How to identify viable AI use case?
AI/ML technology provides enormous opportunities for organization to gain competitive advantages but how do you separate hype from reality. Identifying a viable AI use case and aligning it with business strategic goal is vital for organization to succeed in AI initiatives. With the rapid pace of AI technological advancement, it can seem to be a challenging task for the organization to get hold of what AI based application are viable for them to adapt. The problem is many of them don’t know where to start. Or worse yet, they want to start with an audacious project to show off what’s possible.
In this article, I will break down categories of problem that are commonly handled by today’s AI and will provide actionable steps to begin with AI use case identification with approach and perspective.
AI Capability
An AI capability map is a concise way to see what’s possible and what’s working with AI. At the start of AI use case identification process, It would be beneficial for domain experts and other stakeholders involved in the discussion for AI use case identification to understand what specific capabilities are these AI based technologies are enabling. This will provide high level context of what type of business problems AI can help you. This will allow business domain expert to have a fruitful discussion with AI experts during brainstorming session and feasibility analysis. The table below provides an overview of the type of decisions/actions that AI will excel at and inform strategic group with AI trends or capabilities most likely to deliver the business value.
Recently, we’ve seen a rapid evolution in NLP applications due to deep learning technology and integrating ML with NLP allows for far greater automation, scalability, and accuracy.
Every organization will have unique needs and different priorities or business-critical problems that AI can help you to solve. You can envisage that AI/ML generates value in three ways:
- Improve existing products/services to serve customer better
- Automate tasks/processes to serve customer more efficiently
- Create new products/services/market to increase customer base
Brainstorming Framework
Identifying viable AI use case is a team effort and requires cross disciplinary meeting with stakeholders. Organize at least a half day workshop with Business, product, AI and data engineering leaders to discover the potential area where AI can help business to create unique opportunities. Domain expert and AI experts play a key roles during AI use case identification phase as both of them have expertise in respective areas in business and Technology.
- Domain Expert deeply understands the business and context / systems surrounding business process and would be able to understand what business value AI will bring.
- AI expert is someone who has done AI and deeply understand about AI capabilities, technology advancements, Industry successful adoption of AI technology and is knowledgeable enough to provide guidance about the complexity and feasibility of the AI use-case.
Potential area to explore to identify AI use case
If you have a lot of data, and you are making lots of daily decisions with that data: These are just a few examples of how organizations have taken AI approaches to solve problems they’ve been working on for a long time and have lots of expertise. We already perform these tasks with hand-crafted rules either done by human or by machine, but machine automatically learn (better?) rules from example data. This type of use case would be good candidate to solve using new approaches via AI.
Good example:
- Fraud Detection in Banking, Insurance
- Advertisement in eCommerce
- Financial advisor in Finance
Identify tasks that are done by Human: If the action taken by human is based on data, highly repetitive, tedious and hard to solve by rule based systems, hence overall process is slow then it would be good candidate for AI.
Good example:
- Visual Inspection in Manufacturing
Identify what mission-critical business information you are dying to know: In the rapid changing market landscape, organization is looking to find hidden information from huge amount of data to take business decisions but can’t currently access it due to variability in data. Maybe it’s about understanding the factors yielding the highest customer lifetime value, or the user behavior most indicative of expected churn.
Identify how information is used, accessed and served: Nearly eighty percentages of enterprise data is unstructured and has been less accessible to wider group. In last few years we have made significant progress in Natural Language processing (NLP) and now able to solve wide range of problem which was difficult to solve few years ago. Enterprise can leverage NLP in multiple ways to improve both customer experience and document insight discovery.
Identify how enterprise applications are used by customers (Internal or external): Looks for ways to make application more predictive and contextualized based on user need. For example route customer support requests to the person or team who will treat them best.
Identify if and how much human decisions can be turned in data to augment human decision making with AI: Human consider different sort of data (internal, external) and context before taking decisions. Next we can ask question about how can those bits of information that humans use to make a decision be turned into data and can further be integrated with AI based product to help and improve human decision making. This in turn allows human to do what they are best at i.e. creative problem solving and computers excel at precision, rigor and consistency.
Feasibility Analysis
AI is very good at executing narrowly defined tasks and struggle in uncertain environments; hence AI experts and domain experts must do a due diligence before the starting AI project. In the earlier section, I have provided a quick reference about AI capability for consideration. There are many business problem which can’t be solved using today’s AI or even possible to solve using AI but are technically very complex to deliver business value.
Prioritise
AI Use case should be prioritized properly and must be driven by business strategy. I have mentioned some of the points below for consideration and you will realize that each one of them has a potential to derail or delay the AI project if not assessed appropriately.
Strategic alignment: AI can impact the way organization functions and operate and the key goal of AI is to support business strategy, hence it must be linked to strategic goal.
Impact and Success criteria: We need to be very specific about which metric can be used to track progress. Precisely aligned metrics will allow organization to monitor progress and in some case allow them to move resources from unsuccessful AI project to other AI project which lacks resources.
Data Availability: How easy is it to access right data for building AI application? If data is not available or very difficult to access then initially we can give low rating and plan to make data accessible for future.
Ethical or Regulatory issues: Ethics, data privacy, bias, model explainability is a concern for organization. So think properly about the legal implication on business, customers, and employees. Accessing this early will allow you to think through the additional work required to ensure that AI solution is complying with applicable laws and regulation.
Talent availability: Unavailability of AI talent might prevent you from achieving your AI goal. Consider different options to onboard AI resource either from Vendor partners or hire externally or may be trained some of passionate domain experts.
Infrastructure: Do you have infrastructure and environment to support AI works? If not then it would be good to start with cloud before building infrastructure in-house.
Technology: AI work requires different set of tools, packages and for some organization it’s not easy to adapt state of the art work directly from open source packages due to multiple factors.
Technical complexity: What’s the complexity score for AI solution? Will it require long training time? Will it a multi stage AI pipeline?
Integration challenges: AI based solution is not a one off activity and requires ongoing monitoring of solution effectiveness and improvement based on new data. Hence you need to be extra careful and plan to integrate feedback loop & monitoring in your E2E AI pipeline.
Implementation challenges: AI solution is rarely implemented in isolation and is usually integrated with product or process. How difficult it is to deploy in production? What will happen if something goes wrong during implementation? What’s the impact?
Conclusion
AI is a powerful technology, so use it thoughtfully and judiciously. If you are starting in your AI journey then pick the low hanging fruits first for the simple well defined problem and AI solution. Select top two-three AI use case to ensure that your AI strategy remains focused and achievable. This is very important as the initial success in the AI projects will create a momentum, build trust within organization and allow senior stakeholders to continuously invest and fund AI initiatives.
With these small wins, you’ll be able to quantify the things that you need to understand and manage when you start scaling your AI/ML initiatives.
Thank you for reading my post. In the coming posts, I will cover in detail about how business problem can be framed as machine learning problem. Stay tuned.
