Crew schedule efficiency across the Heartland operating companies

Each crew-week is costed three ways as a loop out from the yard and back. The crew never changes.

“Yard” in this document means the fixed location each crew-day starts from and returns to — the branch’s operating base. Aspire has no field that identifies one, so it was established two ways: for branches that have a property record named after the branch, those coordinates; and for the rest, the place crews are observed to finish the day, taken from the shift clock’s recorded position. The two agree closely where both exist — within 10 to 120 metres across Signature’s five branches — and crews clock in there, with 92% of clock-ins inside 500 metres. Whether the address is a truck yard, a shop or an office is not recorded, and does not affect the measurement.

• As planned — the planned days, in the planned stop order.
• Reordered — the same visits on the same days, in the best sequence.
• Fully optimised — visits regrouped across the days that crew was already going to work, then resequenced.

The two levers follow directly. Lever 1, stop order, is the gap between as-planned and reordered. Lever 2, day grouping, is the further gap to fully optimised. They add to the total, and the worked examples below show them separately for real weeks.

Sequences are solved exactly, by dynamic programming, wherever a day has eleven stops or fewer — between 80% and 98% of crew-weeks depending on tenant. Longer days use a method that returns an upper bound on the best route, so those days understate the saving. Every approximation in this analysis is set to err low.

A solver run over past data can take advantage of knowing which day it rained, which crew was short, and which job overran. A scheduler on the Friday before knew none of that, so a saving that depends on such knowledge is not one anyone could have captured.

The headline figures therefore use only what was recorded before each week began: the planned day for each visit, the planned stop order within that day (present on 96% of visits), planned hours rather than actual hours, and a daily capacity ceiling taken from that week’s own plan. No outcome enters the calculation. For contrast, the same method applied to what actually happened is shown alongside.

Removing hindsight does not remove the opportunity. Because adherence is high, the planned days and the worked days are nearly the same, so the two columns differ mainly in whether planned or actual hours set the capacity ceiling. JML is the one clear exception, where the hindsight figure is much higher; its planning-time figure is the one to rely on.

High adherence is what makes these figures actionable. If plans were routinely set aside, improving them would change nothing on the ground. At 96.3% across 305,772 matched visits, the plan is what happens.

Lever 1 measures something Aspire already does: selecting a crew-day and choosing Optimize Route resequences that day’s stops. That raises an important question for management, because the answer changes the response entirely. If the feature is ineffective, the gap needs new software. If it is effective but unused, the gap needs a process change.

This is testable. If the feature ran on a given day, that day’s recorded stop order should be the best available order, or within a fraction of it. So the share of days that are already exactly optimal indicates how often it was applied — provided the figure is compared against how often a hand-built order would be optimal by luck, which on a four-stop day is one time in three.

On days of six stops or more, where landing on the best order by chance is effectively impossible, 8–21% of days are exactly optimal. That cannot happen by accident, so the feature is being used and it works. It is simply not used consistently: across all judged days the median gap to the best order is 5.9% at Signature, 3.9% at JML and 10.9% at Keesen, and roughly a quarter of days sit more than 15% off.

The pattern by day size is the most useful part. The largest days — where the saving is greatest — are the ones where it is never applied. No day of 18 or more stops was optimally sequenced at Signature. This is consistent with the feature being used opportunistically on simple days and skipped when a day looks complicated.

Distance is the unit of measurement, not the value. The avoidable planned distance converts to 55,883 crew-hours over the seven months: 563,765 straight-line kilometres, adjusted by a 1.3 road factor, at 40 km/h, multiplied by the measured crew size of 3.05 people per crew-day. Fuel over the same distance is roughly an order of magnitude less valuable than the labour.

Aspire records labour two ways, and the distinction matters for putting that figure in context.

• Time booked to work tickets — 2,617,930 h. This is what job costing sees, and it averages 7.7 to 10.0 hours per person-day depending on tenant.
• Time on the shift clock — 3,239,278 h across 329,027 shifts, averaging 9.85 hours per shift. This is what the business pays for.

The difference is 621,348 hours, or 19.2% of all clocked time, that never reaches a work ticket. Against clocked labour, the avoidable driving is 1.7%; against booked ticket time it is 2.1%.

What this measurement cannot show. Visits only move between days the crew was already going to work, so the method can never indicate that five crew-days would do the work of six. Any reduction in crew-days is outside these figures and is treated separately below.

Whether freed time becomes value is an operating decision. It can become additional billable work, shorter days, or simply slack. Days are already substantial — a 9.85-hour average shift — so additional capacity is the most likely destination, provided there is demand to absorb it.

The figures above hold the working days fixed. A natural next question is whether saving enough time allows a week to be completed in fewer days. It does, and in crew-hours it is worth several times the routing saving. Whether it is worth anything net depends on how long a day crews are willing to work, because that is where it meets overtime.

Eliminating a day does not remove labour hours — the work moves to other days. What a freed day provides is capacity: the same crews covering more properties, or a six-day route becoming a five-day one. Converting that to money requires either demand to absorb it or a premium to avoid.

What each column means, and what to do about it

• Property coordinates — the share of properties with latitude and longitude. Strong everywhere (92–100%), and the foundation of any distance work. No action needed.
• Yard recorded — whether each branch has a property record for its own yard. Only Signature has them throughout. Five tenants have none, so their yard locations had to be derived from where crews are observed to finish the day. Recommended action: record a yard property for every branch. The derivation reproduced Signature’s five known yards to within 10–120 metres, so the coordinates are available and simply need entering.
• Yard confidence — how tightly crews cluster on that derived location. Signature reaches 75%; Heartland NE and LCM sit near 14%, meaning crews there do not converge on a single place. Seven Heartland NE branches and three each at LCM and Heritage fall below 20% and should be treated as approximate. The derivation also shows several branches sharing one physical yard — four Heartland AZ branches resolve to a single Tucson location.
• Untimed labour — hours entered manually with a midnight timestamp and therefore no usable clock times. 9% at Keesen, over 17% at Heartland AZ, Heartland NE and Heritage. These hours are excluded from the analysis, and any time-of-day reporting on them is unreliable. Recommended action: move manual entry onto the mobile app where practical.
• Planned order recorded — the share of planned visits carrying a sequence number. Strong except Keesen at 88.5%. Where sequence is missing, lever 1 cannot be assessed for that day.
• Visits never worked — planned visits with no labour recorded against them at all. This ranges from 3% at LCM to 37% at Heartland NE. Some will be genuine cancellations; the remainder is work completed without time capture. Either way the planned-versus- delivered picture is incomplete, and at the high end it undermines any completion reporting. This is the single largest data gap found.
• Actual vs estimate — hours delivered against hours estimated. Most tenants sit within a few points of parity, which is genuinely good. Heritage is the exception at 120%, roughly 108,000 hours beyond estimate.

Some constraints are not recorded anywhere. Aspire’s property availability table — where a “Tuesdays only, 8 to 4” rule would be held — is empty. No service interval or recurrence field exists in the interface. Property notes are populated on about a fifth of properties and are predominantly gate codes. Where a customer genuinely requires a fixed day, the analysis will show a saving that is not available. This is precisely why nothing here moves a visit outside its planned week or to a different crew.

Distances are straight-line. Real roads are longer, so absolute kilometres are a lower bound while the percentages are robust. Depot placement was tested directly and changes the percentages by at most 0.1 points.

Coverage is partial by design. Crew-weeks with a single working day, fewer than four stops, or no capacity history are excluded; roughly three-quarters qualify. The excluded weeks are mostly those where sequence and grouping cannot be wrong.

Earlier figures were higher. A previous version of this analysis reported a group median of 14.7% and 23.4% for Signature. Building the worked examples revealed two faults: the solver could empty a working day entirely, which counts a crew-day reduction as a routing saving, and the daily capacity ceiling did not bind where a multi-day job was recorded as a single visit. Both are now constrained, and both corrections reduce the reported saving. The figures in this document are the corrected ones.

The next action is not further analysis. It is a review of specific proposed changes with the people who build the schedules: take the weakest crew-weeks to a production manager and record how many of the proposed regroupings survive scrutiny. Begin with Signature KCMO, the widest single branch at 21.7%, and include Heritage as a control, where the analysis indicates there is little to find. The proportion accepted is what converts a percentage into a forecast, and it requires one meeting rather than another data exercise.

In parallel, three data changes would improve every figure here and are worth making regardless: record a yard property for each branch, reduce manual midnight time entry, and investigate why 37% of Heartland NE’s planned visits carry no recorded labour.