effect on visits or pay. No source read reports the effect of routing software on late or missed visits, or on workers' paid hours. CQC reports link late and short visits to rotas without travel time at individual agencies¹⁰¹¹.
• Older evidence. The Laps Care figures date from 2009¹², the midwifery simulation from 2015¹³, the CQC inspection programme from 2013¹⁰ and the HMRC case from 2013/14⁶.

Reasoning

Would simpler solutions do? If the benefit comes from planning at all, the Laps Care record suggests that operations research was enough to deliver it, without machine learning¹². If the binding constraint is how much is paid per visit and whether travel is paid, a better planner could not fix it: the shortest possible journeys would still be unfunded under a contract that pays only for contact time.

Who would bear the risk? If shorter routes were used to pack more visits into the day, workers could face tighter schedules, the pattern CQC describes where rosters do not allow for travel and breaks¹¹. Clients could see continuity traded for density if the planner weighted travel above familiarity. The midwifery simulation suggests the trade can be managed but not removed¹³.

Who would pay? No price or cost figure for home-care scheduling software was found in the sources read. If a provider or council bought it, the return would depend on whether the saving appeared as fewer paid travel hours or as more visits, and where councils pay by contact time only the second would be visible to the buyer.

Whose incentives work against it? The minimum wage rule sets a floor for hourly pay across all hours worked, including travel⁵. A provider whose average pay still cleared that floor could meet the rule while treating travel as unpaid as a separate item, and would then have little financial reason to shorten journeys. That incentive would change only if travel were paid explicitly.

Signs to watch

• Remit letter for the negotiating body: the department says priorities from the consultation feed a remit letter later in 2026. Watch whether it names pay for travel between visits⁸.
• April 2027: negotiations for 2028/29 are projected to begin around this date, and the Fair Work Agency takes over minimum wage functions from HMRC (projected by the department)⁸⁷.
• April 2028: the first fair pay agreement is projected to take effect. Watch whether it itemises travel pay⁸.
• Revised minimum wage guidance: the Government said it would review the calculating-the-minimum-wage guidance in 2025 to 2026. Watch for the published text on working time⁹.
• Employment Rights Act regulations: commencement of the guaranteed hours and notice of shifts measures¹⁶⁹.
• Trigger: the first independent trial or A/B evaluation of planning software on driven routes in UK home care, reporting late and missed visits and workers' paid hours.