Openreach puts Google Cloud AI into fibre and fleet planning—execution remains the test

Openreach is using Google Cloud AI and analytics to improve full-fibre planning, field-service efficiency and fleet emissions management.

• Openreach says Google Cloud tools now support three different workflows: modelling fibre-build routes, analysing a large vehicle fleet and assisting data engineers. None of those functions transfers control of the physical network to Google.
• The headline figures—24,000 vans, almost 200 million annual miles, about 10,000 tonnes of CO2e and a greater-than-50% reduction in time to insight—are company claims. They should be tested against build output, fleet fuel and mileage, charging availability, reliability and an explicit carbon baseline.

The announcement concerns decision support, not automated construction

The joint announcement says Openreach is using Vertex AI, BigQuery and Gemini Enterprise across fibre planning, fleet operations and data-engineering work. The important distinction is between recommending a route or vehicle allocation and delivering it. Openreach remains the network builder and wholesale-access operator; Google Cloud supplies computing, analytics and AI services.

That boundary matters because a model cannot grant a wayleave, clear a blocked duct, obtain a permit, secure a contractor, install fibre safely or make a property ready to order. Better information can shorten investigation and prioritisation. Whether it shortens the end-to-end build cycle is a separate outcome that needs measurement.

Three workflows carry three different claims

For the fleet, Openreach says 24,000 vans travel almost 200 million miles a year. BigQuery geoanalytics combines telematics, routes and charging availability to identify suitable electric-vehicle placements, excess travel, idling and possible maintenance needs. The statement attributes about 10,000 tonnes of annual CO2e reduction to additional electric vans enabled by this work. That is an Openreach/Google estimate, not an independently demonstrated causal result in the announcement.

For fibre planning, a Vertex AI digital twin combines records for 35 million homes and businesses with road, rail, waterway and existing-network data. That describes the input universe, not 35 million premises newly passed, connected or accelerated. Its value depends on address quality, asset records, constraint data, planner adoption and how often recommendations survive field survey.

For engineering, Gemini Enterprise is said to convert complex legacy queries into production-ready BigQuery code and reduce time to insight by more than 50%. This is a data-work measure. It does not establish a 50% improvement in construction time, premises passed, customer connections or fault performance.

Independent context separates coverage from adoption

BT Group's 2026 annual report says Openreach passed 4.8 million more premises during the year, bringing its full-fibre footprint to 23 million, including 6.3 million rural or hard-to-reach premises. It also reports take-up above 38% and 6,900 electric vans. These figures show scale and progress, but the report does not isolate how much of the build rate or fleet change was caused by the Google Cloud tools.

Ofcom's 2025 Connected Nations report found full fibre available to 78% of UK residential premises and take-up at 42% of premises where it was available. Ofcom's UK-wide totals cover multiple networks and use a different date and population from Openreach's corporate metrics. They are useful precisely because they show that premises planned, passed, ready to order and actively connected are not interchangeable.

Openreach's own build information says plans can change and some properties remain too complex to upgrade commercially. That is the execution surface the digital twin must improve: not a map alone, but accurate choices under civil-engineering, access and investment constraints.

The carbon claim needs a visible denominator

The annual report records 6,900 electric vans; a later Openreach fleet update records 7,000 electric vehicles in a fleet of about 23,000, more than 4,000 home or site chargers, and says roughly one in three engineers cannot install a home charger. The difference from the announcement's 24,000 vans may reflect timing or scope. It should not be silently normalised.

To assess the 10,000-tonne claim, reporting should expose the counterfactual diesel mileage, vehicle and electricity emission factors, period, fleet boundary, charging mix, avoided fuel, embodied-vehicle treatment and uncertainty. Route optimisation and electric-vehicle deployment are distinct mechanisms. The relevant operating indicators include total mileage, fuel, electricity, idle time, vehicle availability and maintenance, not only the number of electric vans.

AI creates a governance surface as well as an efficiency opportunity

The systems join vehicle telematics, property and transport records, broadband assets and legacy engineering logic. Openreach therefore needs access controls, data lineage, retention rules, quality checks, model and code review, exception handling and rollback. Gemini-assisted code should be tested like other production code; digital-twin recommendations should be compared with field outcomes and corrected when local conditions differ.

ITPro's report confirms the deployment and quotes the same company metrics, while Google Cloud's later summit account describes Openreach as a field-service optimisation customer. Neither supplies a controlled comparison of sites, routes or vehicles. The defensible conclusion is that Openreach has embedded Google Cloud tools into material decisions. The proof of faster fibre and lower emissions will be a repeatable change in operating results, with methods and baselines visible.