[{"_1":2,"_80":-5,"_81":-5},"loaderData",{"_3":4,"_15":16,"_19":20},"root",{"_5":6},"analytics",{"_7":8,"_9":10,"_11":12,"_13":14},"enabled",true,"key","phc_oB9H4iqQJ8otg7L2NNeTyouNuo8weGuQ8BXJbKkVwjkW","host","https://eu.i.posthog.com","environment","production","routes/lang",{"_17":18},"lang","en","routes/lang.blog.$slug",{"_17":18,"_21":22,"_78":79},"post",{"_23":24,"_25":26,"_27":28,"_29":30,"_31":32,"_33":34,"_35":36,"_37":8,"_38":39,"_40":41,"_50":51,"_68":69,"_72":73,"_74":75,"_76":77},"title","Why Do Most AI Projects Fail?","slug","why-most-ai-projects-fail","author","Sarah Ghidalia","date","2025-12-01","excerpt","Succeeding with an AI project is no accident: a clear roadmap, user involvement, agile methods, co-development with specialists, ROI evaluation and data sovereignty are all levers that turn an idea into a concrete, useful and durable solution for the business.","category","Digital Strategy","image","https://asset.gencior.ai/landing/blog/pourquoi-la-plupart-des-projets-IA-echouent.png","published","publishedAt",["D",1764460800000],"tags",[42,43,44,45,46,47,48,49],"AI","Artificial Intelligence ROI","Agility","AI Audit","Industry 4.0","Bourgogne-Franche-Comté","Digital Transformation","Sovereignty","keywords",[52,53,54,55,56,57,58,59,60,47,61,62,63,64,65,66,67],"artificial intelligence","AI audit","AI roadmap","AI best practices","business process automation","enterprise AI integration","data sovereignty","predictive AI","enterprise GenAI","smart manufacturing","measuring AI ROI","digital sovereignty","Gencior","industrial SMEs","production optimization","lean AI","translations",{"_70":71},"fr","pourquoi-la-plupart-des-projets-ia-echouent","content","\nThe word AI has been everywhere for three years. It has replaced \"digital\", \"big data\" and \"cloud\". Yet, like its predecessors, AI is frightening, especially when it comes to bringing it into a company.\n\nLike many IT projects, AI is perceived as an essential but risky innovation. ESSENTIAL, because if you go by public discourse, by the outsized funding this technology enjoys, and by its integration into everyday tools, everyone seems to be using it today. The injunctions come thick and fast: it is a revolution, you must use it if you want to stay in the race. RISKY all the same, because like any innovation it threatens to disrupt established ways of working. Recent statistical studies, such as the MIT report on the use of generative AI (or *GenAI*), reveal that 95% of projects generated no measurable financial return: everyone takes fright.\n\nHow are we to make sense of this contradiction? And above all, what kind of thinking is needed to get the best out of AI?\n\n## Succeeding with an AI Project: Lessons from the Software World\n\nDigital projects, not just AI projects, run into plenty of difficulties. Most of them can be summed up like this: the designer's insufficient understanding of the end client's real needs.\n\nSince 1994, The Standish Group has published its CHAOS report (for Comprehensive Human Appraisal for Originating Software) almost every two years. The *CHAOS report* is a goldmine for understanding the key success factors (and failure factors) of IT projects, and how the human factor influences them. In 2020, the success rate for software projects was only 31%. Success here means: deadlines and budget met, with all planned features delivered. Of the rest, 19% of projects were simply cancelled (money out of the window), while the remaining 50% were finished over schedule or over budget. Before even setting up an AI project, let us look at what can be learned from software projects in general.\n\nA major success factor is the use of agile methods. That was, incidentally, this year's theme at the \"Créer demain\" event organized by the Bourgogne-Franche-Comté Regional Economic Agency. Throughout the day, the various speakers set out the benefits of agility: better-structured projects, better suited to client needs because they are co-built with end users, controlled costs thanks to incremental development methods, features prioritized on the roadmap according to user feedback, with a clear idea of the expected results. This demands, from the software creator (and therefore the AI developer), systematic transparency towards the client, along with a willingness to adapt and continuously improve. It is also crucial to identify clearly who the end user is: is it the entity funding the project? an employee who will use the solution? the customers of the company for whom the solution is being developed?\n\nIf there is one thing to take away from the MIT report , it is that the use of Generative AI (GenAI) in business is not yet mature, which creates misunderstanding and weak returns on investment (ROI). Generative AI has won over the general public in recent years, but struggles to be genuinely used inside companies. The GenAI offerings on the market are, moreover, hardly geared towards the specific needs of business functions. No solution, no market. As a result, corporate AI budgets are often confined to marketing and sales, whereas it is usually the automation of administrative or operational tasks that generates the best ROI. The weakness of this report is that it is limited to GenAI: it does not show the full range of solutions made possible by the various types of AI available today. That is a shame, because technically mature technologies such as predictive AI or reasoning engines are also powerful tools that can be put to work in industry right now.\n\nA few essential points to remember:\n\n* Software projects fail frequently: only 31% meet deadlines, budget and scope, 19% are cancelled and 50% overrun on time or budget.\n* Agile methods are decisive for AI projects, as they encourage co-construction with end users, feature prioritization and continuous adaptability.\n* Clearly identifying the end client is critical to a project's success.\n* GenAI remains immature in business today: limited uses, weak ROI and a concentration on marketing and sales.\n* Other types of AI, such as predictive AI or reasoning engines, are already mature and make it possible to automate tasks for real productivity gains.\n\n## How Do You Set Up an AI Project That Succeeds?\n\nConcretely, how do you set up an AI project that succeeds — meaning one that keeps its promises in terms of time, cost and return on investment?\n\nFirst of all, the very first thing to do is to map all your business processes with the help of an AI professional. Being an IT professional in general as well, they will be able to spot the automatable tasks in the value chain and how to take advantage of AI. The MIT report highlights the fact that solutions co-developed with specialized external providers double the chances of success compared with solutions developed purely in-house (67% instead of 33%). This first audit stage, which at Gencior we call the [Strategic Assessment](https://gencior.ai/en#accompagnement), is indispensable for understanding precisely where our clients' needs lie. This stage is not always the simplest, but it sometimes puts a finger on internal dysfunctions, or simply on points that can be optimized quickly and cheaply. It is also at this stage that priorities, technical constraints and the expectations of the various stakeholders are defined, ensuring that the project starts on foundations that are both solid and realistic.\n\nThe audit also reveals the constraints the solution will have to accommodate in order to fit as smoothly as possible into the company's processes. The idea is to minimize disruption, whether for the company's staff or for its information system. Ideally, an AI project succeeds when it requires no lengthy training to get started, because the teams have been involved in the project throughout its creation, as recent reports recommend . Like any accomplished digital object, it has to be intuitive for the user in order to support them in their daily work.\n\nPlanning and success in AI projects also rest on evaluating their financial or operational impact on the value chain. Defining that evaluation upfront is good practice. It sets clear criteria for measuring the project's success, anticipating the resources required and prioritizing the most promising initiatives. By clarifying the performance indicators from the outset, you limit the risk of drift and ensure that every stage of the project genuinely contributes to the value expected by the business. Evaluating AI projects is not always straightforward, because results can be hard to quantify; impacts are often indirect, and benefits spread across several functions. It is essential to combine financial measures, operational gains and qualitative criteria to obtain a complete picture of each project's potential. The Rockwell Automation report also recommends securing short-term returns on investment (a few months at most) to encourage adoption and let AI solutions fund themselves.\n\nFinally, a point that matters to us at Gencior is keeping in mind the need to avoid becoming too dependent on external players (particularly foreign ones) and to preserve the sovereignty of your data within your own territory. That means favouring trusted partners, securing data flows and making sure technology choices comply with local regulations. This vigilance is essential not only for security, but also to guarantee the company's strategic autonomy in its AI projects.\n\nThe most important points from this section:\n\n* The [Strategic Assessment](https://gencior.ai/en#accompagnement) is indispensable for mapping business processes, identifying automatable tasks and involving end users.\n* Collaborating with specialized partners doubles the chances of success compared with purely in-house development.\n* Evaluating ROI from the start makes it possible to prioritize projects, anticipate resources and ensure that every stage delivers value.\n* Adoption by teams is made easier by involving them in the creation of intuitive solutions.\n* Data sovereignty and security guarantee strategic autonomy and compliance with local regulations.\n\n> A recurring problem in AI projects is that they stop after the prototype, as we detail in [this article](https://gencior.ai/en/blog/from-pilot-to-deployment-ai-projects).\n\n## Key Takeaways\n\nTo conclude, the success of an AI project rests on a clear roadmap that structures priorities and aligns actions with the company's objectives. It is essential not to limit yourself to GenAI and to explore the full range of available technologies, which are often better suited to real needs on the ground. Involving users throughout the design process and relying on agile methods guarantees fast adoption and continuous evolution of the product. Calling on external specialists, in co-development, significantly increases the chances of operational deployment and makes integration with existing systems easier. Measuring return on investment from the outset guides your choices and limits drift, while preserving data sovereignty protects the company's strategic autonomy. By combining these practices, an AI project stands a real chance of keeping its promises in terms of time, cost and value created.\n\n**👉 Want to identify your AI opportunities or structure your projects reliably? Get in touch to run your **Strategic Assessment** and lay the foundations for a solution that is genuinely useful and ready to be deployed.**\n\n\n## Sources\n\n1. [State of AI in Business 2025](https://www.aigl.blog/content/files/2025/09/The-GenAI-Divide-STATE-OF-AI-IN-BUSINESS-2025.pdf)\n2. [The Standish Group](https://www.standishgroup.com)\n3. [Summary article on the CHAOS report](https://thestory.is/en/journal/chaos-report/)\n4. [Rockwell Automation report](https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/campaigns/state-of-smart-2025/INFO-BR027D-FR-P-noi.pdf)\n","compiledSource","\"use strict\";\nconst {Fragment: _Fragment, jsx: _jsx, jsxs: _jsxs} = arguments[0];\nfunction _createMdxContent(props) {\n const _components = {\n a: \"a\",\n blockquote: \"blockquote\",\n em: \"em\",\n h2: \"h2\",\n li: \"li\",\n ol: \"ol\",\n p: \"p\",\n strong: \"strong\",\n ul: \"ul\",\n ...props.components\n }, {CallbackRequest, LinkPreview} = _components;\n if (!CallbackRequest) _missingMdxReference(\"CallbackRequest\", true);\n if (!LinkPreview) _missingMdxReference(\"LinkPreview\", true);\n return _jsxs(_Fragment, {\n children: [_jsx(_components.p, {\n children: \"The word AI has been everywhere for three years. It has replaced \\\"digital\\\", \\\"big data\\\" and \\\"cloud\\\". Yet, like its predecessors, AI is frightening, especially when it comes to bringing it into a company.\"\n }), \"\\n\", _jsxs(_components.p, {\n children: [\"Like many IT projects, AI is perceived as an essential but risky innovation. ESSENTIAL, because if you go by public discourse, by the outsized funding this technology enjoys, and by its integration into everyday tools, everyone seems to be using it today. The injunctions come thick and fast: it is a revolution, you must use it if you want to stay in the race. RISKY all the same, because like any innovation it threatens to disrupt established ways of working. Recent statistical studies, such as the MIT report on the use of generative AI \", _jsx(LinkPreview, {\n text: \"(1)\",\n url: \"https://www.aigl.blog/content/files/2025/09/The-GenAI-Divide-STATE-OF-AI-IN-BUSINESS-2025.pdf\"\n }), \" (or \", _jsx(_components.em, {\n children: \"GenAI\"\n }), \"), reveal that 95% of projects generated no measurable financial return: everyone takes fright.\"]\n }), \"\\n\", _jsx(_components.p, {\n children: \"How are we to make sense of this contradiction? And above all, what kind of thinking is needed to get the best out of AI?\"\n }), \"\\n\", _jsx(_components.h2, {\n id: \"succeeding-with-an-ai-project-lessons-from-the-software-world\",\n children: \"Succeeding with an AI Project: Lessons from the Software World\"\n }), \"\\n\", _jsx(_components.p, {\n children: \"Digital projects, not just AI projects, run into plenty of difficulties. Most of them can be summed up like this: the designer's insufficient understanding of the end client's real needs.\"\n }), \"\\n\", _jsxs(_components.p, {\n children: [\"Since 1994, The Standish Group has published its CHAOS report (for Comprehensive Human Appraisal for Originating Software) \", _jsx(LinkPreview, {\n text: \"(2)\",\n url: \"https://www.standishgroup.com\"\n }), \" \", _jsx(LinkPreview, {\n text: \"(3)\",\n url: \"https://thestory.is/en/journal/chaos-report/\"\n }), \" almost every two years. The \", _jsx(_components.em, {\n children: \"CHAOS report\"\n }), \" is a goldmine for understanding the key success factors (and failure factors) of IT projects, and how the human factor influences them. In 2020, the success rate for software projects was only 31%. Success here means: deadlines and budget met, with all planned features delivered. Of the rest, 19% of projects were simply cancelled (money out of the window), while the remaining 50% were finished over schedule or over budget. Before even setting up an AI project, let us look at what can be learned from software projects in general.\"]\n }), \"\\n\", _jsx(_components.p, {\n children: \"A major success factor is the use of agile methods. That was, incidentally, this year's theme at the \\\"Créer demain\\\" event organized by the Bourgogne-Franche-Comté Regional Economic Agency. Throughout the day, the various speakers set out the benefits of agility: better-structured projects, better suited to client needs because they are co-built with end users, controlled costs thanks to incremental development methods, features prioritized on the roadmap according to user feedback, with a clear idea of the expected results. This demands, from the software creator (and therefore the AI developer), systematic transparency towards the client, along with a willingness to adapt and continuously improve. It is also crucial to identify clearly who the end user is: is it the entity funding the project? an employee who will use the solution? the customers of the company for whom the solution is being developed?\"\n }), \"\\n\", _jsxs(_components.p, {\n children: [\"If there is one thing to take away from the MIT report \", _jsx(LinkPreview, {\n text: \"(1)\",\n url: \"https://www.aigl.blog/content/files/2025/09/The-GenAI-Divide-STATE-OF-AI-IN-BUSINESS-2025.pdf\"\n }), \", it is that the use of Generative AI (GenAI) in business is not yet mature, which creates misunderstanding and weak returns on investment (ROI). Generative AI has won over the general public in recent years, but struggles to be genuinely used inside companies. The GenAI offerings on the market are, moreover, hardly geared towards the specific needs of business functions. No solution, no market. As a result, corporate AI budgets are often confined to marketing and sales, whereas it is usually the automation of administrative or operational tasks that generates the best ROI. The weakness of this report is that it is limited to GenAI: it does not show the full range of solutions made possible by the various types of AI available today. That is a shame, because technically mature technologies such as predictive AI or reasoning engines are also powerful tools that can be put to work in industry right now.\"]\n }), \"\\n\", _jsx(_components.p, {\n children: \"A few essential points to remember:\"\n }), \"\\n\", _jsxs(_components.ul, {\n children: [\"\\n\", _jsx(_components.li, {\n children: \"Software projects fail frequently: only 31% meet deadlines, budget and scope, 19% are cancelled and 50% overrun on time or budget.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"Agile methods are decisive for AI projects, as they encourage co-construction with end users, feature prioritization and continuous adaptability.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"Clearly identifying the end client is critical to a project's success.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"GenAI remains immature in business today: limited uses, weak ROI and a concentration on marketing and sales.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"Other types of AI, such as predictive AI or reasoning engines, are already mature and make it possible to automate tasks for real productivity gains.\"\n }), \"\\n\"]\n }), \"\\n\", _jsx(_components.h2, {\n id: \"how-do-you-set-up-an-ai-project-that-succeeds\",\n children: \"How Do You Set Up an AI Project That Succeeds?\"\n }), \"\\n\", _jsx(_components.p, {\n children: \"Concretely, how do you set up an AI project that succeeds — meaning one that keeps its promises in terms of time, cost and return on investment?\"\n }), \"\\n\", _jsxs(_components.p, {\n children: [\"First of all, the very first thing to do is to map all your business processes with the help of an AI professional. Being an IT professional in general as well, they will be able to spot the automatable tasks in the value chain and how to take advantage of AI. The MIT report \", _jsx(LinkPreview, {\n text: \"(1)\",\n url: \"https://www.aigl.blog/content/files/2025/09/The-GenAI-Divide-STATE-OF-AI-IN-BUSINESS-2025.pdf\"\n }), \" highlights the fact that solutions co-developed with specialized external providers double the chances of success compared with solutions developed purely in-house (67% instead of 33%). This first audit stage, which at Gencior we call the \", _jsx(_components.a, {\n href: \"https://gencior.ai/en#accompagnement\",\n children: \"Strategic Assessment\"\n }), \", is indispensable for understanding precisely where our clients' needs lie. This stage is not always the simplest, but it sometimes puts a finger on internal dysfunctions, or simply on points that can be optimized quickly and cheaply. It is also at this stage that priorities, technical constraints and the expectations of the various stakeholders are defined, ensuring that the project starts on foundations that are both solid and realistic.\"]\n }), \"\\n\", _jsxs(_components.p, {\n children: [\"The audit also reveals the constraints the solution will have to accommodate in order to fit as smoothly as possible into the company's processes. The idea is to minimize disruption, whether for the company's staff or for its information system. Ideally, an AI project succeeds when it requires no lengthy training to get started, because the teams have been involved in the project throughout its creation, as recent reports recommend \", _jsx(LinkPreview, {\n text: \"(1)\",\n url: \"https://www.aigl.blog/content/files/2025/09/The-GenAI-Divide-STATE-OF-AI-IN-BUSINESS-2025.pdf\"\n }), \" \", _jsx(LinkPreview, {\n text: \"(2)\",\n url: \"https://www.standishgroup.com\"\n }), \" \", _jsx(LinkPreview, {\n text: \"(4)\",\n url: \"https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/campaigns/state-of-smart-2025/INFO-BR027D-FR-P-noi.pdf\"\n }), \". Like any accomplished digital object, it has to be intuitive for the user in order to support them in their daily work.\"]\n }), \"\\n\", _jsxs(_components.p, {\n children: [\"Planning and success in AI projects also rest on evaluating their financial or operational impact on the value chain. Defining that evaluation upfront is good practice. It sets clear criteria for measuring the project's success, anticipating the resources required and prioritizing the most promising initiatives. By clarifying the performance indicators from the outset, you limit the risk of drift and ensure that every stage of the project genuinely contributes to the value expected by the business. Evaluating AI projects is not always straightforward, because results can be hard to quantify; impacts are often indirect, and benefits spread across several functions. It is essential to combine financial measures, operational gains and qualitative criteria to obtain a complete picture of each project's potential. The Rockwell Automation report \", _jsx(LinkPreview, {\n text: \"(4)\",\n url: \"https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/campaigns/state-of-smart-2025/INFO-BR027D-FR-P-noi.pdf\"\n }), \" also recommends securing short-term returns on investment (a few months at most) to encourage adoption and let AI solutions fund themselves.\"]\n }), \"\\n\", _jsx(_components.p, {\n children: \"Finally, a point that matters to us at Gencior is keeping in mind the need to avoid becoming too dependent on external players (particularly foreign ones) and to preserve the sovereignty of your data within your own territory. That means favouring trusted partners, securing data flows and making sure technology choices comply with local regulations. This vigilance is essential not only for security, but also to guarantee the company's strategic autonomy in its AI projects.\"\n }), \"\\n\", _jsx(_components.p, {\n children: \"The most important points from this section:\"\n }), \"\\n\", _jsxs(_components.ul, {\n children: [\"\\n\", _jsxs(_components.li, {\n children: [\"The \", _jsx(_components.a, {\n href: \"https://gencior.ai/en#accompagnement\",\n children: \"Strategic Assessment\"\n }), \" is indispensable for mapping business processes, identifying automatable tasks and involving end users.\"]\n }), \"\\n\", _jsx(_components.li, {\n children: \"Collaborating with specialized partners doubles the chances of success compared with purely in-house development.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"Evaluating ROI from the start makes it possible to prioritize projects, anticipate resources and ensure that every stage delivers value.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"Adoption by teams is made easier by involving them in the creation of intuitive solutions.\"\n }), \"\\n\", _jsx(_components.li, {\n children: \"Data sovereignty and security guarantee strategic autonomy and compliance with local regulations.\"\n }), \"\\n\"]\n }), \"\\n\", _jsxs(_components.blockquote, {\n children: [\"\\n\", _jsxs(_components.p, {\n children: [\"A recurring problem in AI projects is that they stop after the prototype, as we detail in \", _jsx(_components.a, {\n href: \"https://gencior.ai/en/blog/from-pilot-to-deployment-ai-projects\",\n children: \"this article\"\n }), \".\"]\n }), \"\\n\"]\n }), \"\\n\", _jsx(_components.h2, {\n id: \"key-takeaways\",\n children: \"Key Takeaways\"\n }), \"\\n\", _jsx(_components.p, {\n children: \"To conclude, the success of an AI project rests on a clear roadmap that structures priorities and aligns actions with the company's objectives. It is essential not to limit yourself to GenAI and to explore the full range of available technologies, which are often better suited to real needs on the ground. Involving users throughout the design process and relying on agile methods guarantees fast adoption and continuous evolution of the product. Calling on external specialists, in co-development, significantly increases the chances of operational deployment and makes integration with existing systems easier. Measuring return on investment from the outset guides your choices and limits drift, while preserving data sovereignty protects the company's strategic autonomy. By combining these practices, an AI project stands a real chance of keeping its promises in terms of time, cost and value created.\"\n }), \"\\n\", _jsx(_components.p, {\n children: _jsxs(_components.strong, {\n children: [\"👉 Want to identify your AI opportunities or structure your projects reliably? Get in touch to run your \", _jsx(_components.strong, {\n children: \"Strategic Assessment\"\n }), \" and lay the foundations for a solution that is genuinely useful and ready to be deployed.\"]\n })\n }), \"\\n\", _jsx(CallbackRequest, {\n lang: \"en\",\n CTAText: \"Get in touch\"\n }), \"\\n\", _jsx(_components.h2, {\n id: \"sources\",\n children: \"Sources\"\n }), \"\\n\", _jsxs(_components.ol, {\n children: [\"\\n\", _jsx(_components.li, {\n children: _jsx(_components.a, {\n href: \"https://www.aigl.blog/content/files/2025/09/The-GenAI-Divide-STATE-OF-AI-IN-BUSINESS-2025.pdf\",\n children: \"State of AI in Business 2025\"\n })\n }), \"\\n\", _jsx(_components.li, {\n children: _jsx(_components.a, {\n href: \"https://www.standishgroup.com\",\n children: \"The Standish Group\"\n })\n }), \"\\n\", _jsx(_components.li, {\n children: _jsx(_components.a, {\n href: \"https://thestory.is/en/journal/chaos-report/\",\n children: \"Summary article on the CHAOS report\"\n })\n }), \"\\n\", _jsx(_components.li, {\n children: _jsx(_components.a, {\n href: \"https://www.rockwellautomation.com/content/dam/rockwell-automation/documents/pdf/campaigns/state-of-smart-2025/INFO-BR027D-FR-P-noi.pdf\",\n children: \"Rockwell Automation report\"\n })\n }), \"\\n\"]\n })]\n });\n}\nfunction MDXContent(props = {}) {\n const {wrapper: MDXLayout} = props.components || ({});\n return MDXLayout ? _jsx(MDXLayout, {\n ...props,\n children: _jsx(_createMdxContent, {\n ...props\n })\n }) : _createMdxContent(props);\n}\nreturn {\n default: MDXContent\n};\nfunction _missingMdxReference(id, component) {\n throw new Error(\"Expected \" + (component ? \"component\" : \"object\") + \" `\" + id + \"` to be defined: you likely forgot to import, pass, or provide it.\");\n}\n","readingTime","8","dnsName","gencior.ai","actionData","errors"]