Really scaling AI as a retailer? Get these 4 challenges out of the way!
Experimenting with AI, that's what retailers are doing in abundance. But as soon as they want to scale up, they often get stuck on several fronts at the same time.
Take customer service, for example. The menu of options that we are all used to ("Choose 1 for a question, choose 2 for a request...") will soon disappear. You are increasingly talking to AI, which helps you directly or connects you to a human when needed.
Many retailers see such a change as an extra feature on top of what is already there. And therein lies the misunderstanding. Scaling up is not a matter of adding features to an existing process. What really makes progress is overhauling your process and creating a new foundation. Four challenges illustrate why!
The 4 challenges that stand in the way of scaling up
1. You don't know exactly what AI will solve for you
What can you actually do with AI agents? What can the technology do and what bottlenecks can you solve with it? At its core, this is a leadership issue. It's essential that the CEO, CTO, and other C-level executives agree on where AI makes a difference. From there, you can roll out targeted initiatives that actually add value to the organization. If you don't think about this at the leadership level first, you run the risk of using separate use cases to create fantastic solutions to problems that you don't really need to solve.
2. Your processes are not yet AI-first
In retail, the profit is in the process from A to Z: you buy something, you ship it, you sell it. As long as you 'shove' AI somewhere in that process, little will change. You want to get rid of the question that most organizations are currently asking: 'How do we make our existing processes cheaper with AI?' Instead, it is important to ask yourself: 'In which actions does humans really add value and what can AI handle independently?'
In that context, it is important not to think of an AI agent as a tool that makes a small step of an existing process easier for a team member. For example, an agent who independently adjusts pricing or stock is a team member who gives you a place in your organization.
3. Your data is there, but 'usable' is different
Retail organizations that know how their data is built can get value from it. Do you measure in your store which brands someone always picks up first? Or which products go off the top or bottom shelf? Then you can analyze that with AI and base predictions on it. The problem: according to The Thinking Company , only a small fraction of omnichannel retailers have a unified data layer that can feed AI models in real time. The rest have data in abundance, but can't do anything with it at scale. This creates data silos that block agentic AI.
4. Compliance risks may arise that you don't immediately see
When you use AI, data can be processed by third-party vendors, including U.S. tech companies. This is not automatically prohibited, but you do need to know what data is shared, where it is processed and what legal agreements and safeguards have been arranged. In the case of personal data, this is subject to the GDPR rules for international data transfers, among other things. This also applies within existing business environments. For example, since July 24, 2026, models managed directly by OpenAI have been available by default for eligible Microsoft 365 environments, unless an administrator adjusts the settings. In a previous blog, we explained what this change means for organizations that use Microsoft 365 Copilot and which settings they need to check.
It is expected that a large part of the obligations under the EU AI Act will apply from 2 August 2026. Exactly which rules apply depends on the way you use AI and the risk involved. Violations of the AI Act, the GDPR or consumer law can lead to high fines.
For retailers, this is relevant, for example, in customer profiling and personalized pricing. Where an online price is personalised for an individual consumer on the basis of automated decision-making, that consumer should be clearly informed. Organizations must therefore have insight into what data and algorithms they use and be able to substantiate that their application complies with the relevant legislation.
Four challenges, one solution: lay your foundation
It may seem as if the challenges discussed have little to do with each other. But in practice, they all point to the same underlying problem: a foundation that is not yet AI-ready. Put that in order (with data governance first) and you will solve all four.
Does this mean that you have to tackle everything at once? Absolutely not. First, identify the critical data assets for your key AI initiative. Assign data stewards for those domains and improve data quality for one high-impact use case. This is how you prove the value of governance and create the space to scale further. And that brings you step by step to the future-proof AI-first retail organization you want to be.