By Frode Nybo, Project Manager - Norway
The rising need for AI knowledge aligns with MIT Technology Review’s A Playbook for Crafting AI Strategy, which reports that 95% of companies use AI and 99% expect to in the future. In fact, half of all companies expect to fully deploy AI across all business functions within two years.
At the same time, according to the Project Management Institute (PMI), the reality is stark: 70–80% of AI projects fail. They fall short on ROI, miss deadlines or budgets, or deliver technically flawed outcomes.
So, companies are investing heavily in AI — but many initiatives still fail. And in many organizations, the knowledge gap persists.
Should we really expect different results if we keep doing the same thing? There are many approaches to starting your AI journey, but I’ve observed two particularly common ones:
Many organizations are drawn to “moonshot” AI projects — ambitious initiatives aimed at transforming business models or disrupting markets with cutting-edge technology.
While inspiring, these efforts are often misaligned with current AI maturity, carrying high technical complexity, unclear ROI, reliance on immature technology, and a heavy demand for scarce AI talent. When they fail — and many do — they create fatigue, damage internal trust, and slow AI adoption.
So, is it FOMO that drives companies to chase these kinds of projects? Interestingly, according to MIT’s Playbook, 98% of organizations say they are willing to forgo being the first movers in AI if it means deploying it safely and securely. This suggests that using AI as a competitive advantage is not considered critical – at least not at the cost of stability and risk.
LLMs Without Data Access? Personal Productivity Yes, Strategic Value No
Many businesses are advised to familiarize themselves with AI by giving employees access to LLMs like ChatGPT, Claude, or LaMDA to experiment and explore. However, in conversations with several companies, I’ve encountered a high level of skepticism — even resistance — toward integrating LLMs into their infrastructure. That skepticism deepens when it comes to granting these models access to sensitive business or customer data.
This is reflected in how ERP vendors like Oracle design their AI solutions. Oracle offers over 50 generative AI use cases. These are specifically designed to respect enterprise data privacy and security, and Oracle emphasizes that no customer data is shared with external LLMs.
LLMs can be excellent tools for boosting personal and team productivity, but their strategic value is clearly limited in a business context if they can’t access the data that drives decisions.
I think both approaches reflect a tendency to overestimate what AI can do today, while underestimating what’s possible using tools companies already have — often right inside their existing systems.
So, is there a better way for companies to start their AI journey?
Given the high failure rate of AI projects, the immaturity of the technology, the need for safe adoption, and the ongoing knowledge gap — isn’t it time more companies consider a more pragmatic starting point?
After finishing my Master’s in Information Technology (specializing in Enterprise Resource Planning, ERP), I spent around 10 years working with ERPs — mainly SAP — as both an application consultant and a project manager. Nearly all medium and large companies use ERPs to support critical business processes across functions like manufacturing, asset management, logistics, sales, finance, and HR. According to Gartner, traditional ERPs are being heavily influenced by AI-based technologies, and ERP leaders are advised to adapt their strategies accordingly — toward a smarter, more automated reality. Many vendors offer AI capabilities and are now beginning to embed generative AI into their products. By 2027, at least 50% of AI-enabled features in ERP applications will be powered by generative AI.
What about starting with what you have?
But, is what ERP vendors offer good enough to get started?
SAP supports over 200 business use cases, such as reducing accounts receivable matching effort by 71%.
Oracle supports over 50 generative AI use cases, including tools that help sales teams create customer success stories from account history.
Microsoft integrates Copilot into Dynamics 365. In Supply Chain Management, it helps decrease supplier delays by 27%.
IFS reports over 60 customer-tested industrial AI capabilities, such as the Demand Plan- ner AI Forecast, which improves forecasting accuracy to reduce safety stock.
Visma integrates into 155 products, and reports that its new NLP technology has delive- red customers a 100% increase in invoice automation rates.
Even though nearly all medium and large companies use ERPs — and these systems clearly now offer a wide range of AI features and use cases — most organizations haven’t moved beyond pilot projects. According to MIT’s Playbook, 76% of companies have deployed AI in only one to three use cases.
That’s exactly why I suggest a simple four-step ERP-first approach for kicking off AI in your company:
1. Start with the low-hanging fruit – use AI features already in your ERP
Open a dialogue with your ERP vendor to learn what AI features are available in your cur- rent version, and what’s planned for the next year or two. At the same time, understand what’s needed to use them — things like data quality, system setup, infrastructure, licen- sing costs, or basic training. Involve your managers and subject matter experts (SMEs) early. Many SMEs, with decades of experience in ERP-related functions like manufactu- ring, logistics, and finance — often with deeper business knowledge than the vendor teams creating the features — can quickly spot where a feature isn’t right or isn’t working well and suggest improvements to the vendor.
2. Collaborate with your ERP vendor on quick-win AI opportunities
Once managers and SMEs start using existing AI features, they’ll often spot gaps or suggest new ideas. Their deep business knowledge makes them valuable in shaping solutions that work in practice. Work with your vendor to turn these ideas into reality. This might mean submitting a feature request — sometimes added to the standard product at no cost, other times funded as a custom feature you keep. To save costs, you could help write the requirement specification or test during development, user acceptance, pilot, or go-live. Active involvement improves the tools you get and strengthens your vendor relationship.
3. Consolidate where it makes sense
Gartner predicts that by 2027, 60% of customers replacing ERP applications will choose software for its ability to integrate, automate, and use AI to manage end-to-end processes — rather than the traditional transactional functions they take for granted.
One of the main reasons AI projects fail is that many organizations underestimate what’s needed to achieve high-quality data for AI models. The entire “data process” — from collecting, storing, and cleaning to monitoring and maintaining — is a significant challenge, especially when about 90% of enterprise data is unstructured.
If your company wants real value from AI, consolidating into a single ERP can help. A unified system delivers consistent data, simpler integration, and often lowers costs.
4. Integrate where necessary
Sometimes full ERP consolidation just isn’t possible, whether for political reasons, technical constraints, or simply because different systems have been in place for a long time. You may also have other information systems outside your ERP stack that hold valuable business data.
When integration is required, choose a solution that can handle both structured and unstructured data, integrate openly with any type of system, and be ready to support future AI innovations. In some cases, the integration tools in your main ERP may be the best choice.
Moonshot Projects And LLMs Can Have Their Place, but Why Do I Believe the “Boring” ERP-First AI Approach Is The Smarter Starting Point?
Starting with AI in your ERP offers a low-cost, low-risk way to make real progress. The use cases are already tied to the business processes you run every day, and the features have been tested, secured, and refined by the vendor — often in close collaboration with customer experts. Your ERP already holds structured, governed data, and your teams know how to work with it. This means you can move quickly from identifying a need to putting solutions into production, improving decisions, and automating processes while keeping costs under control and avoiding unnecessary risks. For many organizations, it’s the most practical way to build momentum and create a solid foundation for future AI adoption.
