A recent Harvard Business Review article by Nathan Furr and Andrew Shipilov, titled “Beware the AI Experimentation Trap,” addresses the issue of return on investment in generative AI.
The article recalls an MIT Media Lab/Project NANDA report that 95 percent of investments in Gen AI have not produced economic returns. This finding, combined with the judged disappointing launch of GPT-5, has fueled skepticism about AI’s ability to generate value at scale. It is no coincidence that Gartner now places Gen AI in the “trough of disillusionment” stage of its Hype Cycle.
Recall that Gartner’s Hype Cycle is a model that describes the stages of adoption of emerging technologies: from initial enthusiasm, to disillusionment, to maturity and widespread adoption.
Indeed, many signs indicate that generative AI is experiencing the “trough of disillusionment.”
- Initial expectations inflated by hype;
- concrete limitations that have emerged (high cost, uncertain results, difficulty in enterprise-scale adoption);
- projects that fail to scale, generating frustration and negative economic returns.
This does not mean that AI is a bubble, but that we have entered the stage where companies distinguish real applications from illusory ones. After disillusionment, in fact, the Hype Cycle provides the
The authors point out that, in reality, the MIT report is more nuanced: individuals are indeed increasing productivity through Gen AI tools, but the benefits are not yet showing up in corporate income statements.
We can imagine that the two authors are referring, for example, to tools that bring personal micro-efficiencies (time saved, less cognitive fatigue) but not yet business macro-efficiencies (reduced structural costs or increased revenues):
- Employees and consultants use ChatGPT or Copilot to draft emails, reports, or summaries. So they save time, but the impact on turnover or overall costs remains marginal and unmeasurable.
- Managers and sales people use tools such asCanva AI or Gamma.app to prepare presentations faster. As a result, efficiency improves, but revenues do not change.
- With GitHub Copilot, developers produce more lines of code in less time. However, software projects always have long cycles and the savings are not yet quantified in budgets.
- Marketing teams use ChatGPT or DeepL Write to translate texts and localize campaigns. Time is reduced, but the economic value is invisible at the P&L level.
- Analysts use AI to synthesize documents or explore market trends faster. Individual productivity is high, but no transformation is evident in business results.
In addition, most investments are focused on marketing and sales, while the most profound, lasting, and high-return transformations are in core processes. Core processes are defined as the central and distinctive activities of a company, those that create direct value for customers and form the heart of competitive advantage. Examples:
- Manufacturing companies: design, production, logistics
- service companies: in the financial (credit management, customer services, risk analysis) and health (access to treatment, visit waiting lists) areas
Core processes are distinguished from supporting processes (e.g., IT, HR, administration) that, while necessary, are not the core of value. In this respect, we do not entirely agree with the authors: in fact marketing and sales may be core in those industries or contexts where they contribute directly to revenue generation and customer perceived value (e.g., luxury and fashion, food & beverage, tourism and hospitality).
Instead, we agree that often companies-under pressure to show quick results to stakeholders-invest in Gen AI for marketing and sales (content, campaigns, chatbots). These are useful applications, but in most cases they create tactical efficiencies without innovating the business model.
AI applications on core processes, on the other hand, generate much larger and longer-lasting returns. A few examples:
- Manufacturing: predictive maintenance and quality optimization;
- Retail/Logistics: demand forecasting and dynamic inventory management;
- Health care: clinical support and optimization of hospital flows;
- Finance: risk analysis, fraud prevention, dynamic pricing.
These are more complex projects, requiring more integration and organizational change but, because of this, they create competitive advantages that are difficult to replicate.
The risk, according to Furr and Shipilov, is to encourage unfocused experimentation by repeating the mistake already seen a decade ago with digital transformation (i.e., moving from a company that “uses digital tools” to a company that has digital in its operational and strategic DNA). Recall that digital transformation is about the entire enterprise (processes, culture, models), across sectors; instead Industry 4.0 focuses on factory and production (IIoT, robotics), with specific industrial and, in Italy, also regulatory/fiscal valence.
At the time, the logic of “let 10,000 ideas flourish” generated scattered and poorly strategic projects with poorly scalable results. We can give some examples of these unsuccessful projects to better clarify the two authors’ thinking:
- Showcase business app with no real utility. The app merely replicates the website and offers no distinctive services (reservations, support, order tracking). Result: downloaded by few, soon abandoned.
- Generic customer service chatbot, which neither knows internal processes nor integrates CRM. It ends up answering only trivial questions, generating frustration in customers and additional overhead.
- Social media marketing project without strategy. The company spends a lot on social campaigns without connecting them to the sales funnel, CRM or customer journey. Result: some momentary likes and visibility, but no significant impact on revenue.
- Blockchain for traceability not required by the market. A manufacturer launches a blockchain project to certify the supply chain, but neither customers nor suppliers perceive the value or are willing to use it. The initiative remains a “pilot” without widespread adoption.
- Digital twin of a machine not integrated into processes. The digital twin is sophisticated, but not connected to maintenance or production planning systems. The prototype works technically, but remains confined to R&D with no real operational return.
Disappointment has led many companies to discontinue digital transformation initiatives, reverting to more conservative solutions. The lesson is clear: experimenting is fine, but only in connection with real business opportunities and customer value. Instead, AI is too often presented as revolutionary, but the essential purpose of the enterprise is forgotten: solving real customer problems.
The authors point out four steps to avoid falling into this trap:
- Understanding the role of AI in the broader digital transformation journey;
- Focus on serving customers better;
- Conduct targeted experiments, with attention to scaling;
- scale with dedicated teams (the “ninja teams,” small agile groups tasked with spreading innovations).
Value arises when experiments are directed at solving practical problems and are designed to scale. Indeed, a company’s mission does not change with technologies: it remains to create real value for its customers and stakeholders.
With this in mind, experimentation cannot be left to chance: a company is not a pure laboratory; it is an economic entity. Therefore, experimentation must always have a focus:
- Strategic alignment with clear business objectives;
- Time horizon of pay-back, balancing quick win and transformative investment.
Each experimentation has different levels of risk and expected return. Without focus, experimentation becomes dissipation; with focus, it becomes a powerful engine of strategic learning and economic return.
Widecons® supports SMBs in adopting AI solutions applied to core processes: concrete, scalable, focused and always aligned with the strategic and time objectives of the enterprise.
Learn how to transform AI from a scattershot experiment to a concrete competitive advantage. Widecons® helps you identify, test, and scale the right projects for your business.
Author: Gaetano Rizzitelli