The EU AI Act

Compliance you can check, not a badge

Plenty of vendors will tell you they are "AI Act compliant." That sentence on its own is worth very little: obligations under the Act depend entirely on how a system is classified and on whether you are the provider or the deployer. So instead of asking you to take our word for it, this page states our position in terms your procurement and compliance teams can test.

Bull is the provider of Shelf Inspector. A retailer or brand running it in their own stores is the deployer. Both roles carry obligations, and they are not the same ones.

What the system is for

Shelf Inspector's intended purpose is narrow and deliberately so. It analyses photographs of retail shelving to assess:

• Whether the products that should be on a shelf are present, and where they sit
• Shelf share and facing counts against an agreed layout
• Out-of-stock and out-of-shelf situations
• Compliance with a planogram
• Consistency between price tags and the prices that should be showing
• Whether promotional material has been put up as specified

It reads shelves, not people. The subject of every assessment is merchandise and its arrangement. The system does not identify individuals, infer anything about them, or produce an output about a person.

Risk classification

We classify Shelf Inspector as not high-risk under the AI Act, and it is not a prohibited practice under Article 5. The basis is straightforward: assessing the physical arrangement of goods on a shelf does not fall within any of the eight categories listed in Annex III, and the system is not a safety component of a product covered by Annex I.

The boundary that matters

One Annex III category deserves to be addressed head-on rather than left implicit. Category 4 covers AI used in employment and workers management — including systems used to monitor and evaluate the performance and behaviour of people at work. An AI system that scored individual employees would sit squarely inside it.

Shelf Inspector reports on locations, not personnel. Its outputs describe the state of a shelf in a store at a point in time. It is not designed to assess, rank or monitor the people who work there, and using it that way falls outside the intended purpose we place it on the market for. We are explicit about this because the distinction is what keeps the classification correct, for you as much as for us.

Prohibited practices

Article 5 prohibitions have applied since 2 February 2025. Shelf Inspector engages none of them. It performs no emotion recognition, no biometric categorisation, no social scoring, and no subliminal or manipulative techniques. Emotion recognition in the workplace is specifically prohibited, and we want to be unambiguous that nothing of the kind is present here.

People who appear in photographs

Shelves are photographed in working stores, so shoppers and staff can pass through the frame. They are incidental to what the system does — no assessment is made about them, and no attempt is made to detect, match or recognise them. Handling of those images is governed by data protection law and by the agreement with each customer, which sets out retention and access on a deployment-by-deployment basis.

People stay in the loop

Shelf Inspector produces findings, not verdicts. Scores, gaps and trends go to the people who run the stores, and it is those people who decide what — if anything — happens next. The system takes no action on its own, and there is always a photograph behind every finding, so any result can be checked by a human against the evidence it came from.

AI literacy

Article 4 has applied since 2 February 2025 and requires providers and deployers alike to ensure staff working with AI systems have a sufficient level of AI literacy. It is an obligation that lands on your organisation as well as ours. We support it by documenting what the system does, what it does not do, and where its outputs should not be relied on without a human look.

Where the Act stands

• 1 August 2024 — the AI Act entered into force
• 2 February 2025 — prohibited practices and AI literacy obligations began to apply
• 2 August 2025 — obligations for general-purpose AI models began to apply
• 2 August 2026 — the remainder of the Act, including obligations for high-risk systems under Annex III, begins to apply
• 2 August 2027 — obligations begin for high-risk systems that are safety components of regulated products

Questions from your compliance team

If you are evaluating Shelf Inspector and need detail beyond this page — documentation of the intended purpose, the basis for the classification, or how the above maps onto your obligations as a deployer — ask, and you will get a straight answer from someone who knows. Contact us through this site, or reach the Bull Data Protection Office at dpo-bull@bull.com for data protection questions.