A serious AI automation agency starts by measuring, not by stacking no-code scenarios dressed up as artificial intelligence. AI is already everywhere at your company: Claude, ChatGPT, Gemini, Copilot, in-house agents. Yet 30 to 50 % of AI initiatives fail to return what they should. Unlike a generalist AI agency, at Idun Group we analyse your real usage with our usage-analysis tooling, we estimate the ROI of each use case, then we tell you what to industrialise and what to cut. No process automation starts before the analysis has proven its value, and we are not the ones who build it: your teams, your integrators or a partner take it on. 62 % of companies are already testing artificial intelligence but only 23 % have deployed it at scale (Ifop-Converteo study, 2026). First step: a 30-minute discovery call.
of companies are already testing AI (Ifop-Converteo, 2026)
only have deployed it at scale
for the discovery call
A serious AI automation agency starts with the analysis of your company's real AI usage: what your teams already do with the tools in place, what it costs, what it returns. The real scope goes well beyond connecting applications to customer service. It covers mapping usage, estimating the ROI of each use case, recommending what to industrialise, consolidate or cut, through to choosing an autonomous agent for the few needs whose value is proven. That technical expertise is what separates a genuinely specialised agency from a reseller of preconfigured scenarios that automates without ever measuring.
A no-code scenario runs a chain of fixed rules. It breaks as soon as a field changes format or an unforeseen case appears. An autonomous AI agent reads the context and adjusts its decision accordingly, a capability that generative artificial intelligence alone does not provide, since it produces text without acting inside your systems. That structural difference explains why so many no-code automation projects plateau quickly. These workflows work for simple, repetitive tasks but collapse as soon as variability rises, holding back the growth the executive team expects.
The provider proposes agents without ever looking at what your teams already do with Claude, ChatGPT, Gemini, Copilot or your in-house models.
A savings percentage guaranteed before any analysis is a sales argument, not a measurement. Serious gains are targeted, then measured on your data.
Without prioritisation by ROI, effort and adoption, the roadmap proposed reflects the seller's intuition, not your reality.
A real analysis also identifies waste: under-used licences, zombie projects, redundant agents to consolidate or cut.
If everything starts with months of integration, be wary: the analysis should start from the logs and exports you already produce, read-only.
Idun Group, AI automation agency, 2026 verification criteria
The grid above sets out five concrete signs to check before you sign. At Idun Group, a serious AI automation agency starts by reading, read-only, the logs and exports you already produce, installing nothing on workstations. We analyse usage: results are aggregated by team and by tool, never by person, and the exact status of the processing is validated with your data protection officer. The debrief is grounded in your reality, not in a market average: use cases detected, waste identified and a roadmap prioritised by ROI. That verification saves you from paying for expertise the provider never actually brought to your sector.
only have deployed AI agents at scale, against 62 % running experiments.
62 % of companies are experimenting with AI agents but only 23 % have deployed them at scale in at least one business function. Fewer than 10 % extend them across several functions (Ifop-Converteo study, 2026). The gap does not come from a lack of available technology but from execution that often stays stuck at the pilot stage.
That measurement confirms what we see in the field with our clients: the difficulty is not launching one more project, it is knowing what the ones already running return. A company that picks the right approach gets that answer without months of internal tinkering. We work first with companies of 500 to 5,000 employees that have already deployed AI, counting the headcount of the entity that buys, not of the group. That size marker excludes no one: to frame an AI strategy, what counts is usage, with at least 50 AI licences deployed, and a decision taken in France.
A workflow built on a third-party platform stays subject to its pricing terms, its API limits and its roadmap decided elsewhere. That vendor dependency weighs heavily as soon as volume grows. And waste piles up on the usage side too: across the estates we analyse, 20 to 40 % of the AI licences paid for go unused, while zombie projects keep burning tokens without serving anyone. Automating without a prior analysis of what exists means adding a layer of cost on top of waste nobody has measured yet.
30 to 50 % of AI initiatives fail to return what they should (Gartner: at least 30 % of proofs of concept abandoned; McKinsey: more than 80 % with no impact on the operating result). ROI is estimated use case by use case, from usage actually measured, never from a generic promise. That is the whole point of our analysis: a concrete figure to present to your leadership, grounded in your data. Target set and measured at your site: 30 to 50 % lower cost on the targeted usage and 2 to 4 times fewer errors.
We start from the logs and exports you already produce, on the platforms you administer (Azure OpenAI and in-house models, Dust, Copilot, custom agents), installing nothing and re-instrumenting nothing; handing over exports or a read-only connection, the exact arrangement is settled in the scoping meeting. For consumer AI used on individual accounts we do not collect that data: we name the blind spot and address it through the internal offer, a deliverable of the audit. The journey has four stages: a 30-minute discovery call; scoping, through interviews with the IT department and the business units (a mandatory step: the tooling does not replace them, it makes them factual); the analysis, from the logs and exports provided; the debrief. No access to your data before signature: pre-sales happens in conversation. The deliverables:
An analysis is only worth the decisions it enables. We prioritise your use cases by ROI, effort and adoption, we arbitrate your licences at renewal time, we flag what is duplicated from one tool to another (the same use case found on three tools deserves one single implementation) and we carry the decision through to the steering committee.
The diagram above contrasts the current state, AI you endure and whose value nobody measures, with the target state: usage mapped, waste identified and a roadmap your leadership can arbitrate with full knowledge of the facts.
We tell you what to automate, in what order and on what conditions; we do not do it for you. Development stays with your teams, your integrators or a partner, and we can help you choose one. That is what keeps the recommendation credible: we sell neither the agents, nor the agent platform, nor the training we recommend, and we have no interest in advising a project we would then bill you for.
A company's AI usage shifts every month: new tools, new teams, new waste. The continuous insights option (recurring re-analysis, the same deliverables kept up to date) extends the engagement: the map, the list of what to abandon, the full cost of AI, the ROI roadmap. Your leadership steers the value of AI over time, instead of rediscovering the problem once a year.
Demand that the analysis runs read-only, on the logs you already have, installing nothing on workstations. We analyse usage: results are aggregated by team and by tool, never by person, and the exact status of the processing is validated with your data protection officer. That principle runs through our whole AI agency practice, not just the automation side.
Be wary of a provider that builds before it measures. Demand a ROI estimated use case by use case and a prioritised roadmap before the first project. Serious gains are expressed as targets set and measured at your site, never guaranteed in advance.
The analysis should consume the logs and exports you already produce, on the platforms you administer: Azure OpenAI and in-house models, Dust, Copilot, custom agents. A reading limited to one tool leaves most of your usage in the blind spot. For consumer AI on individual accounts we do not collect that data: we name the blind spot and address it through the internal offer.
Two distinct deadlines coexist in the European AI regulation. An agency able to set out that distinction precisely demonstrates real regulatory expertise rather than a generic urgency argument. Compliance informs the roadmap: it is never the way into an engagement, proof of value comes first.
Two distinct regimes coexist in the European AI regulation (the AI Act). The transparency obligations of article 50, telling users they are interacting with a machine, have applied since 2 August 2026 (article 113 of Regulation (EU) 2024/1689). For the high-risk systems of annex III, the AI Omnibus, in force since 27 July 2026, deferred the application of the rules to 2 December 2027 (European Commission). Confusing the two regimes leads to badly calibrated compliance choices, while the penalties provided for by the regulation go up to 35 million euros or 7 % of worldwide turnover.
Others measure usage, fix an interface friction or push you to build agents. We arbitrate the AI portfolio (industrialise, consolidate, cut) with a quantified return on investment and a roadmap you can present to a steering committee. The grid below sets out the questions to ask and our answer to each.
At Idun Group, the advice rests on our usage-analysis tooling: it reads the logs and exports of the platforms you administer (Azure OpenAI and in-house models, Dust, Copilot, custom agents), brings your tools into one view and detects use cases from real usage rather than from what people report. That is what makes the analysis faster and more factual than a generalist consultancy can be, grounded in your company's reality. We sell neither the agents, nor the agent platform, nor the training we recommend: the analysis serves your decision, not a catalogue. The analysis delivers, among other things:
lower cost on the targeted usage
fewer errors, set and measured at your site
Target set and measured at your site: 30 to 50 % lower cost on the targeted usage and 2 to 4 times fewer errors, never guaranteed in advance. The analysis runs read-only, installing nothing on workstations. We analyse usage: results are aggregated by team and by tool, never by person, and the exact status of the processing is validated with your data protection officer. By default the analysis starts from the exports provided and your data stays in France, at Scaleway in Paris. If your data protection officer or your security policy requires it, the chain runs inside your own perimeter and the data never leaves your premises. The arrangement is settled in the scoping meeting. No access to your data before signature: pre-sales happens in conversation.
LVMH: an Idun engagement on the generative AI platform of all the perfume and cosmetics brands. Leeto: strategic audit then production rollout of AI agents, automated classification under human review and validation. Airbus is also among our references. More than 40 AI diagnostics run in a year. The big logos are evidence; our primary target is companies of 500 to 5,000 employees that have already deployed AI, counting the headcount of the entity that buys.
Book a discovery callAn AI automation agency is judged on what it measures at your site, not on its promises. We analyse your real usage read-only, we deliver a roadmap prioritised by ROI and we tell you what to industrialise. Among our references, LVMH: an Idun engagement on the generative AI platform of all the perfume and cosmetics brands. First step: a 30-minute discovery call.