Building AI that businesses truly own
Exclusive Q&A with Alex Haywood, CEO of MetisJean
You talk about moving from AI adoption to AI ownership. What does “owning” AI actually mean for a UAE government entity or company, and where do most organisations stand today?
In conventional AI use, each time a business user makes a request to a Large Language Model, the AI vendor gains valuable intelligence on roles, responsibilities and workflows – how information travels through your enterprise to achieve results.
Owning AI means that the business controls this intelligence and gives the AI vendor, whichever LLM you use, no visibility. It also means being able to change providers without losing your organisation’s knowledge or workflows.
Many organisations use AI for individual tasks today. The next step is integrating it into their wider operations while retaining control over their data and how the technology is used.
At MetisJean, we believe AI should build lasting capabilities, not just recurring costs. Jean is designed to run within an organisation’s own infrastructure, connect its existing systems and learn how the business works. If you change AI providers tomorrow, what do you keep? Genuine AI ownership means retaining your data, institutional knowledge and control over how AI operates.
Many UAE organisations already use externally hosted AI tools. What are the real risks of that, and how can they keep control of sensitive data without giving up the capabilities those tools offer?
The main risk is becoming dependent on an external provider for access to your own operational intelligence. Depending on the tool and its terms, sensitive data and prompts may be processed externally, potentially exposing business processes and internal knowledge. Organisations must also consider how unexpected changes to AI pricing, functionality or access could disrupt operations.
The answer is not to give up advanced AI, but to retain control over how it is used. Organisations should understand where data is processed, enforce existing access permissions, govern which models can handle sensitive information and ensure that outputs and actions can be audited.
Jean is designed to run within an organisation’s own infrastructure or a designated sovereign environment. To mitigate the risks posed by reliance on individual LLMs, we route queries across multiple models under policy, so no single provider ever holds the full picture.
The goal is to benefit from AI innovation without surrendering control over sensitive data, operational knowledge, or critical workflows.
As AI starts to inform decisions in business and public services, what does “verifiable and auditable” look like in practice? What should leaders demand from any AI system before trusting its outputs?
Leaders should demand evidence, not just a convincing answer. In practice, that means being able to trace an AI output to its original sources, understand what information and rules were used, see how confidence was assessed and review an audit trail of the system’s actions.
They should also ask how accuracy is tested, what happens when information is missing or conflicting, and whether the system respects existing access permissions. For consequential decisions, there must be clear human accountability and a way to challenge the output.
Jean is designed to ground answers in an organisation’s own data, while Synthia (the validation layer) tests accuracy against mathematically known ground truth. The principle is simple: trust should be earned through verification, not assumed because an AI system sounds confident.
Many AI pilots never reach full deployment. Based on your 30 years in automation, why do they stall, and how should organisations identify the use cases that deliver measurable productivity gains?
In my experience in automation, technology only delivers value when it solves a clearly defined operational problem. Pilots stall when organisations focus on demonstrating what the technology can do, rather than what the business actually needs. A good demo is not the same as a solution that fits into daily workflows and earns employees' trust.
I would start with repetitive, time-consuming tasks with a clear cost: finding information across systems, reconciling records, and preparing reports. Before piloting, set a baseline and agree on success measures, such as time saved or error rates. Involve the people who will use the system, test it against real workflows, and scale only once the evidence is clear.
How should UAE employers approach AI so that it empowers their workforce rather than unsettles it, and what should leaders prioritise in the next 12 to 18 months?
Employers should use AI to help employees work better, not just to replace tasks. AI can save time on repetitive work, information searches and report writing, allowing employees to focus on problem-solving, relationships and decision-making.
Over the next 12 to 18 months, employers should focus on three priorities: choosing use cases that deliver real value, improving employees’ AI skills and setting clear rules for its use. Success should be measured not only by efficiency, but also by service quality and how effectively employees can do their jobs.
