Unreliable plumbing route optimization is usually a data and dispatch-design problem—not proof that AI cannot work

An AI route optimizer can calculate an efficient path and still create a terrible day for your technicians. Plumbing dispatch is not a traveling-salesperson problem in which every destination is equally ready, every stop takes a predictable amount of time and every worker can perform every task. A useful schedule has to reflect the operational reality behind each address: whether the customer is available, whether the job description is accurate, which technician has the right experience and equipment, how urgent the call is and whether earlier work is running long.

Wild goose chases often begin before the optimizer runs. A mistyped address, a service request labeled too broadly, an unrealistic duration or a duplicate appointment gives the system a false picture of the day. The algorithm then optimizes bad inputs with complete confidence. Dispatchers may compound the problem by changing assignments through phone calls or text messages without updating the central schedule, leaving the software to plan around jobs that are no longer current.

Orange County adds the usual complications of a dense service area: conditions differ by time of day, nearby addresses are not always quick to travel between, and an urgent call can disrupt an apparently efficient sequence. The important lesson is that geographic closeness is only one variable. If your wider concern is how to evaluate outside help rather than another off-the-shelf tool, use a disciplined process for finding a company to build a custom AI solution.

Do not begin by tuning dozens of settings. First establish whether the schedule supplied to the system represents what your team actually knows. Pick a recent group of problematic dispatches and reconstruct each one from intake through completion. Compare the original customer request, the address entered, the assigned technician, the estimated duration, the route recommendation, every dispatcher intervention and the final result.

This same discipline applies beyond routing. If automated estimates are also creating operational risk, address AI estimates that keep underbidding construction work as a separate workflow. Do not assume that fixing dispatch data will automatically correct estimating logic; the inputs, consequences and review controls are different.

The phrase “the AI is unreliable” is understandable, but it is too broad to guide a repair. Divide each bad dispatch into four layers: source data, business rules, optimization logic and field execution. A wrong address is a source-data failure. Assigning a technician outside the intended service area is probably a rule or constraint failure. Repeatedly choosing routes that look efficient but collapse under normal traffic points toward optimization assumptions. A technician receiving an old itinerary after dispatch changed it is an execution or synchronization failure.

Keep a plain incident log and assign every failure to one primary category, even when several contributed. Record what the optimizer recommended, why the team rejected or changed it and what happened next. Patterns will emerge quickly enough to tell you whether the dominant problem is incomplete intake, poorly expressed operating rules, stale schedule information or weak routing performance.

Avoid using a single dramatic dispatch as the entire case against the system. Equally, do not let a vendor dismiss repeated failures as isolated user error. A structured log turns frustration into evidence. The same approach is useful when evaluating AI material predictions that require constant correction: isolate the bad input, rule or forecast before adding more automation.

Route software often presents an optimized map, projected travel time or utilization figure. Those displays are useful diagnostics, but they do not prove that the schedule served customers or improved the day. Your scorecard should connect recommendations to field outcomes. Track avoidable mileage, jobs completed as planned, late arrivals, repeat visits caused by assignment problems, overtime, dispatcher interventions and technician reassignments.

Establish a baseline from the current operation before changing configurations or vendors. Then make one meaningful change at a time and compare the same measures. If you alter job categories, service-time assumptions, territories and dispatch policies simultaneously, an improvement or decline will be difficult to explain. This matters whenever leaders need defensible evidence from AI operations, including cases where a board is demanding proof of savings from an AI system.

The immediate choice is not simply “keep AI” or “remove AI.” You can use the system as a recommendation engine, automate only stable portions of the schedule, or allow broader automation with exception rules and human oversight. The right level depends on the quality of your data, the volatility of daily work and how easily dispatchers and technicians can identify and correct a poor recommendation.

Do not automate exceptions merely because they consume dispatcher time. Exceptions are where the highest uncertainty and operational risk often live. Start with repeatable jobs that have dependable intake information and realistic duration histories. Keep ambiguous calls visible to a person until the business has a reliable way to classify them.

The same principle protects customer relationships in adjacent workflows. For example, AI can support speed and consistency without replacing judgment or personal communication. If pricing is the next concern, consider using AI for plumbing quotes without losing the personal touch. If several workflows are failing independently, step back and evaluate one partner for AI strategy and implementation instead of purchasing another isolated tool.

Once you have identified whether the real problem is data, workflow design, decision rules or the optimization itself, you can have a more productive AI conversation. OceSha Ventures and its AI-first work builds and operates AI-first solutions for businesses and organizations, including course creation, branded academies, AI assistants such as Lumi and business intelligence. That background is relevant when your challenge extends beyond one route map into how business information, customer inquiries and operational intelligence should work together.

Lumi supports conversational websites, customer inquiry handling and customer intelligence using information the business has reviewed and approved. This can help an organization make confirmed information accessible through an AI assistant, but route optimization should still be assessed as its own operational system. The practical starting point is a clear incident record, a defensible baseline and a prioritized description of the decisions that need improvement.

Rohan Hall has built AI and blockchain systems and has worked extensively across the United States, Europe and Asia. His professional technology career began in 1984, and his work includes technology strategy, architecture and emerging technologies. You can explore Rohan Hall’s ventures, book and broader work or examine whether an AI initiative is also connected to the wider problem of turning business expertise into discoverable answers.

For a broader treatment of AI alongside blockchain, cryptocurrencies, neuromorphic technologies, cognitive intelligence and other emerging fields, Rohan covers the subject in depth in The Convergence of AI and the Top 10 Emerging Technologies. That wider perspective is useful when leaders need to distinguish a fashionable technology purchase from a system that is designed around reliable business decisions.