How your AI Initiatives Can Draw Inspiration From an Old Greek Fable.
Artificial intelligence has thrown the enterprise world into a state of frenzy: AI projects are mushrooming, holding C-levels and IT budgets hostage. The block of hardliners still resisting the formidable wave is melting under the influence of enthusiastic social media posts and urging industry analysts. How can one withstand the pervasive pressure without risking being labeled a laggard?
Being the last to adopt a new technology, being resistant to change – what an awful image in a world where digital transformation and change management fly high! Rather jump headfirst into the AI pool than risk being late in the race – the fear of missing out often drives a quick decision, sometimes too quickly. Has someone checked that the pool was filled with water before jumping in? Not always. Due to market pressure and the supervisory board constantly breathing down your neck, the planning phase tends to be reduced to a minimum.). All too often the most important KPI for an AI project seems to be “has started”. Tick.
While delaying the adoption of innovation can be disastrous for your competitive advantage, taking the time to observe and learn can provide critical benefits, such as saving your organization significant time and money.
Have you ever heard of this Greek fable, refreshed by 17th century French author Jean de La Fontaine, The Tortoise and the Hare? It delivers a universal moral that one should meditate before hasting into the next AI initiative: “More haste, less speed”.
Let’s illustrate this proverb with some statistics:
- 74% of CEOs think AI will significantly impact their industry – they were only 18% to think so in 2021[1].
- The average spend for an enterprise AI POC ranges from $ 300k to $2.9 million dollars (note we are just talking about the costs of a proof of concept!)[2].
- Cost estimates for generative AI projects can be off by 500 to 1000%[3].
- 91% of CIOs report that managing costs limits their ability to get value from AI[4].
- 47% of CIOs say their enterprise AI initiatives have not met ROI expectations4.
These numbers give you the picture: Out of enthusiasm and market pressure, generous budgets are allocated to AI initiatives which must start as quickly as possible, at the cost of thorough planning. A very common mistake is to think that developing, integrating and scaling an in-house AI solution is straightforward. The reality looks different: the path to productive in-house AI applications resembles more a pig tail than a straight line. The moment your team begins experimenting with machine learning and engaging with large language models, they step into a highly sophisticated arena that requires not only literacy and skills but also, critically, time and experience.
Time is of crucial essence, because developing an AI solution implies a fair part of trial and error. Time is constantly underestimated in AI projects because of the multitude of factors to be considered to bring an AI application to full production: data availability and quality are often a showstopper, as are security and data protection issues, or insufficient knowledge of business processes, which leads to applications not meeting actual user needs. With time spent on AI projects, experience also arises: the know-how of how to best implement a tool, avoid mistakes, getting better results; the acknowledgment that adapting a pre-built solution might bring a quicker win than developing a solution from scratch.
Experience includes the capacity to better assess the costs of AI projects. Those are overwhelmingly underestimated, for two reasons: LLM pricing is difficult to understand and predict, and LLMs are not necessarily the best technology fit for every use case, leading to longer experimentation phases, poor results and swallowing up too much money in the end.
Meanwhile, industry analysts agree that the systematic use of LLMs represents a problem-technology misfit that some describe as “Pointing teenager technology at grown-up problems”, others as “Shooting at a fly with a bazooka.”
As a C-level executive, what can you do to increase the chances of success for your AI initiatives in 2025? Here is a checklist to pin on the discussion wall for your planning meetings:
1 – What is the expected impact of this AI initiative on our business goals?
It is not enough to align your AI initiatives with your business priorities at a high level, for example by stating “This AI project will help us improve operational efficiency” or “This AI initiative will enhance customer experience”. You must define as precisely as possible the problem(s) you want to solve with AI, including, and this is the most difficult part, a quantification in terms of costs or missed opportunities, and of the concrete benefits or added value you expect from AI. For example, “Currently, the quoting team spends about 20% of their working time (about 20,000 hours per year) manually extracting relevant information from certificates and entering this information into our quoting system. This process is highly inefficient because this mundane task prevents our quoting specialists to dedicate more time to risk analysis or personalized customer interaction. By automating the information extraction, we expect to reduce the time necessary to prepare quotes by 30% minimum, shortening response times to customers from 3 days to 1 day and improving closing rates by a minimum of 10%. We expect a positive impact on revenue in the range of USD [put your estimation here].”
Quantification is difficult, but critical: you should really put a lot of efforts into measuring the costs of your current processes and estimating the added value of AI.
2 – Is the approach we have selected the best fit for our use case?
You must allocate enough time to select the right approach for the problem you want to solve, which means conducting a comparative survey of potential approaches, technologies and solutions before starting the project. For example, there is a difference between generating text (= creating new information, like crafting an email or answering a question) and extracting text (= identifying the relevant bits within existing information). You won’t need the same technology nor the same level of efforts to get good results for both tasks. In the “Build or Buy” discussion, consider this: Building an in-house AI solution makes sense if this solution brings a decisive competitive advantage and/or involves complex workflows that are specific to your organization. If none of these criteria are met, you’ll probably be better off buying an existing solution. If you tend towards building in-house, make sure you address any potential key capability gap by hiring the right skills and/or partnering with third party providers.
3 – Have we conducted an ROI calculation for our AI project that balances both direct and indirect benefits?
While the direct benefits like “Number of hours saved” are easy to quantify, the indirect benefits like “10% less errors in contracts” or “10% more customer satisfaction” are more difficult to translate in dollars. You’ll probably have to begin with estimations and refine them as you monitor results. To begin with, you can search the web for comparable case studies and help tools to calculate the ROI – like this white paper that gives a method for estimating the indirect benefits of intelligent document processing.
Last but not least, make sure your ROI calculation integrates a holistic view of the costs (incl. external costs like inference costs, energy, data preparation, cloud & security infrastructure, and internal costs, mainly the labor costs of your AI team, IT and project management) – Beware of the hidden costs of AI!
4 – Does this project involve all necessary competencies?
Involve business users from the start to validate the business case and ensure the solution aligns with their needs. This is crucial to create AI-based solutions that really add value to the organization. Also involve the IT team to guarantee proper integration into the existing enterprise architecture. If you have enough resources, create an AI Center of Excellence to accompany your AI initiatives, oversee governance and compliance considerations, and induce the necessary changes to make your organization AI-fit.
5 – Do we have an adequate monitoring system including regular milestones and red flags?
Follow the progress of your AI initiative along a set of KPIs and regular milestones. Agree in advance on acceptable deviations and red flags that compromise the viability of the project. Be transparent about the results and accept failure early rather than sinking additional costs.
Once you have thoroughly planned your next AI initiative, there is one more step I recommend taking: Ask your team to think of the worst-case scenario. What will happen if all goes wrong? For example, what will happen if the project does not deliver any satisfying results after one year? Or if your team needs double as much time to develop your application? If the data is not usable as envisioned? If the application does not communicate with your core systems? Developing worst-case scenarios will help you identify potential risks you might have not anticipated yet and improve the planning of your projects.
In conclusion, embrace the path of AI and automation, but take the time to reflect and prepare.
Adopt the cautious approach of the tortoise, who steadily and surely reaches her goal, rather than mimicking the hurried steps of the frantic hare, overly confident of winning the race and underestimating the forces at play.
Not sure yet what kind of added value AI can add to your organization? Discover four use cases from the insurance industry to improve operational efficiency, reduce risks or enhance customer service with this white paper.
[1] CEO & Senior Business Executive Survey – Gartner 2024
[2] 2023 Gartner “AI in the Enterprise” Survey
[3] Published Gartner estimate: Toolkit AI & GenAI Cost Calculators
[4] 2024 Gartner AI Survey
