Your whole schedule lives
in one person’s head.
They know which tech handles that account, which customer will not accept a late window, and what is actually on the truck. When they take a week off, the operation runs at half speed. We build the system that carries that knowledge, reading your live data instead of a whiteboard.
The morning board is built by 6:30 and destroyed by 9:15. A tech calls out, a customer is not home, a two-hour job turns into five, and a part that the system says is in stock is not on the truck. From there the day is triage, run by a dispatcher with a phone in each hand.
Everybody in the building knows what this costs. Windows missed, a second trip that should never have happened, overtime that was not in anyone’s plan, and a customer who will remember the wait longer than the repair. What nobody has is the number.
The software you already bought does not fix it, because it was built to record decisions rather than make them. It knows the appointment exists. It does not know that this particular customer has cancelled twice, that this tech is forty minutes faster on this equipment type, or that the part on the ticket is sitting in the other truck.
One callout at 6:40am
Ninety minutes of triage and three unhappy customers.
A decision before the first truck leaves the yard.
What We Build
Constraint-Aware Scheduling
Skills, certifications, territory, customer history, and equipment type factored into the assignment. Not just who is closest, which is the answer that keeps producing second trips.
Mid-Day Re-Optimization
A callout or an overrun triggers a re-plan in seconds instead of ninety minutes of phone triage. This is where the day is actually won or lost.
Realistic Duration Prediction
Job length estimated from your own completion history rather than from a default block that has been wrong for six years.
Parts & Truck Stock Awareness
The schedule checks whether the part is genuinely available and on the right vehicle before committing a window. Second trips are pure margin loss.
Demand & Inventory Forecasting
Seasonal load, failure patterns, and reorder points read off your real history, so stockouts stop being a surprise every year at the same time.
Explainable Recommendations
Every suggestion shows its reasoning and can be overridden. Dispatchers adopt a system they can argue with and abandon one they cannot see into.
Direct Database Integration
Reads and writes against the ERP, field service platform, or scheduling system you already run. No replacement project, no parallel source of truth.
Owned versus rented, over three years
Scheduling systems re-evaluate constantly, which is exactly the usage pattern metered pricing handles worst. This compares infrastructure cost only, and ignores the recovered trips and overtime, which is where the real return sits.
Estimate only. Metered pricing assumes roughly 4,000 tokens per decision at a blended $5 per million tokens, a mid-range figure across current commercial providers. The local column adds a 12 percent annual reserve for hardware refresh and maintenance. Your real numbers come out of the Operational Snapshot, not out of a slider.
Built for operations that run on judgment calls
Field Service Operations
Crews, windows, skills, and parts. The four variables that fight each other every morning and get resolved by memory.
AI for professional servicesContractors & Trades
Multi-site work with shifting crews, weather delays, and inspections. The schedule is a negotiation, and it changes hourly.
AI for professional servicesSupply Chain & Distribution
Reorder timing, dock scheduling, and load planning read off your live inventory rather than a report that was accurate yesterday.
On-site automation systemsMulti-Location Practices
Provider availability, room and equipment constraints, and patient preference across sites, coordinated centrally without one person holding it together.
AI for healthcareThe morning schedule is not the hard part
Route and crew assignment has been studied formally for sixty years. What changed is which version of the problem matters. Pillac, Gendreau, Gueret and Medaglia draw the line cleanly in A Review of Dynamic Vehicle Routing Problems: a static problem knows everything at the start, and a dynamic one has information revealed while the day is already running. Real dispatch is the second kind. The call that changes your afternoon arrives at ten forty. A plan built at seven is stale by mid-morning, no matter how good it was at seven.
That review is the reason we do not sell a better morning schedule. The research on dynamic routing concentrates on re-optimization under partial information, using rolling-horizon methods and metaheuristics that can produce a revised plan fast enough to act on. The value is in the recovery, not the plan. A dispatcher who can reshuffle six technicians in ninety seconds when a job blows up beats a perfect plan that assumed nothing would.
The solver itself is not where the money goes. Google publishes OR-Tools, a production-grade constraint and routing engine, under an open license. Time windows, skill matching, capacity limits, and mandatory breaks are all first-class constraints in it. The expensive, bespoke work is the part nobody can package: encoding the rules your business actually runs on, including the ones that live only in the head of the person who has always done the schedule.
Which makes the pricing question worth asking early. Field service AI is increasingly sold on a per-conversation or per-interaction meter, as Salesforce Agentforce pricing shows, and the model layer underneath is metered too, as Amazon Bedrock pricing makes plain. A dispatch system re-optimizes constantly by definition. Metering the thing you want it to do all day is the wrong incentive to build a business on. Run it on hardware you own and re-optimization is free, so the system does it every time something moves instead of only when it is worth the charge.
We scope the governance side against the NIST AI Risk Management Framework, because a scheduling system quietly becomes a system of record for who was where and when. That is evidence in a dispute, and it should live somewhere you control.
How We Work
We start with an Operational Snapshot and a morning next to your dispatcher. The decisions that matter are almost never written down anywhere, and you cannot encode a rule you have not watched somebody apply.
Then the system runs alongside the humans before it runs anything. It proposes, your dispatcher decides, and the two get compared. When the suggestions are consistently as good or better, people start taking them, and adoption stops being a change management problem.
It reads your existing systems rather than replacing them, and you own what we build outright. That is the premise of the Sovereign AI System: infrastructure in your control, with no meter on the decisions your business runs on.
References and further reading
- 01Pillac, Gendreau, Guéret and Medaglia, A Review of Dynamic Vehicle Routing Problems (European Journal of Operational Research)
- 02Google, OR-Tools vehicle routing documentation
- 03Salesforce, Agentforce pricing (per-conversation billing for field service agents)
- 04NIST, AI Risk Management Framework 1.0
- 05AWS, Amazon Bedrock pricing (metered token model)
- 06MLCommons, MLPerf Inference: Edge benchmark results
Sources are cited for context and verification. Vendor pricing pages change without notice, and figures quoted here reflect published rates at the time of writing. This study is general information, not legal, clinical, or engineering advice.
Find out what your second trips are costing you this year
Almost nobody has this number, and it is usually larger than the build. Request an Operational Snapshot. Thirty minutes, a full diagnostic, and the real math on missed windows, overtime, and repeat visits.
Request an Operational SnapshotNo sales pitch. No guilt trip. Just clarity on what's costing you.
See also: on-site automation systems and the sovereign service backbone.
Built by Perry Luzier, Founder of Luzran.