Trust, data and costs: the most frequently asked questions

No. If the AI layer runs through a cloud service, that is done under a business contract in which the provider does not use your data to train AI models. If your data may not leave the company, the model runs in your own environment. Your data stays yours and is only used to answer your own questions.

All traffic is encrypted. The assistant has no free access to your systems: it only works with the datasets prepared for your dashboards, and access keys are stored only on the server, never in the browser.

The set-up takes the GDPR into account from the start: only the data that is needed is used, personal data can be left out or anonymised beforehand, and agreements on the processing are recorded. Which measures your organisation needs, we look at together during the quick scan.

That risk is greatly reduced, because the AI does not make up the figures: the calculation is first done directly on your data, and only then does the AI explain that result in plain language. That way you can trace every answer back to the figures it is based on.

Because they build on each other. A stable process delivers consistent data, and consistent data makes AI answers reliable. Skipping a step means building on a shaky foundation.

A generic chatbot does not know your company and answers from general knowledge. This assistant is linked to your own structured data and knows the context behind your figures: departments, KPIs and exceptions.

No. If dashboards already exist, we build on them. If not, they are set up during the programme, from one central data model, so dashboards and the AI assistant always work with the same figures.

Yes. Some organisations start with process optimisation or with a dashboard, and add the AI layer later. The order stays the same, though: first the foundation, only then AI.

The tools are available to everyone today, often already in your office software. But such an assistant is only as good as the process and the data underneath. If your figures are scattered or the process is not stable, even the best tool gives answers that are wrong. That is where the work of DVD Intelligence begins.

That depends on the process and the starting point, so no standard figure is promised. The gains usually come from fewer errors and delays, less waste and fewer hours of manual reporting. In the quick scan, a first concrete business case is drawn up together.

No. The assistant takes over searching and reporting work, so your team has more time for analysis and decisions. The AI gives explanation and advice; you make the decisions.

For companies with a warehouse or distribution centre, in any sector: from chemicals to retail. The demo shows orders, stock, invoicing, staffing and customer reporting. The same set-up also works for other processes with recurring steps and measurable data.

An ERP records what happens, but rarely explains why a figure deviates. This approach brings the ERP data together with other sources in one data model, and lets you ask questions about it in plain language. Your ERP simply stays in use.

That AI only sees what is in its own package. In a warehouse, the figures are also in the TMS, the ERP, the planning and the time clock. DVD Intelligence brings them together in one data model and links actions to them. Your WMS simply stays in use.

A consultancy firm usually deploys three profiles: process, data and AI, each with its own day rate, plus overhead and margin, and meetings and hand-overs in which context gets lost. At DVD Intelligence that knowledge sits with one person: no hand-overs, one point of contact and one day rate, with a fixed price to start.

That depends on the scope and on the state of your processes and data. After the quick scan you get a plan with clear phases and decision points, so you know in advance what will be delivered when.

Mainly at the start, to map processes and explain the data. After that it is limited to fixed review moments and the training before go-live.

Usually not. We work as much as possible with what is already there, such as exports from your existing systems and your ERP. If something extra is needed after all, it is discussed in advance and included in the business case.

Through a separate training phase before go-live. After go-live there is a period of aftercare, so the new way of working really sticks.

Both, depending on what fits best. If your current software can support the AI layer well, we build on it. If your situation needs more control, a custom solution is set up. That choice is prepared in the quick scan and justified in the design phase.

Mapping how the work really runs today, making visible where waste, errors and delays arise, and then working out one standardised way of working that the team actually follows.

A fixed approach in five steps: define the problem clearly, measure, find the cause, improve and sustain. That way every improvement is backed by data and still stands after the project.

Yes. Any process with recurring steps can be improved, including administration, customer service, planning or HR.

The quick scan shows that. In three days we look at whether the process runs stably, whether your data is complete and reliable, and where the gains are. If your company is not ready yet, you will hear that too, along with what needs to happen first.

Yes. For example introducing a new warehouse system, onboarding a new customer or a re-layout. The project runs in fixed phases with a decision point after each phase, and the project manager follows it up until the new way of working runs stably.

Get in touch and briefly describe what is going wrong. You will receive a personal reply, with when someone can be on site.

A fixed moment at the end of each phase. Only when the agreed results are finished and approved do you decide whether the next phase starts.

Then the next phase does not start yet. Together we look at what is missing, and that is solved first. That way problems are not pushed to a later and more expensive moment.

Each phase has results and review moments agreed in advance. You always see what is finished, what is still running and which decision is next on the table.

No. The assistant does not search the internet and does not carry out actions: it only gives explanation and advice. Decisions and actions stay with your team.

Every programme is documented: a register of the AI applications, a risk sheet per application and a test report with fixed questions and pre-calculated answers. That way it remains demonstrable what the system does, even months later.

For a programme like this, usually not yet. The transparency obligations of the European AI Act have applied since August 2026; the heavy obligations for high-risk applications only start on 2 December 2027 and 2 August 2028. ISO/IEC 42001 is not legally required anywhere, but large customers and tenders already ask for it.

ISO/IEC 42001 is the international standard for an AI management system. AIGP is the certification for people who set up AI governance in practice. The documents you receive follow the structure of ISO/IEC 42001, so you can certify later without starting over.

Your organisation. The assistant gives explanation and advice; your team makes the decisions. The register shows per application who is responsible, who checks the answers and what happens in case of an error.

If a customer, auditor or tender asks how you manage AI, you put the register, risk sheets and test reports on the table. If the person who set it up leaves, someone else can take over. And if you want to certify later, that is an addition, not a fresh start.

That depends on your data and your rules. The quick scan looks at which data the AI would see, how sensitive it is and what your policy allows. Options include: a business cloud service with processing in the EU, a model on a server at your site, a dedicated server in a European data centre, or a combination.

Then the model runs in your own environment, or on a server rented only for you in a European data centre. Because the calculations are done in code and not in the language model, a smaller model than you might expect is often enough.

Yes. Confidential questions are processed locally, general questions through a stronger cloud model. Or the code sends only anonymised or summarised results to the cloud, never the raw data.

In the cloud there is no investment and you pay per use. Local requires an investment in hardware, from a few thousand euros for one department to a few tens of thousands for a larger company, plus power and management. You get a three-year comparison before you decide.

For free reasoning, the biggest cloud models are stronger. For well-defined tasks, such as questions about your own figures or summarising documents, a well-chosen local model often achieves the same result. Which model is enough is tested on your own questions.