AI in Business
Even though tremendous economical and scientific relevance is attributed to both, general Artificial Intelligence
(AI) and machine learning in specific, it is still not enough for businesses to implement these technologies. Matthew
Evans, vice president of digital transformation at Airbus, puts it like this:
"Well, strictly speaking, we don’t invest in AI. We don’t invest in natural language
processing. We don’t invest in image analytics. We’re always investing in a business problem."
In other words, AI is a tool that should be utilized when the implementation costs are outweighed by the expected
gain in value. This said, there are three business areas that can be addressed by AI.
AI can (semi-)automate business processes. For example, an image recognition
software for cancer detection could pre-classify the obvious cases, freeing up radiologists’ time to focus on more
AI can support decision-making by gaining insights through data analytics. In
a world of increasing competitive pressure, excelling in analytics could be a point of differentiation. However, it
is important to point out that data analytics does not generate business value itself. It can only provide a better
understanding of performance and reveal opportunities to improve profitability. Ultimately, the onus is on the top
management team to leverage these insights to gain business value.
AI can create new products, services, and even new business models, by engaging with
customers and employees. Especially when AI is applied to customer engagement, it is important to address
explicit or implicit customer needs – the basics in building a business.
In practice, AI is applied in all three areas of business needs, with a little more emphasis on enhancing features,
functions and performance of existing products. The question remains, why companies and executives are interested in
AI in general. In fact, a global survey by MIT Sloan Management Review of over 3,000 business practitioners found
that, despite the hype around AI, more than the half of the respondents had no adoption of AI at all. Nevertheless,
83% of the participants perceive AI as a strategic opportunity for their organization, while 37% also recognize the
risk of missing out. Most of them feel pressure to adopt AI from all dimensions. Customers and suppliers will ask for
AI-driven products and services, incumbent competitor will use AI (e.g. for cost efficiency), and new market entrants
will use AI and substitute them. In a nutshell, most of the respondents (84%) believe that AI will allow them to
obtain or sustain a competitive advantage.
In summary, to obtain or sustain competitive advantage, most businesses perceive it as crucial to exploit the
opportunities of AI. Of course, not everyone needs to be an AI expert, but for decision makers it is definitely an
advantage to understand the basic principles. To put it in the words of Brynjolfsson, E. and McAfee (2017):
“Over the next decade, AI won’t replace managers, but managers who use AI will replace
those who don’t.”
Technical, Managerial and Organizational Challenges
Companies want and should learn about AI, experiment with its possibilities, and deploy pilot projects. However,
there are some specific technical, managerial, and organizational challenges for applying AI in a business context.
From a technical perspective, AI, or specifically machine learning, is mostly based on statistical modelling, which
again is based on mathematical theories grounded on precise and explicit premises. These premises often cannot be
proven to hold in the real world but are pragmatically assumed to hold due to the lack of alternatives. For example, a
fundamental assumption is that the sample data, on which the models are trained, is “representative”. That means, they
assume this small sample shows similarities in terms of values, combinations, and their distributions, with the real
population. Therefore, conclusions about the sample can be extrapolated to the whole population. The hope is that
those insights can be generalized and used for predicting unseen samples in the future. However, in business
environments, circumstances are always dynamic, which implies the ever changing nature of the data over time. In those
dynamic systems, fixed parameters can become invalid in the likely case of structural changes of the environment and
thus of the data, resulting in a worse “fit” of the model and ultimately a loss of “predictive performance”.
Consequently, to successfully implement AI in your business, you need technical experts who know the limitations and
the implicit assumptions of the underlying statistical models. Furthermore, they also need a sufficient amount of
domain expertise, to be able to assess the magnitude of the discrepancy between the model and the messy real world of
When it comes to applying AI, most people focus on the difficulties of the implementation side. But besides these
technical challenges, there are also managerial challenges regarding the application of AI in a business context. One
of the most discussed issues with AI models is the lack of interpretability. Especially in the context of Deep
Learning, those neural networks can be too complex to understand why this model comes up with its decisions. Even
though this lack of interpretability is clearly a disadvantage in theory, the question arises, if this plays a role in
practice. Many practitioners would argue, it doesn’t matter how the system derives its decisions, as long as the
black-box produces statistically validated and accurate predictions. But the reality is, there are three significant
risks, that result from imprudent application of AI:
Models can incorporate hidden and unwanted biases, e.g. consideration of
gender or ethnicity background for predicting suitability within a recruiting process. This is why extensive data
analysis is critical in order to gain the necessary understanding of the data and its influence on the model.
Models can make mistakes, due to the fact that machine learning is based on
statistical probabilities, rather than actual truths (correlation vs. causation). This is especially true if the
training data is flawed or if new situations occur that were not represented beforehand. This lack of verifiability
of the system is a major risk in critical situations, like life-or-death decisions in autonomous driving. In
practice, good evaluation protocols are essential in these situations.
Model biases and errors are hard to correct even if they are found. Most
machine learning systems are very sensitive to adjustments, since they follow the “CACE principle”: Changing
Anything Changes Everything. Therefore, a carefully collected training dataset is extremely important, not just for
the internal validity of the predictive model, but also for the external validity within the respective business use
case. Furthermore, constant performance monitoring and regular model re-trainings are inevitable for the long-time
success of those predictive systems.
These risks can be addressed in two ways: during the technical implementation, you need technically-trained staff who
know about the importance of good data selection, pre-processing, modelling, and monitoring mechanisms. From the
interpretation side, you need decision makers, who understand the basic data science principles, and can differentiate
between correlation and causation.
Even if you have specialized staff on the technical and managerial side, applying AI also brings a multitude of
additional challenges to the organization as a whole. In fact, the barriers to the adoption of AI depend on to what
degree the company has already adopted to AI. While for highly adopted companies the lack of AI talents clearly is the
bottleneck, the novices with no AI adoption are struggling with identifying a business case for AI in the first place.
However, identifying use-cases for AI in your own business require not only deep domain knowledge, but also innovative
minds with a strong customer orientation. Those companies in the middle adoption level are rather challenged by
deciding between competing investment priorities. In those cases, sound business cases can help to clarify between
competing investment opportunities. In general, there are three organisational challenges in introducing AI:
Businesses need to develop an intuitive understanding of AI. In order to do
so, companies can build AI-related skills inhouse through training the existing staff or hire AI talents. This might
be more expensive but pays off in the long run. In the short run, these activities can be combined with external
services by outsourcing to gain momentum and quickly create value.
Businesses need to re-organize the company structures, processes, and culture for
AI. The successful implementation of AI in products and services requires interdisciplinary cooperation and
clearly defined ownerships. Even process re-designs might be necessary. Finally, to be able to leverage the value of
data analytics, the gained insights need to be seriously considered during decision-making processes. This might be
a cultural challenge to become more data-driven and should not be underestimate.
Businesses need to re-think the competitive landscape, in which their
businesses operate. That means they must recognize new drivers for competitive advantage (e.g. exclusive access to
valuable data) and develop a strategic positioning regarding them. But although the majority of companies see an
urgent need for the development of an AI-strategy, only half of those already have one.
In summary, business applications of AI naturally contain some inherent technical, managerial and organizational
challenges. Therefore, you need trained staff with technical experience and domain expertise, who are aware of the
technical challenges and know how to address these issues, as well as technologically experienced leaders and decision
makers. Otherwise, the company is blind-folded, without knowing when AI should, could and why it would generate value
and ultimately competitive advantages.
Brynjolfsson, E., & McAfee, A. (2017): The Business of Artificial Intelligence. Harvard Business Review.
Davenport, T. H., & Ronanki, R. (2018): Artificial Intelligence for the Real World: Don't start with moon
Harvard Business Review.
Gerbert, P., Justus, J., & Hecker, M. (2017): Competing in the Age of Artificial Intelligence. The Boston
Ransbotham, S., Kiron, D., Gerbert, P., & Reeves, M. (2017): Reshaping Business With Artificial
Closing the Gap Between Ambition and Action. MIT Sloan Management Review.
Ransbotham, S., & Kiron, D. (2017): Analytics as a Source of Business Innovation. MIT Sloan Management
Rajpurohit, A. (2013): Big Data for Business Managers - Bridging the gap between Potential and Value. In X.
(Ed.), IEEE International Conference on Big Data, 2013. 6-9 October 2013, Santa Clara, CA, USA (pp. 29–31).
Piscataway, NJ: IEEE.
Sculley, D., Holt, G., Golovin, D., Davydov, E., Phillips, T., Ebner, D., Chaudhary, V., Young, M., Crespo,
Dennison, D. (2015): Hidden Technical Debt in Machine Learning Systems. In NIPS (Ed.): Vol. 28. Advances in
Information Processing Systems, Proceedings of the 28th International Conference on Neural Information
Systems (NIPS 2015). 7-12 December 2015, Montréal, Canada.
About the authors