The Three Buckets Framework: Bucket #3 - Digital Representations of Assets
After 25 years working with data-fuelled business development, one pattern keeps recurring in the current market: the AI technologies that generate the most boardroom discussion are rarely the ones generating the most business value. Bucket #1 - Generative AI is loud. Bucket #3 - Digital Representation of Assets, is quiet, but in many industries, this is where the real money is.
From insurance mathematics to modern AI
My career started in the insurance industry. Insurance is a bet, a numbers game, which makes it a masterclass in analytical discipline. Long before machine learning entered the mainstream vocabulary, we were calculating and simulating assumptions, risks, and outcomes at scale. In that context, AI is not something entirely new, but rather the next generation of the same analytical capability, with considerably more muscle.
When I worked in insurance, unstructured data was largely inaccessible for analysis at any meaningful scale. Now, though, with LLMs (large language models) and LVMs (large visual models), both text and images can be analysed, modelled and utilised for a wide range of purposes. Most of all, machine learning adds a set of strong analytical muscles to the massive numbers game inherent to insurance.
Insurance, finance, and telecom are three mature industries with an abundance of data representing all aspects of the business. However, most other industries also have significant and growing volumes of data about everything from products, services and employees to operational processes. In short, digital representations of any and all assets in a company are or can quite easily be established.
These data are not part of the training data for the LLMs. From an analytical point of view, the value of the data primarily lies in good old-fashioned structured analysis and modelling. In other words, mathematics. AI in the form of machine learning is a useful tool**, and with enough data,** these digital representations, also known as digital twins, can be built to simulate, analyse, and optimise processes and assets. Examples may be recommendations on eCommerce sites, listening recommendations in Spotify, load balancing of networks, fraud detection in payment services, fuel reduction in shipping, production stop prevention in production lines, weather forecasts, etc.
The three buckets framework
A while back, I wrote an article about the Three Buckets Framework for constructive AI discussions.
My premise is that in order to have constructive and useful discussions about AI's role in strategy and risk assessment, the contributors to the discussions must have a shared conceptual understanding of AI. To facilitate this, I created a very simple framework of three buckets, where each bucket contains a subset of AI technologies, accompanied by the strategic value, risks, regulations, and complexities relevant for each bucket.
Bucket #1contains personal productivity tools.
Bucket #2 contains AI agents.
Bucket #3 contains digital representations of assets, whether they are virtual or physical, or objects or processes. This article focuses on this third and final bucket, Bucket #3: digital representations of assets.
What should the board understand about the capabilities of digital representations of assets?
Contrary to the first two buckets, generative AI is of limited importance to this bucket. The main value lies in all the data points about your company's assets, such as the driving pattern of a construction-site excavator, monitoring of production-line stoppages, passenger shopping and movement at airports, city centre traffic jams, cell tower load patterns, payment fraud patterns, and insurance underwriting.
Imagine if you could shorten the time to completion on the construction site through smarter utilisation of the excavator, anticipate production line fault stops, increase revenue through smart store location, reduce rush-hour traffic jams, make load-balancing more resilient and power efficient, reduce the number of fraudulent claims, or increase insurance revenue whilst maintaining the number of customers? You can. If you know your operational processes and where to implement which improvements. This is where good old predictive AI, aka machine learning, comes in. The prerequisite is that you have relevant data in sufficient volumes of useful quality.
It's important that the board also understands that the toolbox for extracting value from digital representations contains a number of different tools, such as:
Classical statistical analysis
Optimisation algorithms (typically belonging to the domain of Operational Research)
Computer vision
Machine learning
The methods used are data engineering, mathematics/statistics, and software programming. Not prompting.
The environmental aspect is also of importance. Generative AI is notoriously known for being computing-intensive, fuelling a massive data centre growth with devastating environmental implications. Non-generative AI, though, requires orders of magnitude less computing power, and can, in most cases, be both trained on and run from a local computer. In fact, many such models are already running on your own computer and smartphone, as part of the applications you use on a daily basis.
What should the board understand about the risk of digital representations of assets?
Within data governance, the good old saying "garbage in, garbage out" still holds. With AI applied to the mix, the phrase needs modification: "garbage in, garbage factory out". The key message here is that data integrity and quality are paramount.
Any productive discussion at board level about how to enable strategic moves or improve operations through applying mathematics to digital representations requires at least a basic knowledge of the company's processes and data. The alternative is wishful thinking with a very high price tag and low probability of success.
Non-generative AI is in no way exempt from regulatory requirements. On the contrary, the EU AI Act was initially designed to regulate non-generative AI. However, the information security risks are less problematic for non-generative AI than for generative AI. "Hallucinations" and prompt injections, for example, are features of the transformer technology used in generative AI, and are not relevant for non-generative AI.
A key risk to understand is the cost of making the wrong decision. Almost all methods under the AI umbrella are based on predictions. Outputs of prediction models come with a probability of being correct, preferably as high as possible. However, that also means that every outcome has a certain, albeit hopefully small, probability of being wrong. For a board director, this may become directly tied to legal liability.
A well-known example is the Dutch childcare benefits scandal, where thousands of families were wrongly accused of childcare benefit fraud. In that case, the cost of being wrong was extremely high. Families ended up in severe financial distress, and some parents died by suicide. The related Dutch SyRI case also illustrates how algorithmic welfare-fraud detection can raise serious human rights, privacy, and governance concerns.
Not all decisions carry that weight. At the other extreme, Spotify serves as a pedagogical example; how much does anyone care if a listening recommendation in Spotify is wrong? The only consequence is potential loss of streaming revenue.
From a board perspective, the actual implementation of AI may be too operational to deserve a place on the board's agenda. It is important, though, that the board defines a framing for the extent and scope of decisions made by these mathematical prediction models.
What should the board do about digital representations of assets?
The use of mathematics to analyse, optimise, foresee, prescribe, and generate is the natural next step of the digital transformation. Data is - or can easily be made - abundant, processing power is readily available, and the tools and methods exist. They have done so for decades. Whichever industry your company is in, there is value to reap by understanding and utilising your assets and processes better or in new ways. The AI toolbox offers a powerful set of muscles to do this. This is AI, but has nothing to do with ChatGPT.
When discussing strategic and tactical issues, start with the outcome you want, not the technology you've heard about. What do you need to know about your operations that you currently don't? That question is more valuable than starting with any technology currently being marketed to you.
What should the board expect from digital representations of assets?
When clear desired business outcomes meet data, mathematics and deliberate process design, value creation, innovation, and disruption may happen. Consider the examples already mentioned in the sections above, and you quickly realise that the potential ROI are more than significant. However, these initiatives also require structured approaches to data management and integration of AI models into operational systems. The price tag may therefore also be significant.
Where generative AI still has a role in Bucket #3
The distinction between generative and non-generative AI is not as clear-cut as this article may indicate. LLMs can be supporting tools for prototyping and analysis of digital representations of assets and processes. To some extent, LLMs may serve as a natural language interface between the human intention and an overwhelming data set, enabling quick first-draft hypothesis testing. LLMs can also structure textual data into structured formats which can then be analysed with structured methods.
But be warned, the use of generative AI to access and process structured data for analysis and modelling requires thoughtful choices of model design, parameters, and trade-offs between different aspects. Understanding context, assumptions, and limitations is essential. You cannot trust an LLM-based agent to do the modelling for you. This is where you need data engineers and data scientists with, or complemented by, in-depth knowledge about your operations and assets.
Why Bucket #3 may be your most valuable bucket
The irony is that most companies are already sitting on the data that Bucket #3 requires. They don't need to buy it, scrape it, or build it from scratch. It exists in their operational systems, their sensors, their transaction records, their supply chains. The question is not whether the data is there, but whether and how it is being put to work.
In my work as an advisor and board director, I see this gap repeatedly. The AI conversation is dominated by generative AI. This is understandable, given the powerful capabilities of language models, the availability of the tools, and the intensity of the media coverage. But the consequence is that Bucket #3, the most mature, most commercially proven, and often highest-ROI form of AI, stays below the strategic radar.
The potential for European businesses to utilise their own internal data is still to a large extent untapped. That represents both a risk and an opportunity, depending on which direction your competitors are looking.
If your board is not yet asking what your operational data can tell you about your processes and assets, start there. That question is worth more than any vendor pitch about generative AI.
Feel free to contact me if you want to continue the conversation.
About the Author
Elin Hauge is a keynote speaker, AI strategist, and trusted advisor to business leaders and boards. She specialises in helping organisations make sense of artificial intelligence beyond the hype, connecting technology to strategy, governance, and real-world value. With a multidisciplinary background in physics, mathematics, business, and law, Elin brings both analytical rigour and practical perspective. Her talks and advisory work empower leaders to ask better questions, make wiser decisions, and navigate AI with confidence.
Frequently asked questions:
-
Digital representations of assets are data-based models of physical or virtual assets, objects, processes, or operations. They can be used to simulate, analyse, and optimise business performance through methods such as machine learning, statistical analysis, optimisation algorithms, and computer vision.
-
They help boards focus on business outcomes rather than AI hype. By using operational data, companies can identify opportunities to improve processes, reduce risk, optimise assets, detect fraud, prevent failures, and create measurable business value.
-
No. Bucket #3 is primarily about non-generative AI, including machine learning, statistical analysis, optimisation algorithms, and computer vision. Generative AI may support prototyping, analysis, or structuring textual data, but it is not the main source of value in this bucket.
-
Boards should understand risks related to data quality, data integrity, regulatory compliance, model accuracy, and the cost of wrong decisions. Prediction models always carry some probability of error, so boards need to define the appropriate scope and limits for decisions made or supported by AI models.
-
Companies should start with the business outcome they want to achieve, not with the technology. A useful first question is what the company needs to know about its operations, processes, or assets that it currently does not know.