AI in Production Depends on Data Readiness
Mon 13 Jul 2026
Across London, Frankfurt and Paris, this month’s AI in Practice coverage is centred on one practical question: what needs to be in place before AI can move beyond pilots and into production?
The answer this week is data readiness. Not simply whether organisations have enough data, but whether that data is trusted, governed, accessible and understood well enough to support AI at scale.
That question takes a different shape in each market. In the UK, the focus has been on the point where weak data foundations begin to undermine AI pilots. In France, the discussion has moved to what happens when those pilots are pushed into production. In Frankfurt, the five practical checks were discussed, offering a practical list for professionals to consult before deciding to deploy AI.
Data Readiness is a Business Issue
The clearest signal comes from Zoho’s Digital Health Study 2026.
It found that 97% of businesses with good digital health reported ROI from AI data insights, compared with 45% of businesses with poor digital health.
The gap points to a simple reality: AI performance depends heavily on what already exists underneath it. Data quality, governance, ownership, and access all shape whether a model can produce useful results.
Sachin Agrawal, Managing Director of Zoho UK, said organisations often discover they are not ready for enterprise AI when pilots stall and fail to deliver measurable impact. In many cases, the issue is not the AI system itself, but the data estate supporting it.
Production Exposes the Execution Gap
That same issue surfaced in Lyon, where data and AI leaders from Michelin, Interflora, and Relyens discussed the realities of putting AI into production.
The conversation moved quickly beyond model capability. Shared data definitions, user adoption, governance, measurable business value, and the willingness to stop weak use cases all emerged as practical conditions for success.
Alexandra Lorenzo de Brionne, Lead Data Scientist at Michelin, highlighted the challenge created when different parts of an organisation interpret the same data differently. Without shared definitions, metadata, and clear governance, technically promising projects can still fail to produce usable outcomes.
Vincent Faburel, Head of Data & Backend Products at Interflora, added another important point: maturity can also mean walking away from a project when the evidence shows it is not creating enough value.
Frankfurt Highlighted the Basics of Data Readiness
Frankfurt’s insight page returned to five practical checks:
Do you know where the data lives?
Can you trust it?
Is it organised?
Is governance keeping pace?
Are you solving a real business problem?
The takeaway: data needs to be reliable enough for AI to operate, though it doesn’t have to be flawless. In essence, AI reveals the existing gaps.
The Harder Work Sits Behind the Model
The picture emerging from these regional discussions is not about a shortage of AI capability, but rather the operating conditions that surround it.
Reliable data, clear ownership, appropriate permissions, shared definitions, and measurable outcomes matter long before a model is deployed. They are less visible than a new tool launch, but they soon determine whether an AI project remains a pilot or becomes part of day-to-day operations.
As the AI in Practice programme continues across London, Frankfurt and Paris, data readiness is emerging as one of the clearest indicators of where AI adoption succeeds and where it stalls.
The technology may be advancing quickly. The organisations that make progress will be the ones that can turn that capability into repeatable, measurable value.
