Top Reasons why Artificial Intelligence projects Fail

This article is consolidated from my LinkedIn post and contains information contributed from my LinkedIn colleagues in addition to my own view implementing AI projects

You can view this article as more of wisdom of crowd, similar to how random forest algorithm works :)

Link to LinkedIn post is below and a big thanks to all who have contributed and shared their experience to this:

Multiple factors come into play when implementing AI project. Data, skill, domain understanding, Culture, AI Strategy, Data Strategy, Capabilities are few that plays key role in success of AI projects

Below are some reasons why Artificial Intelligence projects fail to deliver

• Lack of Data Strategy - Starting with AI projects, layout and implement necessary data strategy. Without having atleast baseline data strategy, one might find quick success in AI projects but might hardly get translated into production capability

You can read my article below highlighting core capabilities of Data Strategy

• Focusing on technology and not on high impact use cases. Identify projects that can create highest impact on your KPI (Increasing topline, reducing operations cost, Increasing customer experience etc.) rather picking project where one see's scope of technical break through
• Data Science by itself does not always directly translate to production. Not having your Data and Software Engineers at the start of business understanding and modelling exercise cycle might result in models that either are hard to productionize or take way longer to do so

Results in training deployment skew

• AI initiatives must be thought ground up. Data, Infrastructure, Tools and accelerators that allows data scientist to focus on problem rather spend more time searching for data or working on plumbing activities. Without right Infrastructure in place Hiring Data Scientist is like

Hiring astronauts to drive a bullock cart

Checkout my article below that highlights need for ground up thought

• Lack of design thinking mindset, when dealing with AI problems, One need to get into hypothesis and experimentation mindset. Validate ideas quicker, eliminate or accept based on experimentation outcome
• Leaders who don’t fully understand AI are likely to have unrealistic expectations and quickly get frustrated with the trial-and-error process, detractors can be quick to call efforts failures and urge abandonment of AI projects. Check Reference section for more details
• Lack of Capabilities to accelerate data pipeline build - Build capabilities that can automate data collection, data quality and other data management activities. Ensure data assets are easily searchable and accessible along with right control in place. AI needs right and clean data to deliver better outcomes
• Substituting target labels with human or business intuition to avoid complex pipelines - There might be cases where labels for dataset is outcome of complex pipelines. There is no shortcut to this process. Ex: If you are trying to predict if a customer might complain in future, you might have customer historical complain history spread across call center, online channels, email, social media etc. It will be tedious to integrate from all channels and also create single view of customer
• Finding passionate and skilled team - Lack of confidence/trust in solution by competent team members can derail AI Journey. There is huge difference between one who talks AI to the one who does AI
• Starting with Data, Algorithm or tool rather understanding of business problem - Framing the business problem is key determinant to understand if business objective is met and it is the first step in data science process. Focus on business problem first and then think of data, tools and algorithms
• Directly Jump on to buzzword or trending algorithms at first - Design for simpler models and incrementally transition to more complex model as necessary. Balance between model performance, speed, explain-ability and ease of deployment
• Lack of right data and skill availability - I feel in today's enterprise this is slightly less of an challenge. Having the right skill who understands end to end AI cycle is key to take AI projects to implementation. Have right mix of Data Analyst, Data Engineers, Data Scientist and Software Engineers working in collaboration for successful outcome

Note: I might have modified some of LinkedIn comment to fit in the topic format and in some case skipped it if found already addressed in some form in other categories. Again, Thank you to all for the contribution.

Reference: