Before Investing in an Artificial Intelligence Project, think of these Business Considerations.
Neither “doing things right” nor “doing the right things” are optional.
Introduction
For some years Artificial Intelligence (AI) has been moving away from a purely academic discipline. By now it has developed into an indispensable tool in business. In many cases, the path has been rocky. Most projects in this area are still unsuccessful in the sense that they fail to create a sustainable economic result for their company.
But that’s what projects in a corporate environment are all about: They need to create Value, i.e. an improvement in the company’s profitability. To reach this goal, further professionalization of the use of AI is often necessary.
With the arrival of A.I. in business life, the environment has changed dramatically for practitioners. I already mentioned the importance of economic considerations. For certain industries, regulatory/legal demands can be very important also. The emphasis in training data scientists firmly remains with technical and mathematical aspects, however. Data Science is a fairly new discipline in many companies. Many corporate data scientists have not yet had the time to become fully familiar with the non-technical aspects of their work environment.
For A.I. projects, not considering Economic, Marketing and/or Legal aspects can have serious consequences . Too many projects are announced, only to be discontinued after 12–18 months because of irrelevance, failure etc. As the required specialists are rare and expensive, a discontinued project typically involves significant sunk costs. Such an event can undermine the reputation of A.I. and those involved in the project with top management for years to come. Clearly, a thoughtful, strategic approach is needed.
Starting an A.I. project in a complex corporate environment requires a careful setup, covering cooperation between several departments/specialist disciplines, their organizations and processes. The terms “operationalization of AI” or “MLOps” are often used.
How to start
Ideally, you ensure the relevance of the projects from the perspective of top management right from the start. Relevance & profitability result from additional sales and/or reduced costs. In the present article, let’s focus on sales.
Sales can be increased by selling either more of the same, selling the same at a higher price point or selling new products / services:
- Selling more of the same means either winning over new customers or selling more to existing customers
- Selling at a higher price point implies either a market split or creating a more premium brand
- Selling additional products / services means either developing new products / services or entering new markets
We could continue to develop this topic tree for a while, but we will stop here instead. As you can see, a company has a number of options for increasing sales. However, which options are most promising depends:
- On the one hand, on the individual company and its market environment. This is a complex field of analysis in its own right and usually the daily work of its own team of specialists. This team needs to be involved in any discussion related to exploring the priorities of potential AI projects. Specialists for these questions can usually be found in the Strategy and / or Marketing departments
- On the other hand on the existing data, its accessibility, quality and governance. Responsibilities are not always clearly regulated here and can range from IT operations to Marketing, Business Intelligence as well as Business Division Heads
A systematic approach towards corporate AI initially requires an identification and ranking of realistic potential projects. In this article, I’ll call this phase “ideation”. This must be followed by a review of these project-ideas in the context of legal and commercial risks, a phase I’ll call “risk management”.
Ideation
To make sure that a first (pilot) A.I. project is both relevant and feasible, initial discussions need to include representatives of strategy, marketing and data governance as well as data scientists. Such discussions could for example take place as part of a (series of) ideation workshops. Such workshops require careful preparation by those involved as each team needs to contribute their specific expertise:
- The Strategy / Marketing teams present the company’s strategic focus topics and/or market positioning. Key question: “What is vital for our company in the current competitive landscape?”
- The data responsibles summarize which data is available and how. Key question: “What data is available for A.I. or can be made available with reasonable effort? “
- The data science team discusses examples of how A.I. supports company goals elsewhere. Key question: “What is the state of the art of A.I. use in the company’s industry/competitive environment? “
One of the pitfalls of such a multidisciplinary workshop is that those involved often speak different jargon/”specialist languages”. This can easily lead to the situation of “talking past each other”. It has been found in practice that the use of a knowledgeable external facilitator in the preparation and implementation of such workshops can help handle such challenges.
In the final phase of the workshop, the actual “ideation” is carried out in which ideas for relevant projects are developed in a brainstorming-style session. These are prioritized in a subsequent “ranking”.
Where necessary, it can make sense to organize several workshops, as the above discussions usually reveal that further information is required, etc. In addition, evaluations, e.g. investment cases, must usually be calculated.
Risk Management
Before project proposals can be submitted to top management, further aspects and risks must be evaluated:
- Protection of personal data. The data to be used often have a person/individual aspect (e.g. customers / suppliers). The use of such data has been subject to increasingly strict regulatory frameworks for several years. A violation of this regulation can lead to significant penalties and/or reputational damage and should therefore be avoided. Often the solution is to anonymize or pseudo-anonymize the data sets used. The topic is, however, more sensitive than it appears at first glance. Even the potential identifiability of the individuals behind the data, while they are not mentioned by name, can be critical. A case-by-case evaluation of this issue by specialists, usually the company’s legal department, is recommended. Public law firms that have specialized in this area may provide further insights.
- Bias and Public image. Companies regularly need to take decisions which substantially impact customers’ lives. While reality dictates that both the general and business public may not be pleased with every decision, companies should be careful when it comes to unwarranted bias and/or discrimination, as in recent years certain segments of society have become increasingly militant about this. Companies have suffered considerable reputational and/or financial damage after customers obtained the impression of discrimination based on skin color, gender, sexual orientation, age or other inappropriate factors. The awareness that such bias must be carefully guarded against is now widespread in Data Science circles and methods to prevent it need to be part of the toolkit of any well-trained Data Scientist. The question remains whether documentation of such care is always available once the company is faced with a (possibly frivolous) law-suit at a (much) later point in time. Against this background the question arises to what extent decisions made by the A.I. system for individual projects must be:
o Understandable
o Explainable
o Auditable.
These questions must be clarified before the design / implementation of the system take place, as they may restrict the design degrees of freedom (e.g. with regard to the selected algorithm) or make a new system design altogether necessary. Creating full audit-security is costly. Not creating it may prove even more expensive. It is important to get the extent right under the applicable legal framework and project goals. The contacts for clarifying these questions are the PR and Legal departments.
Once these aspects are covered, possibly in workshops that alternate with the ideation workshops, the project can be submitted for approval to top management.
Obtaining Management Approval and Commitment
The procedure for the approval of A.I. projects is company-dependent, so it cannot be debated in detail here. However, it can be very beneficial to submit an A.I. project to top management for approval.
As mentioned, it is important that the available, expensive resources are focused on the most relevant projects. It may be appropriate to make sure that the assumed priority of projects indeed coincides with the CEO’s view. Explicit top management approval also ensures the relevance of the project.
The creation of approval documentation, as a rule requires the preparation of project, milestone and financial plans. Mostly there will also be corresponding reporting obligations during the project and possibly the creation of a steering committee . This ensures a certain discipline but also cross-departmental communication during the preparation and implementation phases of the project.
The commissioning of the project by top management also gives the project internal visibility and legitimacy. This can help obtain the cooperation of all parties involved e.g. facilitate the availability of good quality data.
In addition, the use of A.I. may lead to questions/controversy both internally and externally. Involving management in the early phase of a project means that management is prepared for such questions.
Conclusions
We have seen that an introduction of AI projects by Data Science departments working in isolation carries grave risks. In the above I showed how the systematic identification of promising and relevant projects can be achieved through the involvement of corporate specialists in the areas of Strategy, Marketing and Data. Risk Management can be achieved through workshops with Legal and Public Relations’ specialists. Finally, I mentioned how inviting explicit approval by top management has wide-ranging advantages.
If you liked this article, please don’t forget to applaud here below, leave a comment and/or follow. A more playful approach to the topic can be found here. Feel free to drop me a line on LinkedIn if you’re struggling with this.
