
The pitch for route optimisation is consistent: ingest the day's deliveries, apply the optimiser, save 12–25% on fleet cost. The demonstration is compelling. The pilot at a single depot confirms the saving. The rollout follows. Twelve months later, the operating P&L has not moved. Understanding why is more useful than the original pitch.
The three places the saving disappears
- The pilot was measured against the planned route, not actual driver behaviour — and actual behaviour was already 30% more efficient than the plan, with experienced drivers making informal optimisations the system never captured
- The optimiser was accurate on Monday and wrong by Friday because the underlying inputs — customer time windows, traffic conditions, parking restrictions — were assumed rather than measured and updated in real time
- The dispatcher could not override the optimiser in a straightforward way, so they routed around it entirely; adoption looked high in the system and was effectively zero in practice
What separates the projects that hold the saving
Three patterns repeat across the implementations that deliver sustained results. The baseline is measured against actual driver behaviour over a representative window, not the legacy plan. The optimiser ingests live operational signals and improves week over week rather than running on static assumed inputs. And the dispatcher remains the authority over exceptions — the system surfaces the cost of every override without blocking it.
Route optimisation technology has matured considerably. The economics of route optimisation projects have not followed at the same pace. The buyers who realise the saving are those who insist on the right baseline before they sign, and the right operational integration before they roll out beyond a single site.
More insights

Where enterprise AI creates durable business value
Most enterprise AI programmes fail not because of the technology but because of how organisations approach them. The firms capturing lasting value share three disciplines that have nothing to do with model performance.

The network operations centre is a strategic asset waiting to be unlocked
Telecom operators have invested heavily in network infrastructure. The returns on that investment are increasingly determined not by the infrastructure itself but by the intelligence layer running on top of it.

AI strategy in high-growth markets: lessons from the Gulf
Governments and enterprises across the Gulf are committing significant capital to AI transformation. The organisations seeing results are those that treated AI as a strategic management question before treating it as a technology question.

Rethinking enterprise security: from access control to risk intelligence
The enterprise security function is evolving from a compliance obligation into a source of operational intelligence. The transition requires more than new technology — it requires a different conception of what security is for.

The data architecture decisions that determine transformation outcomes
Most organisations approach data architecture as an IT infrastructure question. The firms that extract the most business value from their data treat it as a strategic design question — one that belongs in the boardroom, not just the server room.

AI in financial services: separating the durable from the hype
Financial institutions have been among the earliest and most enthusiastic adopters of AI. They have also produced some of the clearest illustrations of where AI programmes go wrong and what it takes to make them last.

Building a customer intelligence capability that actually changes decisions
Retailers collect more customer data than ever. Most of it sits in systems that don't talk to each other, analysed by teams that don't talk to merchants. The result is insight that arrives too late to matter.

How to tell if your organisation is actually ready for enterprise AI
AI readiness is not a technology question. It is a diagnostic across strategy, data, governance, talent, and operating model. Most organisations that think they are ready are not — and the signals are detectable before any money is spent.

AI in healthcare: the organisational challenges matter more than the technical ones
Healthcare AI projects fail at a rate that should give every executive pause. The failure mode is rarely the algorithm — it is the operating environment the algorithm is dropped into without adequate preparation.

Energy transition strategy: the grid challenges that determine the pace
The energy transition story is usually told as a generation story. The harder problem — and the one operators are investing the most in — is grid flexibility: matching variable supply and variable demand in real time across infrastructure built for neither.

From smart city pilot to permanent capability: what changes at scale
Smart city pilot projects are familiar. Smart city permanent operating capabilities are considerably rarer. The difference between cities that make the transition and those that do not comes down to a small number of decisions made at the start.

Government digital transformation: the implementation gap
Government digital programmes report high levels of service digitisation. The metric that matters — end-to-end completion rate — tells a different story. Most programmes have not yet made the measurement straightforward enough to be honest about it.

Quality intelligence in manufacturing: the case for moving upstream
Real-time quality inspection technology has matured. The constraint on returns is no longer the model — it is the operating environment, the taxonomy, and the discipline to keep the system trustworthy through normal manufacturing variation.

Insurance claims transformation: speed, accuracy, and the fraud trade-off
Claims automation has a sound business case. It also has a set of trade-offs that are routinely underweighted in the evaluation process. A candid note for insurance executives assessing the move.

ESG performance reporting: from compliance exercise to strategic data asset
ESG reporting is moving from voluntary disclosure to assurance-grade obligation. Most enterprises are not ready, and the reason is rarely policy — it is the data layer underneath the report.

Data governance under AI regulation: the four decisions that matter most
Data governance was a back-office concern until AI made every production model a regulated artefact. The four decisions that determine whether an enterprise's governance approach holds up under supervisory scrutiny.

MLOps in regulated industries: the discipline gap in enterprise AI delivery
MLOps reference architectures are well documented. None of them describe what changes when the model is regulated and every promotion to production needs documented approval from a named accountable person.

Five questions every board should be asking about their organisation's AI programme
Boards have moved past asking whether the enterprise should use AI. The question is now what the enterprise is exposed to by using it, and whether management can demonstrate the answer is known and managed.

AI-augmented defence: how the threat landscape is changing for enterprise security
AI in the security operations centre tends to fall into one of two patterns: the kind that converts existing alert noise into actionable signal, and the kind that adds more noise. The distinction is detectable before you sign the contract.
