More than 75% of industrial AI pilot projects never reach large-scale deployment, and over 80% of manufacturers cannot extend AI beyond isolated use cases, according to a report by Everest Group and Capgemini Engineering — with the failure attributed not to weak algorithms but to factory systems that cannot move data between each other.
The report, Operationalizing Adaptive AI Across the Manufacturing Shopfloor, was relayed by Moroccan business publication Challenge, which applied its findings to a country whose industrial base makes them unusually relevant.
The diagnosis
The obstacle, the report argues, is architectural. In many factories, IT systems including ERP and PLM platforms operate separately from OT environments — MES, SCADA, controllers and production equipment — leaving data poorly connected or arriving too late to support automated decisions.
Nicolas Rousseau, head of engineering technologies at Capgemini Engineering, put it plainly: digital ambition is not the constraint in manufacturing today. Execution is.
Four obstacles are identified: rigid processes, legacy technology, fragmented data and a shortage of cross-functional skills.
That diagnosis will be familiar. PwC’s survey of South African mining found more than 85% of respondents rating their data management capabilities average or poor, with one chief executive observing that “data doesn’t help if it’s just data. It needs to be put through the funnel to get to wisdom.” Nearly 70% of mines rated their AI readiness poor or very poor, and identified skills rather than technology as the binding constraint.
Different sector, different country, same conclusion.
Infrastructure first, applications second
The report’s prescription reverses the usual sequence. Rather than starting with algorithms, it recommends manufacturers first build architecture capable of connecting equipment and structuring data — through a Unified Namespace, the MQTT protocol for local connectivity, and in some cases streaming platforms such as Kafka. The rollout can be progressive, starting with a single production line before expanding.
Hybrid edge-cloud architecture is the second element, processing data close to equipment to cut latency and reduce dependence on constant cloud connectivity. Challenge argues this suits Moroccan manufacturers working with ageing infrastructure and technical debt — outdated data formats and proprietary protocols.
The report also warns against rushing to full automation. With nearly half of manufacturers still struggling to integrate IT and OT, deploying autonomous agents onto ageing infrastructure risks adding complexity rather than removing it.
As operational systems open toward cloud and AI applications, cybersecurity and compliance become part of the architecture rather than an afterthought — a point Challenge notes carries weight for Moroccan manufacturers handling sensitive data for international clients, alongside growing interest in sovereign AI offerings from major cloud providers.
Its final observation is that no single vendor controls the full technology chain, with industrial platforms, cloud providers and AI infrastructure specialists occupying complementary roles.
Why Morocco
Morocco has more industrial base to modernise than most African markets. It is the continent’s largest automotive manufacturer, with Renault, Stellantis and BYD operations anchored around Tanger Med, alongside significant aerospace and phosphate processing sectors. Much of that capacity was built or expanded over the past two decades and modernised incrementally — machinery and software added without a full architectural rethink, which is precisely the condition the report describes.
The country has also been building AI capability at pace. The $1.2 billion Nexus AI Factory is under construction outside Casablanca, chosen over South Africa. The AI Made in Morocco initiative targets a 100 billion dirham GDP contribution and 50,000 AI jobs by 2030. UM6P and Syensqo launched a joint AI materials lab. France’s Onepoint is building an AI centre of excellence focused explicitly on industrial and administrative applications. The Al-Jazari Institute of Industry X.0 was established for exactly this purpose.
What the Everest Group findings suggest is that none of that guarantees adoption on the factory floor. The gap between national AI investment and operational deployment runs through data architecture — and no amount of compute or strategy substitutes for systems that can talk to each other.
That is the same argument Roger Jantio made in these pages last month, from the investment side: Africa does not suffer from a shortage of pilots, but from too few pathways converting them into working operations.