Getting the most out of AI implementations: 6 dimensions of value creation

Calculating the expected return on investment of AI projects is a challenging task that often leads to delays in or deferral of investment. That is a pity because AI has much value to add.

AI applications are so broad that finding a unifying approach to estimating its added value is hard.

AI is a so-called “general purpose technology”. It aims to translate real-world tasks into mathematical ones and solve these new problems using computers.

AI’s biggest strength and weakness is that numbers can represent anything, and many seemingly completely different problems (like identifying one’s personality or predicting the weather tomorrow) can thus be solved using a similar approach.

The proposed framework consists of four steps:

1. Identify the impacted processes;
2. Perform what-if analysis in 6 dimensions;
3. Estimation of added value for each of these dimensions;
4. Pivot your business model towards scalable value creation.

Step 1: Identify the impacted processes

We advise moving away from calculating the added value for a specific AI application but to start from the processes that will be impacted and might benefit (or suffer).

This step is vital because AI often “cuts through” silos in an organisation, and there may be many hidden costs as well as benefits:

• Data collection at another department may need to change (e.g., clean a database or collect new pieces of information);
• Suppliers may need to adapt their way of working (e.g., deliver data or provide feedback through an API );
• Innovative business models may impact internal costing methods (e.g., an activity may be loss-making but generates valuable data that can be monetised);
• Improved insight from operational data may lead to better decision-making at a strategic level;
• Automation of a task may lead to new business cases.

Step 2: What-if analysis among six dimensions

The second step is to organise a brainstorming with a diverse group of operational, strategic, open-minded, creative, and critical-minded people. Attempt to include people of different ages, genders, backgrounds and familiarity with the process.

For each of the following six dimensions, perform a “what-if” analysis to picture the “to-be” scenario and contrast it with the “as-is” situation. As you will notice, we position the dimensions at a “meta-level”, which should not come as a surprise as AI is a general-purpose technology.

Economies of certainty

The first dimension — too often overlooked — concerns the question, “What if we had full certainty/knowledge about…”. Ensure you “idealise” the situation to reach strategic AI-first visions. This aspect may be relevant to your case when:

• You are often caught off guard by unexpected situations;
• There is a too large variance in the quality of products or services;
• It is hard to predict how long something will take, how much it will cost;
• Standardisation efforts do not take off;
• There is a mismatch in the offer and demand, too much slack, over/under capacity;
• You over-rely on educated “guess-work”.

A good way to get the discussion started is to find “blind spots”, internally and externally.

Economies of performance

Typically, economies of performance are the first to come up in people’s minds when dreaming up the advantages of AI, reflecting the question: “What if we had a perfect performance for task X?” In other words, you attempt to do what you already do, but better and faster.

There is a catch, however. When you start from existing processes, the danger is that you do not look beyond the direct impact and miss the true potential of AI. Drastically improved or faster performance can lead to new markets or business cases (e.g., being able to monitor all conversations in a call centre opens up the possibility of detecting or even predicting sudden drops in quality and planning preventive actions)

Economies of scale

One of the evolutions that AI is bringing about is the “inversion of the operational pipeline.” Currently, processes are mainly executed by humans who use IT systems to support them. In a fully digitised setting, algorithms orchestrate service delivery, and humans are asked for support when needed. This extreme automation is a prerequisite to achieving massive scale and has profound consequences for organisational processes.

In this dimension, we ask, “What if we could scale this process or task indefinitely at quasi-zero cost?”

When picturing the to-be situation, make sure to “break free” from the current organisational structures, as there will be a change in processes, roles and tasks that humans will perform, and business models (e.g. adding business requires fewer new people).

Economies of complexity

Humans are incredibly smart creatures, but also have their limitations. We are bad at thinking in high-dimensional spaces and understanding nonlinear effects or delays. Economies of complexity tackle the question: “What if we could handle the most complex cases we currently avoid?”

The added value lies in going beyond current capabilities. Ideally, tackling more complex scenarios scenario leads to better insight for humans and a better understanding of your business, which in turn leads to an improved design of the software systems, resulting in a positive feedback loop.

Economies of scope

Two questions are addressed: “What if we could reuse solutions for different use cases?” and “What new possibilities open up with AI?”.

As AI systems implement solutions to template problems (e.g. predict the next item in a sequence, where this sequence can represent any concept), one can transpose an existing solution to other contexts. The digital nature means that this typically can happen faster than traditional human-driven processes ever could.

One can also start bottom-up from existing AI tasks to discover new applications and brainstorm how they could lead to new value creation. Many techniques are less “popular” but yet very powerful, like logic reasoning or graph algorithms. However, we suggest starting the other way around till a proper AI mindset has permeated the organisation.

Economies of action

Having digital processes in place that are delivered fully automatically opens up the ability to perform continuous experimentation.

This approach goes beyond “basic” optimisation because experiments happen online and in real-time. They thus interact with the service delivery and measure the impact in vivo rather than in pilots or laboratory experiments.

Step 3: Quantify costs & benefits

Finally, quantify the expected investment and added value. Ensure to document your assumptions, the parameters your analysis depends on, and your metrics. Crucially, remember to monitor these during implementation!

Benefits

As explained throughout the text, benefits can come from all six dimensions:

• Reduction of slack, waste and overhead due to uncertainty;
• Better service due to improved planning and certainty
• Improved efficiency and productivity due to automation;
• Access to new market segments (because of improved performance);
• New business cases due to economies of complexity & and scope;
• Extra business due to economies of scale;
• Increased operational excellence, faster innovation and deployment;
• Increased margins due to continuous experimentation.

Costs

When quantifying the costs, take a full perspective on your business, reflect on what makes you unique, and embrace open innovation. You need to understand what is happening in the ecosystem (this can be outside your sector) and across the value chain.

For example, you could train an AI model on one of your datasets. But is this the best data in the market? Can someone with a more extensive dataset outcompete you? Or can a competitor outcompete you by doing smart acquisitions? Are you willing to stay a leader by investing in the best information driving your models?

Also, be careful to account for any technical debt you introduce, the necessary technological changes, and the impact on staffing (who do you need to hire? How will existing jobs change?) Pay attention to the needed change management to make the transformation.

Calculate the sampling cost for collecting one data point and find out if you can use or buy existing datasets (or make particular acquisitions) to bootstrap learning.

Finally, remember that algorithmic systems need proper quantitative/technical and qualitative monitoring to ensure responsible implementations and avoid degradations in performance.

Step 4: Pivot your business model

The economies of AI require a considerable capital investment. To make them profitable, often new business models are needed. Activities that previously were profit-making may become loss-making, and new opportunities will arise.

An illustrative example is John Deere. Traditionally, their main activity was manufacturing — and their main product, tractors and other agricultural equipment. Quite some years ago, they started equipping their machines with sensors.

This move allowed them to collect valuable intelligence about farming: what are optimal strategies for fertilising? How are farmers currently farming? How do crops grow in different places? How does weather influence harvests?

This information may be worth more than enough to offset potential losses in the sales of tractors.

Manufacturing, previously a profit centre, has turned into a cost centre. Their products now serve as a “mine” for data that can be monetised at a much higher price.

In conclusion

It is up to you to decide whether you are the best party in the market, with the best access to data & knowledge, and whether you can and are ready to scale and reorganise your business to make the investment profitable and stay ahead of the competition.