Logistics: optimizing routes without changing software

When routes cost too much, the reflex is often to evaluate new software. In most transport and distribution SMEs, that is not the right first move. The tool already in place, whether a TMS, an ERP module or a dedicated solver, can usually do more than it is asked to. This guide sets out six steps to get the most from it before deciding anything. It reflects our approach to operations and processes and applies to transport and logistics companies in Switzerland, Europe and the US alike.

Why the software is rarely the problem

A routing solver computes the best possible plan from what it is given. If addresses are approximate, stop times are averages and constraints have been frozen for years, the plan is optimal on paper and unworkable on the road. The planner then corrects it by hand, the tool becomes a rough starting point, and the conclusion is that the software does not fit.

This is a general principle at work: automating a bad process only makes it faster. New software fed with the same data and the same implicit rules would typically produce the same results, after an expensive migration.

Step 1: measure the starting point

Without a baseline, no gain can be demonstrated. Over four representative weeks, collect a few simple indicators:

• kilometres per stop and stops per route;
• fill rate by weight and by volume;
• gap between planned and actual route duration;
• deliveries outside the promised window, and empty kilometres;
• overtime hours per vehicle.

These figures usually come from TMS exports, telematics and payroll. Their consistency matters more than their precision: they will be used to compare before and after.

Step 2: clean the master data

This is the least visible step and the most profitable. Four workstreams come up almost every time:

• Addresses. Verified geocoding, with the real delivery point (loading dock, rear access, pedestrian zone entrance) rather than the postal address.
• Service times. A time per customer or customer type, not a single average. Typically, the gap between a customer with a dock and a city-centre shop with no parking can be a factor of two to three.
• Vehicles. Real capacities in weight, volume and pallets, tail-lift availability, access restrictions.
• Time windows. The ones customers genuinely require, not the ones entered in a record years ago and never reviewed.

Step 3: review the tool's parameters

Many features stay unused because the initial configuration was never reopened. Go through:

• the distinction between hard and soft constraints (a window can often be exceeded by ten minutes at a penalty, instead of adding a vehicle);
• speed profiles by time of day and road type;
• breaks and driving times, under the Swiss ARV 1 ordinance or EU Regulation 561/2006;
• the trade-off between cost per kilometre and cost per hour, and whether the number of vehicles is fixed or adjustable by day;
• start and end locations, sometimes imposed by habit alone.

Step 4: understand what planners correct

Every manual change to a plan carries information. For four weeks, ask planners to record the reason for each change under a few categories: customer constraint not entered, local knowledge, driver preference, commercial urgency, data error.

Sorting is then straightforward. A legitimate rule ("this customer is never delivered after 10 a.m.") goes into the tool. A habit with no justification is discussed with operations. Either way, knowledge moves out of one person's head and into the system, which also reduces dependence on a key planner.

Step 5: close the loop with reality

A plan is only useful if compared with what actually happened. Each week, reconcile planned and actual using proofs of delivery and telematics: stop durations, delays, detours. A thirty-minute review is enough, provided it leads to corrected service times and constraints in the tool.

The logic is the same as in field technician routing: a realistic plan depends on a regular learning loop more than on the sophistication of the algorithm.

Step 6: act on commercial rules

Part of routing cost is decided outside operations: order cut-off time, minimum drop size, delivery days by zone, width of promised windows. Tightening a zone to two fixed days, or offering a wide window at standard price and a narrow one at a premium, can be worth more than any solver setting.

These are choices about service and margin, therefore management decisions. They belong to strategy and value creation as much as to logistics.

What it can represent: a worked example

Take a distribution company with the following assumptions, to be adapted:

• 12 vehicles, 220 operating days a year;
• a full cost of CHF 450 per vehicle per day (driver, fuel, maintenance, depreciation);
• hence about CHF 1.19 million a year in routing cost.

A 5 percent reduction in kilometres and hours, achieved through corrected service times, relaxed windows and better zone design, represents about CHF 59,000 a year. A few person-weeks of effort pay back quickly, with no additional licence. If the measurement in step 1 does not confirm a potential of this order, that is useful too: the issue lies elsewhere.

Where AI helps, and where it is not needed

The solver remains the right tool for building routes. AI can help estimate service times from history, anticipate delays during the day or spot recurring deviations. It is a targeted lever, useful when data is clean, which brings us back to steps 1 and 2. For the wider question of automated decision-making, see our thoughts on AI agents in operations. As for predictive maintenance on a fleet of a dozen vehicles, its value is often limited, as our analysis of when it pays off explains.

Two compliance points deserve attention. Driver geolocation is personal data processing: the revised Swiss FADP, in force since 1 September 2023, requires transparency, a clear purpose and proportionality, and the GDPR applies to activities in the EU. Separately, the EU AI Act, most of whose obligations apply from August 2026, treats certain tools that allocate tasks to workers based on individual behaviour as high-risk. A conventional solver is not normally concerned, but any feature that evaluates individual drivers warrants a check.

Where to start

Start small: one zone, one depot, four weeks of measurement. The aim is to find out whether the gap comes from data, parameters, habits or commercial rules before committing a budget. If you would like an outside view on this first reading, our diagnostic helps identify the priority levers in your routing and decide, with figures in hand, whether changing tools is truly justified.