prevision.io : Can an AI Project Succeed without Delivering any ROI?
 
 

Can an AI Project Succeed without Delivering any ROI?
 
by Florian Laroumagne
 

The expectations around the power of artificial intelligence (AI) is quite high these days across several industries, including banking, finance, insurance, retail, media, healthcare, defense, and more. Technologies that help solve AI projects are available on the market, even on the shelf, whether we consider going with open source or commercial solutions. Computing power at our disposal in our data centers or in the cloud have never been more powerful. Also, if you have data readily available  (which is probably true if you use CRM, ERP or any decisional tools within your company) you are ready to move into the AI journey.
 

However, this adventure isn’t at risk. As of today, nearly 80% of AI projects don’t move into production. This means that they are financed and most of the time done by companies (with or without the help of consultants) in data labs or in specialized sections of the IT or R&D department and then never get used by the business in the end. 
 

The reasons for this failure can be complex and varied: technological barriers, a lack of qualified human resources, and bad identification of valuable use cases. 
 

What is certain and what even applies to the more mature companies is that an AI project that isn’t capable of delivering a sufficient return on investment doesn’t have much chance to succeed and I’ll explain to you why. 
 

First of all we need to define what a return on investment (ROI) is. A project’s ROI can be calculated by making the difference or even dividing gains coming from the project and its costs. Costs are generally speaking easy to estimate (software license, consulting, IT infrastructure, internal HR costs, etc). However, gains are less easy to define. We can think at first about what the company is going to save from a financial point of view. In fact, this is a fair point but an incomplete answer.
 

Indeed, gains can include:

  • Financial. The capacity of an AI algorithm to outperform your actual decision process is relevant. It could help you better target customers, improve efficiency of marketing campaigns, increase production and quality of your product. In general, this will have an immediate impact on your company financials’ key performance indicators ().

  • Drastic increase of operational efficiency of your in-place teams related to the topic addressed. Indeed, in most cases what you’ll do is delegate “trivial decisions” to AI algorithms that are still time consuming for humans. By doing so, your teams will have more time to focus on the more complex tasks and decisions which is valuable for them and also for the company.

  • Increase customer satisfaction. For instance, the release of a new service or a new project will have, generally speaking, a positive impact on your customer base. This doesn’t translate to an improvement of financials’ KPIs per say, but can have significant impact in terms of customer happiness and then, decrease at term churn possibilities.

  • A marketing argument toward the market. This type of initiative might be interesting when you are seeking to recruit or when you want to showcase a level of excellence within a specified domain.  Please note that projects that have only this type of gain will mostly have a low ROI.
     

Given these considerations, it appears that some impacts are easy to measure which is especially true about financial ones. To do so, one could measure outperformance before / after the delivery of the AI project or even use A/B testing. This strategy, which is an excellent one, consists of splitting your process into two sub processes:

  • A minority part of your population (let’s say 20%) will still be addressed by the old process

  • The complementary population (so 80% here) will be linked to the AI based one
     

With this method, the estimation of the outperformance will be trustful, easy to do and will take into account all the external factors (such as COVID or any other major effect). 
 

On the contrary, some gains are way more difficult to measure, for instance customer satisfaction(these can still be done with surveys),but even worse: they are difficult to correctly anticipate at the AI project’s kickoff. 
 

All in all, it is a good practice to list at the beginning of the project the expected achievable gains in a realistic and honest way in order to avoid bad surprises. By then, you’ll need to put in place, for every potential gain, a key performance indicator (KPI) that will measure it. If this is not done, the project may be questionable and will most likely fail. By failing, I mean the incapacity to build a stable project over time that goes well beyond the handshake of companies directors that usually happens at a project delivery (even if this is more of a ceremonial handshake during the COVID era).) Indeed, which C-level executive would agree to pay for a project without knowing the expected results for its company, especially with the high cost related to AI. And while AI is not cheap, the barriers to entry are being lowered through the work of companies like Prevision.io.  With easier ways to bill clients and manage platform use, you have a better chance of a successful AI project when outsourcing the work.  
 

In summary, if you are about to launch a new project, especially in the AI field, please make sure that it is feasible and try to estimate at best and within all honesty the project’s probable return on investment by taking into account not only financial aspects, but also human and technological ones that shouldn’t be overlooked on these kinds of projects. If after this first study you can’t provide a sufficient return on investment for your company, think twice before launching a project that is probably going to fail, while there are probably other ones waiting to be solved. 
 


 

ABOUT PREVISION.IO

Prevision.io brings powerful AI management capabilities to data science users so more AI projects make it into production and stay in production. Our purpose-built AI Management platform was designed by data scientists for data scientists and citizen data scientists to scale their value, domain expertise, and impact. The platform manages the hidden complexities and burdensome tasks that get in the way of realizing the tremendous productivity and performance gains AI can deliver across your business.
 


Suivez-nous sur Twitter ! Suivez-nous sur Linkedin ! Suivez-nous sur Instagram !