A Framework for Creating Business Value with AI
How artificial intelligence can empower any business process.
Everyone is talking about artificial intelligence (AI). It’s the buzzword of the century. Some are excited, others are concerned. Some have a good grip on it, others feel lost. Even the most mainstream of news publications, streaming services, and radio shows are discussing artificial intelligence in some form or another. Countries are defining nation-wide AI strategies, with some stressing AI as the most important technology of today. Vladimir Putin, president of Russia, went as far as to say that the nation that leads in AI ‘will be the ruler of the world’. As people are experiencing the dreaded fear of missing out, AI appears to be increasing in urgency, as early adopters have much to gain. Everyone seems to want AI. Yet most don’t really know why, and certainly not how. In this article, I hope will be able to help you answer those questions.
Everyone seems to want AI. Yet most don’t really know why, and certainly not how.
I won’t waste your time talking about the infinite possibilities of AI, and the incredible solutions that have already been implemented by organizations small and large. If you’ve made it to this article, you’re probably already interested in creating value with AI. Perhaps for your own organization, or perhaps you intend to sell your services as a consultant for others. No matter what, if you want to become successful with AI, you must always focus on the value that AI can bring — not the tools themselves. And so in this article, I will present a concrete framework that will show you how to discover that value.
- Definitions. First, I will spend a small amount of time discussing the definition of AI in the context of the framework.
- Origin. Next, it is in order to explain the origin of the framework, and how I first came into contact with it.
- Framework. This is the key part of the article. Here, I will walk you through the framework step-by-step, covering every detail.
- Usage. After presenting the framework in detail, I will carefully explain how to best use the framework.
- Afterword. Finally, I will end the article with some advice, and suggest some next steps to take.
Let’s begin with the necessary headache of any concrete AI discussion: the definition of AI. As you’re probably well aware, artificial intelligence has no widely accepted definition. On the contrary, it’s a buzzword shrouded with vague descriptions and harsh misconceptions. In truth, it’s a term that could encompass just about anything. Thankfully, this framework is so adaptable that the aforementioned description is sufficient. You can define AI almost any way you want, and this framework will still be applicable. In fact, the framework has been made for AI in particular, but can also be used for digitization in general. The one thing we must agree on, however, is that artificial intelligence can improve any given process through either automation or augmentation. It is a common misconception that AI is synonymous with automation, but it very important to acknowledge that augmentation is an equally large part of AI today.
Automation means to remove a human from a process. Augmentation means to empower a human in a process. A process can be any activity. Processes can always be broken down into more specific sub-processes (i.e. “living” -> “working” -> “human resources” -> “recruiting” -> “interviewing”). The more specific a process is, the easier it will be to map it onto the framework. This will make more sense in a bit.
Automation means to remove a human from a process. Augmentation means to empower a human in a process.
The origin of the framework
I first discovered this framework when I was writing my master’s thesis on the subject of creating business value with AI in the context of enterprise resource planning (ERP) systems (off-the-shelf (pre-made) IT systems that can span an entire organization). The framework was first made by the consultancy firm Accenture in 2016. Thus, the framework is not the result of proper academia, but of corporate research. There are two important things to note when dealing with corporate research:
- Unlike academic research, corporate research does not need to be peer-reviewed. In fact, corporations can write whatever they want.
- Corporate research is often a simplification of reality. Corporations print research papers to strengthen their brand, and they often write it simple enough so that their customers can get a rough understanding of the topic (or just enough that their customers will be impressed).
When I was writing my master’s thesis, however, I needed a solid framework upon which I could test my theory. After conducting extensive research into the papers of the academic world, I came up short of any AI frameworks that could be applied on an organizational level. The academic world was clearly too far behind the corporate in this area. Yet I did find a lot of interesting papers from some very famous AI researchers — most of them from MIT — and all of their theories overlapped with this framework. After strengthening Accenture’s framework with academic research, I was able to convince the academic world of its veracity, and was able to successfully include it in my master’s thesis. Since then, I have used the framework extensively, to great success.
Below is the original framework, as created by Accenture:
And here is my version, which I will be using as I present the framework:
Now, let’s go through this framework, step by step.
Presenting the framework
The framework was created for analyzing what business value can be created with artificial intelligence. It is a first step that helps you determine which strategy you should adopt for any given process. It does so by allowing you to place a process in one of four models, determined by two variables: data complexity and work complexity.
Data complexity is the easiest of the two variables to explain. Data with low complexity is typically structured and simple: it is usually strings of text or numbers. The data is easy to interpret for a computer. High data complexity, meanwhile, is often unstructured and up for interpretation. Images, videos, music, and voices, are examples of complex data.
Work complexity is mostly about routines and rules. If a work has clearly defined rules and routines, it becomes predictable, and has low complexity. If work is unpredictable and ad hoc, however, it requires judgmental skills, which would result in high work complexity. Computers are excellent at predicting, but not so good at judging.
Work complexity works like this. Processes consist of tasks. Most tasks have a predictive step and a judging step. For example, driving a car is a process, and making a left turn is a task. The predictive step is turning on your blinker, and checking whether or not it is safe to turn left. The judging step is the act of determining whether it is safe or not to make the turn. Computers are much better at humans at the first step, but generally not so good at the latter step, especially in more complex processes. Tasks that contain a difficult judging step are usually of high work complexity.
It is important to note that time is not a factor in determining work complexity. A process with low work complexity could take a year to execute, while a process with high work complexity could take just ten seconds. What matters is whether or not the process can be handled without, or with very little, human input.
If a process has low data and work complexity, it is suited to be automated. If a process has high data and work complexity, it is suited to be augmented. We can update the matrix above with these two actions.
Again: automation means to remove a human from a process, while augmentation means to empower a human in a process. The further to the bottom-left a process is, the more it can be automated, and vice versa. The framework doesn’t just stop there, however. Depending on where a process is placed in the matrix, we can empower it using one of four models: efficiency, effectiveness, expert, or innovation. You can think of these as four different AI strategies.
If a process has low data and work complexity, it can most certainly be automated. This is the simplest model, and the one that most companies are trying to adopt today. It is a clear first step for any company wishing to start their AI journey. Efficiency is all about optimizing your business, generally reducing costs. Processes that are placed here have very clearly defined rules and routines.
- Automated credit decisions based on any number of data variables.
- Automatic purchasing of common products to your warehouse from wholesalers depending on your stock level, upcoming events, the season, and any other number of factors.
- Automated recruitment. By having AI bots analyze LinkedIn profiles, you can automatically invite appropriate candidates to interviews, with personalized invitations.
- Any form of text can be written automatically, be it a content description, a summary of an article, or even a fictional novel. Likewise, there are already tools out there to automatically create videos for articles, or to create music and art.
- Automatic package delivery, be it through self-driving trucks, drones, or otherwise. Warehouses can also be automated, with robots packaging products to send to consumers or stores.
Some companies take efficiency to entirely new levels, automating most, if not all, of their business. For instance, the founder of the massive Chinese e-commerce JD.com hopes to fully automate his entire business in the future. Some smaller companies with smaller value chains have already done just that.
If the work is simple, but the data is complicated, you will want to focus more on effectiveness rather than optimization. This model revolves around the communication and coordination of workers, with AI serving the role of an assistant. Processes placed in this model are used to make the work of employees more effective, by either eliminating or simplifying the act of scheduling, communicating, or monitoring.
- Automatically scheduling meetings for employees based on their availability or preference, simplifying the act of communicating. Meeting notes can also be generated automatically, by first transcribing the meeting, and then having AI create a summary.
- Some meetings can be avoided altogether, by having AI observe human workers, and explain their work to relevant team members, at a level they understand. For instance, experts and novices in the same field of work may have different levels of understanding of the same topic, a gap which AI can help to bridge.
- Automatically determining which consultant is a good fit for a particular task. In larger consultancy firms, it can be difficult to know exactly which consultant to send for any given project. AI can help by selecting consultants based on their experience level, preferences, goals, availability, and cost.
- Automated customer support. A classic example that has been widely implemented already. If a task is too complicated, a real human agent can join the support errand.
- Ordering through natural voice. Rather than calling a human to order a pizza, a person can simply communicate with a virtual assistant (such as Google Assistant, Facebook Messenger, Siri, or Amazon Alexa), after which the humans making the pizza will receive the order.
Very few companies actually utilize the full potential of the effectiveness model. Companies that work extensively with this model will find their employees spending more time on meaningful work, and less time on communicating and coordinating predictive work-related errands. Out of fear of misunderstandings: this is obviously not about stopping employees from communicating with each other, but merely to streamline the boring, slow, or difficult parts of communication.
When the data is simple, but the work complexity is high, AI can be used to leverage expertise. Processes placed in this model relies heavily on expertise and judgment. Unlike the left-most models, any decision made here is taken by humans, with AI offering a supportive role, opting to offer advice and insights. In this model, humans are always in control. Expert systems deal with anything from large amounts of money to human lives, meaning humans must always be responsible for decision-making.
- Medical diagnosis. By entering the symptoms of a patient, an AI-powered system can suggest potential causes of the patient's illness, and recommend appropriate treatments.
- Selective purchasing can give stores a competitive edge. Niche shops that sell exclusive items may use expert systems to find products that will appeal to their customer segment.
- Financial systems can assist investors and economists in making decisions such as financial investments.
- There are online travel fare aggregators who use AI to predict how the price of flight tickets will change over time, suggesting its users to buy now or wait for a lower price.
As routine-based jobs become automated through the efficiency and effectiveness models, companies will need to help employees gain the skills needed to execute more complex jobs. This can be accomplished by investing in the expert model. By making complicated jobs easier, employees can more easily make the transition from performing rules-based jobs to executing unpredictable jobs instead.
Processes that have both a high data and work complexity can be used to enable creative work. Humans are in complete creative control, while AI is used to identify recommendations and alternatives. As you might imagine, this model is particularly interesting for jobs that are creative in nature, such as designing, researching, writing, cooking, composing music, drawing, filming, and so on. Yet most jobs actually contain some amount of creativity. For instance, every time a person writes a text message to a friend, or has a conversation by the coffee machine, they are actually executing a creative task.
- While conducting an interview with a job candidate, an HR employee can be augmented with an AI that suggests follow-up questions to ask the candidate in real-time, based on the conversation the two are having.
- An AI might recommend a music composer to add a certain bass to a song depending on the other instruments the composer has added.
- When a user writes some piece of text in a Microsoft Word document, the software may recommend synonyms or alternative phrases, augmenting the person writing.
- A person in the midst of editing a video could receive a recommendation from an AI, suggesting appropriate music or cutting for a scene.
- Corporate management can be empowered to run their business better, suggesting ways forward to increase revenue. This is perhaps the ultimate form of AI — empowering the leadership of an organization to realize unexpected improvements to the value chain.
As the most complicated and least adopted of the models, companies investing in the innovation model can experience an enormous early advantage and a massive return on investment. Realizing the unexpected strategies this model can provide is not easy, but definitely worth attempting.
Companies investing in the innovation model can experience an enormous early advantage and a massive return on investment.
How to use the framework
The framework is an excellent first step in identifying how AI could empower any given business process. Find a piece of paper or a whiteboard and draw out the framework (or download an image of it from this article). Map out your processes onto the framework. If you are uncertain where a process fits into the framework, you are probably looking at a framework too broadly, and need to split it into sub-processes. Also, a process that fits into a certain model for one of your competitors may not fit into the same model for your business. For instance, you may have noticed that I wrote “purchasing” under example applications for both the efficiency and expert models. This is because the process of purchasing items from a wholesaler could vary depending on your business model.
I must also mention that while this article has focused on applying processes to the framework, you could also use the framework to map out tools. When looking at value creation, though, you should never focus on the tools themselves, however you may later need to do so when selecting a tool for value creation.
Furthermore, it is important to note that the definition of “complex” is ever-changing. A process that was once considered to have a high data or work complexity may now be considered to be simple. Over time, processes will move from top to bottom and from right to left — but likely not vice versa. For example, financial investment jobs are typically placed in the expert model, but there are already investment funds that are fully managed by an AI, which would belong in the efficiency model. Similarly, educational and medical processes may find themselves moving into efficiency, as AI could potentially execute such processes automatically. It is likely that there will be two separate forms of healthcare, for instance, one that is simple enough to be executed fully by AI, and one that requires human expertise.
In order to decrease the work complexity of a process, you must identify routines or create rules for that process. In order to decrease the data complexity of a process, you must create (or purchase) better algorithms. The innovation model, however, will be the last to be automated. In fact, even if our entire society were to become automated in the distant future, it is very likely that services in the innovation model will be more popular than ever, as a larger amount of the population will be able to fulfill their aspirations of producing meaningful, creative work.
A process that was once considered to have a high data or work complexity may now be considered to be simple.
This framework is not only an excellent first step into creating value with AI, but a valuable tool to have alongside you during your entire AI journey. Always work towards creating a proof of concept. When building a vehicle, don’t go straight to building a car. Make a skateboard first, then a scooter, then a bike, and so on, to incrementally create something that provides actual value. You need to identify the business processes to focus on, and develop in-depth strategies for how to elevate AI for those particular processes. Applied properly, you could even redefine your entire business model, or your entire value chain, with AI technologies. What’s important is to understand the four different models, and how they can automate or augment your processes in entirely different ways.
I hope you found this article helpful in your journey to creating business value with AI. Feel free to reach out to me by posting a comment below, or by messaging me on LinkedIn, should you have any questions.