Adding a managed route optimization integration to an existing delivery app costs SAR 63,000–111,600 in the worked scenario below. A custom solver integration costs SAR 129,600–235,200 under a separate scope. Both figures cover implementation and pilot work. Recurring service usage and support are calculated separately.

This example assumes 25 vehicles, one depot, an operational driver app and an existing order database. All prices are illustrative estimates calculated from explicit hours and rates. They provide a method for building a budget, with inputs you can replace using your team's estimates.

What exactly are we adding?

Consider a fictional Riyadh delivery operator. Each morning, dispatchers distribute 300 orders across 25 vehicles. They account for vehicle capacity, delivery windows and driver availability using spreadsheets and local knowledge. Orders change during the day, and dispatchers adjust the plan manually.

The proposed module takes a validated order set, produces candidate routes and lets a dispatcher review and publish one version. Drivers receive the approved plan through their existing app. Late orders and exceptions go back to the dispatch team.

This scope assumes ordinary parcels, one departure location and no temperature-controlled loads. It also assumes that the existing system can identify orders consistently and record assignment changes. If either capability is missing, include the repair work before pricing an optimizer.

Route optimization assigns stops to vehicles and selects their sequence under constraints. Navigation provides directions for a journey. Address lookup, travel-time estimation, optimization and live navigation can involve separate services and billing units. List each dependency when requesting a quote.

Which inputs must be reliable before you optimize?

A solver can only work with the problem you describe. For each order, record coordinates, service duration, load demand and an agreed delivery window. For each vehicle, record usable capacity, shift limits and any restrictions relevant to the work.

Input Example Budget implication
Order location Verified latitude and longitude Address correction work if coordinates are unreliable
Service duration Five minutes for a normal stop Configuration and later calibration
Capacity Weight in kilograms and volume in litres Multiple capacity dimensions if both limit loading
Delivery window Service starts between 10 AM and noon More constraints and exception handling
Driver availability Shift and break periods Scheduling and dispatch integration
Locked work Driver has already started a stop Partial replanning rules

Choose whether a window constrains arrival, service start or service completion. That decision changes which routes are feasible. Also record whether a rule is mandatory or a preference. A delivery that cannot fit a mandatory window should surface as unassigned with an explanation.

Managed optimization API versus custom solver

A managed service can supply the optimization engine while your team builds the data adapter, review interface, publishing flow and monitoring. The provider's supported constraints and billing model become important dependencies.

A custom solver gives your team more control over modeling and deployment. It adds work for objective tuning, resource limits, runtime monitoring, upgrades and operational support. Open-source solver availability can reduce licensing expense while leaving substantial engineering work.

Decision Managed service integration Custom solver integration
Initial implementation Provider adapter and product workflow Solver model plus product workflow
Constraint choices Must fit the provider's supported model Team implements and maintains the model
Runtime operations Provider service plus your integration Solver infrastructure and your application
Change management Track provider behavior and versions Track model, library and infrastructure changes
Best starting condition Requirements fit an available service Requirements justify maintaining custom optimization

Prototype the difficult constraints using representative orders before selecting either option. A map demo with ten easy stops says little about a day containing narrow windows, oversized parcels and drivers already on the road.

Managed integration build budget

Assume a blended implementation rate of SAR 180 per hour. The rate includes the team's ordinary delivery overhead within the listed work packages. The estimates are incremental work on an existing application.

Work package Hours SAR
Data and workflow audit 40–70 7,200–12,600
Input adapter and validation 60–100 10,800–18,000
Optimization service integration 70–120 12,600–21,600
Dispatcher preview and override 70–130 12,600–23,400
Plan publishing and driver synchronization 50–90 9,000–16,200
QA, shadow runs and rollout 60–110 10,800–19,800
Total 350–620 63,000–111,600

The high end includes more edge cases within this defined scope. A separate reserve can cover unresolved dependencies. For example, a 15% planning reserve adds SAR 9,450–16,740, producing an approved envelope of SAR 72,450–128,340. Release the reserve against documented changes.

This excludes a new driver app, fleet hardware, multi-depot optimization and custom customer scheduling. Estimate calendar time from staffing, dependencies and pilot dates. Dividing total hours by the headcount assumes work can run in parallel, which some integration tasks cannot.

Custom solver build budget

For a separate custom scenario, assume SAR 240 per hour and 540–980 hours. The scope still uses the existing driver app and order database.

Work package Hours SAR
Workflow and data audit 50–80 12,000–19,200
Solver model and baseline comparison 140–260 33,600–62,400
Travel-time integration and input preparation 70–120 16,800–28,800
Dispatcher review and publishing 100–180 24,000–43,200
Jobs, recovery and observability 80–160 19,200–38,400
QA and operational pilot 100–180 24,000–43,200
Total 540–980 129,600–235,200

These are alternative build estimates. The custom figure is not an add-on to the managed budget. If you migrate later, re-estimate which workflow components can be reused and which assumptions need to change.

A small Python feasibility check

Before requesting optimization, validate obvious capacity and window problems. The following illustrative function checks a fixed delivery-only stop sequence. Minutes are measured from the planning day's start. Travel times between successive locations are supplied by the caller, and each stop includes its own service duration.

The function assumes every parcel is loaded at departure. It checks maximum starting weight, service-start windows and return before the shift ends. It does not choose vehicles or find an optimal stop order.

def check_route( stops, travel_minutes, return_minutes, capacity_kg, shift_start, shift_end, ): if len(travel_minutes) != len(stops): raise ValueError("Provide one travel time per stop") if capacity_kg < 0 or shift_start > shift_end: raise ValueError("Invalid vehicle or shift limits") if return_minutes < 0 or any(t < 0 for t in travel_minutes): raise ValueError("Travel times must be non-negative") for stop in stops: if stop["weight_kg"] < 0 or stop["service_minutes"] < 0: raise ValueError("Invalid stop weight or service time") if stop["window_start"] > stop["window_end"]: raise ValueError("Invalid delivery window") load = sum(stop["weight_kg"] for stop in stops) if load > capacity_kg: return {"feasible": False, "reason": "capacity_exceeded"} clock = shift_start service_starts = [] for stop, travel in zip(stops, travel_minutes): arrival = clock + travel service_start = max(arrival, stop["window_start"]) if service_start > stop["window_end"]: return { "feasible": False, "reason": "window_missed", "stop_id": stop["id"], } service_starts.append(service_start) clock = service_start + stop["service_minutes"] return_at = clock + return_minutes if return_at > shift_end: return {"feasible": False, "reason": "shift_exceeded"} return { "feasible": True, "service_starts": service_starts, "return_at": return_at, } 

For a stop with a 10 AM window, a driver arriving at 9:50 waits until 10 AM. A driver starting service exactly at the window's end passes this particular policy. If your customers require completion before that time, change the condition to include service duration and retest.

Production validation needs typed schema checks, consistent units and bounded values before this function runs. Overnight shifts need a date-aware representation. Pickups require a changing load model. Breaks, traffic uncertainty, volume, vehicle compatibility and depot loading queues also require additional logic.

What does monthly operation cost?

Assume 26 operating days, one initial plan per day and an average of one whole-fleet replan. For this budget only, assume an optimizer charges SAR 0.20 per vehicle included per solve. This is a fictional billing model. Replace it with the selected service's actual billing units.

Monthly item Calculation SAR
Optimization 25 vehicles × 52 solves × 0.20 260
Supporting travel-time requests 30,000 assumed requests × 0.02 600
Queue and monitoring Fixed allowance 400
Engineering support 8 hours × 180 1,440
Incremental total 2,700

At 300 orders per operating day, the operation plans 7,800 orders monthly. The listed incremental software expense is about SAR 0.35 per planned order. It excludes the existing app's hosting, drivers, fuel and customer support.

Travel-time matrices can become a major cost driver. A dense matrix for 300 locations contains 90,000 origin-destination elements before depot locations are added. Repeated requests, caching permissions, geographic clustering and provider limits affect actual consumption. Obtain a usage estimate from the proposed architecture instead of budgeting one call per delivery.

Will optimization pay for itself?

Build a baseline before assigning a savings target. Compare kilometres, paid overtime, planning effort, on-time deliveries and unassigned stops on equivalent workloads. Keep a record of traffic conditions, parcel mix and delivery windows.

Suppose the fictional fleet averages 110 kilometres per vehicle per operating day. Monthly distance is 25 × 110 × 26, or 71,500 kilometres. A pilot-observed 6% reduction would save 4,290 kilometres. At an assumed avoidable vehicle cost of SAR 0.60 per kilometre, that is SAR 2,574.

Suppose dispatch planning also frees 39 hours monthly at an assigned value of SAR 60 per hour. That adds SAR 2,340 in capacity value. Combined value is SAR 4,914 against SAR 2,700 incremental software expense.

The resulting SAR 2,214 monthly balance would imply roughly 28.5 months to recover a SAR 63,000 build. The dispatch time may improve capacity without reducing payroll, so show cash savings separately from time value. These assumptions describe a calculation method, not a promised return.

A faster route that misses more customer windows may have worse business results. Include service quality and exception handling when accepting a pilot, even if the distance figure improves.

How should the pilot run?

Start with historical or shadow planning. Dispatchers review proposals while current operations continue. Investigate omitted stops and unusual travel estimates. Then release a limited set of reviewed routes to a small driver group.

Save the input snapshot, configuration version and published plan identifier. If an operator overrides the plan, record the reason. Separate approved routes from drafts so an incomplete replan cannot silently replace work already in progress.

Define a safe fallback for service timeouts. The dispatcher should still see orders and the last approved assignments. Changes to started deliveries need explicit coordination. Publish a plan only after checking that its order set and vehicle availability remain current.

Acceptance case Expected behavior
Load exceeds capacity Clear exception or valid reassignment
Mandatory window cannot be met Order remains visible with a reason
Provider times out Current approved work stays usable
Dispatcher edits a draft New version is reviewed before publishing
Driver is already serving a stop Replan respects locked work
Two planners publish concurrently Conflict is detected and resolved

Questions to ask before approving a proposal

Ask vendors to price the same representative order file and list their assumptions. Include the busiest day, narrow windows and a sample of address corrections. Request separate charges for optimization, mapping, support and any existing-app repairs.

Ask who owns provider accounts, how quotas are monitored, how a failed job is recovered and which constraints the team demonstrated. Confirm that a dispatcher can explain why an order is unassigned. Also ask how users can roll back a configuration change and who supports the system during operating hours.

Frequently asked questions

Can a 25-vehicle fleet begin with a managed service?

Yes, when its required constraints fit the service. Run representative orders through a prototype and inspect the exceptions before approving the implementation budget.

Does adding more vehicles always increase development cost?

Vehicle count affects usage and performance testing. New constraint types, multiple depots and integration gaps often add more implementation work than a simple increase in fleet size.

Do we need machine learning?

A constraint-based optimizer can handle many routing requirements. Forecasting travel or service duration may become a separate project if measured input errors justify it.

What should be approved first?

Approve a small data and constraint audit with a prototype. It gives the delivery team a concrete order model, a shortlist of unresolved questions and a budget grounded in the existing application.