Enterprise AI: lessons learned

As a Business Development Manager I bridge the gap between our clients and engineers. This role allows me to meet companies across a wide range of sectors, from genetic engineering and telecom providers to more traditional industries. In order to apply our AI expertise we need to quickly gain a deep understanding of each client. How does their business work? What challenges are they facing? Where are the opportunities for us to help?

Once we have identified one or more business cases we help a client transition from AI awareness to AI operational. This often requires changes, such as new IT infrastructure (GPU’s), a different way of judging ROI and new development skills. In this blog I will discuss two insights we often come across.

AI is no holy grail

Clients often request for us to have AI ‘solve’ their most complex problems. These are challenges that their own experts couldn’t solve after years of research. The expectation is that AI has superhuman capacities making everything possible however in reality, tasks that are hard for humans to solve are also hard to solve with AI technology. Although we accept the challenge, it is impossible to guarantee satisfying results.

Current AI actually excels in automating educated tasks that are very repetitive. Tasks that humans solve thousands of times a month often follow a pattern that can be learnt by the machine. We train an AI model to detect this pattern by showing it lots of example data. The learned pattern can be used to classify, predict, cluster or detect anomalies. This allows AI to improve or speed up existing core processes and ultimately earn back the financial investment.

An example of this is judging a mortgage application. This requires an educated employee, but task is in practice often very similar. The same applies to an expert in a steel factory checking the product for production flaws, or a call center employee processing a customer email about an address change. It is these kinds of tasks that are very suitable for automation using AI.

Applying AI on challenges which humans currently cannot solve, has potential, however there is also a much higher risk of failure. When searching for a business case we look for areas where AI can have an immediate impact. This helps management to gain confidence and experience in this technology. Once AI is saving you money, you can always start to explore more challenging areas of application.

AI is the means, not an end goal

A lot of enterprises approach us because they want to ‘add AI to their business’. Working with AI seems more important than having it lead to real business results. The current hype around AI is enough ground to start just like blockchain, VR and IoT.

Although it’s important to investigate the potential impact of new technologies, investing time and money in a Proof of Concept is rarely budget well spent. AI projects are intrinsically complex in nature, requiring substantial effort both in terms of data collection and preparation and algorithm fine-tuning. In the limited time of a PoC lots of shortcuts must be taken, often resulting in conclusions which are likely known before the start of the PoC, such as more data would be required to achieve higher performance.

Identifying the use case and business goal clearly and defining the metrics to measure success are fundamental steps, steps which are often overlooked in the initial enthusiasm of getting started with AI. Are you looking to increase efficiency of an existing process or aiming to develop new propositions? Once the goal and metrics are defined, the focus should be on identifying the data sources, specify how to collect data in the most efficient and structured way, and recognize which changes should be carried on through the organization to enable this. Only then the iterative process of selecting the appropriate AI algorithms, optimizing their performance and deploying the solution, would start.


Concrete applications of AI are already saving companies millions and generating extra revenue for early adopters. Rather than allocating valuable resources on PoCs you should apply AI where it has a chance to directly impact the bottom line. Is your organization not ready for this yet? Get in touch with BrainCreators and we can help get you there.

Leave a Reply

Your email address will not be published. Required fields are marked *