AI Consulting

Where we look for AI use cases in sales and eCommerce

We start from tasks, not tools. With each department we go through the high-volume, low-value activities and the decisions that are currently made without data: interpreting orders that arrive in different formats, searching a large catalogue using the customer's words rather than the internal SKU, preparing answers to frequent customer service questions, or summarising an account's history before a sales visit.

That review produces an inventory of candidates described in the same format: which task they cover, who does it today, how often it is repeated and which systems are involved (store, ERP, CRM, PIM).

We also include uses that are already happening informally. It's common for part of the team to use AI tools on their own to draft texts or summarise documents; knowing about them helps identify real needs and bring that use under control.

The goal of this stage is not to find as many cases as possible, but the few that combine enough volume, available data and a process that already works well without AI.

How we evaluate an AI use case

Each candidate is assessed against specific criteria: volume and frequency of the task, cost of an error, availability of the data needed, whether a person can supervise the result, and the integration effort with existing systems. The cost of an error carries a lot of weight: an imperfect product description is not the same as a wrong price on a confirmed order.

The result is a sheet for each case with a clear recommendation: run a pilot, wait until certain conditions are met, or drop it. Ruling out cases with solid reasons is as useful as choosing them.

The assessment is done together with the heads of each department, because they know the real cost of an error in their process and they are the ones who will have to trust the result. For the cases that move to a pilot, we define in advance what will be measured, with which sample and what result would be needed to continue; without that prior criterion, pilots tend to drag on without anyone deciding.

The data an AI project needs

The quality of the result depends directly on the starting data: complete product records, a consistent order history, well-logged customer service enquiries or accessible documents. Often the first project isn't an AI project at all, but organising the catalogue in a PIM or cleaning up the customer master in the ERP.

We also review which personal data is involved, where it would be processed and what terms each model provider offers on the use of that information, so that the use case is compatible with data protection by design.

If a use case needs data that isn't collected today, such as the detailed reason for returns, starting to record it in a structured way may be the first recommendation, even if the benefit comes months later.

Risks and governance of AI in the company

The main risks are well known: wrong answers that look right, dependence on a single provider, usage-based costs that are hard to predict, and employees using tools on their own with company data. None of them rules out using AI, but all of them require rules.

We propose a simple governance framework: who approves a new use case, which data may and may not be used, which decisions are never automated (prices, credit terms or commitments to customers) and how what each solution does is logged and reviewed. That framework also takes into account the obligations under the EU AI Act that apply to each case.

Governance also covers what the people who will work with these tools are told: what they can expect from them, how to review their output and what to do when they spot an error.