In 2006, Geoffrey Hinton’s paper “A fast learning algorithm for deep belief nets” famously demonstrated how large neural networks can work. The nets had more layers than previous models — in other words they were deep, which ultimately rebranded the method as deep learning. After years, where the academic community had almost forgotten about neural networks, Geoffrey Hinton brought them back to life. Since then, a lot has happened and nobody is questioning that neural networks and, with them, Artificial Intelligence (AI) are here to stay. Recent advancements include using AI to beat humans in classifying images or beating humans in playing Go. AI evolves fast. Today’s state-of-the-art model may be outdated tomorrow. Interestingly, almost all major breakthroughs were accomplished by academia or academia-like organizations, making AI a technology that continues to be driven out of the academic community.
Companies, on the other hand, have been rather slow in adopting and integrating the new possibilities AI holds for their businesses. While almost all companies acknowledge that AI is becoming a key strategic lever, recent studies by the Boston Consulting Group and the McKinsey Global Institute found out that only about 20 percent of companies have already incorporated AI solutions into their processes in a meaningful way. Among the early adopters, US and Chinese tech giants already invest heavily to attract the brightest AI scientists right out of university, often promising them complete freedom in their research while simultaneously offering very attractive compensation packages.
Typically, German companies have been more conservative when it comes to new technologies, but AI is simply too important to ignore. As a consequence, German industry heavyweights are also building up dedicated AI teams and join AI research networks — the CyberValley initiative in the Stuttgart-Tübingen region being one of the most prominent examples with industry partners like Bosch, BMW or Amazon. However, AI will not only affect Europe’s largest corporations who are capable of hiring large AI teams, but AI will affect every company trying to stay in business.
Especially for the “German Mittelstand” getting started on a highly complex topic like AI poses a serious challenge as neither the starting point nor the adequate approach seem trivial. With the following five-step approach we want to give some guidance on how to identify first meaningful AI use cases within your company. To avoid any misunderstanding, following these steps does not transform your company into an “AI-first-company”, but allows you to identify and test use cases where AI creates most value for your company.
Step 1: “Follow the money” to decide where to test AI pilot use-cases
In case you do not want to develop a companywide all-encompassing AI strategy right away, but you rather would like to get started testing specific AI use cases, you need to define the best starting point for these. While there are many good reasons to test your first AI initiatives distant to your core business, we argue to “follow the money” to identify the perfect AI testing ground within your company. “Following the money” means defining which capabilities allow you to win in the market today (superior user experience, sales, operational excellence, etc.). In other words: Why are customers buying from you today and which processes in your company allow you to have a competitive advantage? Evaluating how AI can help maintain or create competitive advantages in the key processes of your core business will not only yield the biggest impact but will also prove to your employees that AI is a top priority for your business.
Step 2: Understand the status quo, its pain points and your AI ambition level
First, you need to understand the key decision criteria customers apply when deciding on a purchase and how your key processes allow you to address these decision criteria. This analysis will highlight which key criteria are currently not optimally addressed by your offering — your customer pain points. Fixing some of these pain points with AI would already yield significant value, but defining your AI ambition level before doing so promises even more impact.
Defining your AI ambition level means thinking out of the box and evaluating whether AI can help you solve problems that, not long ago, seemed unsolvable. This can mean sketching out the optimal customer journey of the future by assessing which customer interactions can be optimized with the help of AI. Note that customers do not necessarily need to be external, since also support functions (HR, Finance, etc.) serve internal customers. Thus, developing your optimal target customer journey (=status new) ensures that you do not only optimize the status quo, but that you also take into account future developments. Ultimately, all three analyses (status quo, pain points, status new) are necessary to clearly define the output parameter (e.g. lead conversion, scrap rate, etc.) AI is supposed to optimize when developing AI use-cases in the next step.
Step 3: Develop AI use cases based on customer needs
As discussed in step 2, AI use cases should address concrete customer pain points and help companies maintain or develop competitive advantages. Since this may sound a little technical, let’s have a look at a concrete example from the insurance industry.
Premiums for your car insurance are currently based on a few key variables such as age or the neighborhood you live in. Let us assume that you are young, a very conservative driver and live in a densely populated area with rather narrow streets. The chances that you will end up paying a high premium for your car insurance are very high as traditional insurance companies do not take into consideration how good a driver you are. Luckily, conservative drivers suffering from high policy premiums can switch to a more fairly priced insurance product as innovative start-ups enter the market. Zendrive, for example, charges its customers based on their driving style by including additional data sources into the premium calculation and leveraging machine learning to offer personalized insurance policies.
When trying to brainstorm how AI can improve your product offering, a good starting point is by looking at problems AI has already mastered quite well. AI is particularly suited for any kind of pattern recognition, which particularly allows AI algorithms to help companies in the following three categories:
A. Prediction: For instance predicting when a machine needs maintenance in order to prevent loss of production or predicting when a customer takes some action.
B. Automation: For instance automating repetitive tasks such as invoice processing
C. Classification: For instance optimizing the quality assurance process by classifying manufactured goods into fault-free products and rejects
Step 4: Prioritize use cases based on business value and implementation complexity
When it comes to prioritizing use cases, one should plot all identified use cases on a two-dimensional matrix with business value and implementation complexity being the two assessment criteria on the axes.
Applying AI is not an end to itself. It is probably neither a good idea to marginally optimize the accuracy of your demand forecasting with the help of a sophisticated deep neural network nor does it make sense to offer a new AI-powered functionality if you know that customers are not willing to pay for it. AI projects can have a different setup and timeframe — they should, however, always have a realistic positive business case. To pressure-test the assumptions in your business case, you should always engage in small scale and agile user testing before building a fully-fledged AI solution.
Apart from classic considerations such as number of stakeholders involved, in the case of AI, implementation complexity is always linked to the availability of data. For many use cases the required data sets are simply not available or need to be prepared through a lot of manual work. Ranking your use cases on availability of data does not only allow
you to prioritize first AI use cases, but this ranking exercise can also serve as the first input for an equally needed forward-looking data strategy.
Step 5: Start today
The first mover advantage is particularly important in an AI world. Oftentimes you can already launch an AI product with a rather small set of data allowing you to onboard first users, who in turn give you more data, which can be used to improve your product and thus attract more users again. It is dangerous to believe that companies will be able to catch up on the AI topic in a couple of years as the learning part of machine learning is a clear paradigm shift. Unlike in the classic software development world, companies will not be able to instantly gain a competitive advantage by coding and releasing a better version of their product, anymore. In an AI world offering a trained and calibrated product will be the differentiating factor. Developing such a well-trained and calibrated product usually starts with the availability of data. As in most companies data sets are rarely pooled nor standardized, developing a data strategy should be one key priority today.
When looking for a partner to support your company in this five-step journey, you should look for a provider with the following three key characteristics:
1. Deep expertise in AI: As explained in the first paragraph, the field of AI
is evolving so rapidly that basically only scientists or experts closely
associated with the scientific community can give you a good overview over the
current and future state of AI. Do not trust an “AI expert” whose most recent
publications date back to 2013.
2. Methodological and project management expertise: These are the classic top-management consulting skills. Your partner should propose a clearly structured and battle-proven methodology how to identify, test and implement AI use cases.
3. Domain expertise: Ideally your partner should have expertise in your specific domain as well. However, this expertise may be the most dispensable of the three as you yourself can bring this skill to the table
Hiring a service provider that checks off only one of these three characteristics will most likely leave you unsatisfied. Neither will a classic consulting firm smattering on AI be able to explain the latest AI innovations to you, nor will approaching a university and collaborating with its scientists alone yield optimal results. In fact, the combination of expertise may be the secret sauce to successfully master your specific AI challenges.