Situation
The operating reality
The off-the-shelf routing engine charged per stop, did not respect driver work-time regulations in the carrier’s market, and produced routes that drivers routinely ignored. Costs were rising faster than volume.
Case study · Logistics
A routing engine that cut last-mile costs by 19% in a regulated market
Replaced a third-party routing engine with a domain-specific solver that respected driver work-time rules and reduced cost-per-stop by 19%.
Anonymized delivery dashboard
Logistics outcome cockpit
Cost per stop
−19%
On-time rate
+8 pts
Engine cost
−83%
Before
Constraint
Build
Controlled cutover
After
Measured gain
- Client
- Regional last-mile carrier
- Industry
- Logistics
- Service
- Custom Software Development
- Engagement closed
- July 2024
Storyline
From constraint to measurable change.
Every engagement is framed around the business situation, the constraint that made it hard, and the decision that turned delivery into a controlled path to value.
Constraint
Why it was hard
The off-the-shelf routing engine charged per stop, did not respect driver work-time regulations in the carrier’s market, and produced routes that drivers routinely ignored. Costs were rising faster than volume.
Decision
The path we chose
Built a domain-specific routing solver because the commercial and regulatory constraints were the product, not edge cases.
Build
Domain-specific solver
We built a vehicle-routing solver in Go using a metaheuristic appropriate to the problem size, with hard constraints encoded for legal work-time, vehicle weight, and customer time windows. Soft constraints captured the driver preferences that surfaced in the field interviews.
Routes were reviewed by depot managers before dispatch via an interface designed with them. Acceptance and override rates were captured per route and fed back into the solver weights weekly.
Outcome
Measured impact
Cost-per-stop dropped 19% in the first full quarter on the new engine.
On-time delivery rate improved by 8 percentage points.
What changed after launch
New behavior
Depot managers could review, override, and improve routes, turning local knowledge into weekly optimization signals.
Outcome
The numbers that mattered.
- Cost per stop
- −0%
- On-time rate
- +0 pts
- Engine cost
- −0%
“The new routes respected the real world. Drivers trusted them because their constraints were finally represented.”
Before and after
The transformation the client could see.
The work was not abstract modernization. It changed day-to-day behavior, ownership, and the evidence leaders used to make decisions.
Optimization engine
Before
- Per-stop SaaS costs rising with volume
- Routes ignored driver regulations
- Drivers overrode plans manually
After
- Rules encoded into the solver
- Override feedback improved weights weekly
- 83% lower engine running cost
Architecture and delivery
A controlled path from discovery to launch.
The delivery plan made the system boundary explicit, then used rehearsals, gates, and telemetry to optimize safely before launch.
Delivery architecture
Logistics control loop
Discover
Domain-specific solver
We built a vehicle-routing solver in Go using a metaheuristic appropriate to the problem size, with hard constraints encoded for legal work-time, vehicle weight, and customer time windows. Soft constraints captured the driver preferences that surfaced in the field interviews.
Launch
Operationalised, not just optimized
Routes were reviewed by depot managers before dispatch via an interface designed with them. Acceptance and override rates were captured per route and fed back into the solver weights weekly.
Build
How we shaped the work.
Domain-specific solver
We built a vehicle-routing solver in Go using a metaheuristic appropriate to the problem size, with hard constraints encoded for legal work-time, vehicle weight, and customer time windows. Soft constraints captured the driver preferences that surfaced in the field interviews.
Operationalised, not just optimized
Routes were reviewed by depot managers before dispatch via an interface designed with them. Acceptance and override rates were captured per route and fed back into the solver weights weekly.
What changed after launch
What shipped, and what it changed.
- Cost-per-stop dropped 19% in the first full quarter on the new engine.
- On-time delivery rate improved by 8 percentage points.
- Engine running cost fell 83% vs. the previous per-stop SaaS contract.
After launch
Depot managers could review, override, and improve routes, turning local knowledge into weekly optimization signals.
Stack
What we built it with.
Go
PostgreSQL
Kafka
Next.js
AWS
Terraform
Related case studies
View all workSimilar delivery patterns.
A similar problem?
Let’s talk about your project.
A senior engineer will follow up within one business day with an opinionated take on the shape of the work.