How do you find a project best suited for your company's first, second or even 10th AI Project?
There are many ways AI can help companies. Automation, Optimization, Innovation and many more. There are tons of companies that use AI effectively today. Yet, it is still a fact that most early AI initiatives fail to take-off. Why? How do you minimise the chance of your next AI project becomes one of those? What can you do early on in order to increase the success chance of the AI initiative? Let’s take a look at the recipe.
- Firstly, is there a relevant business problem?
- This needs to be as defined by business unit/team that is facing the problem
- Solving these problems doesn’t have to be the Top objective of the last quarter but needs to be in the Top 3 for the near future (2 – 3 planning cycles)
- An attempt has been made through current operating structure to solve this problem. i.e., there are learnings valuable to drive the design of a data & AI-led approach to the problem. Remember that most AI techniques are learning systems. Without the learnings & directions on how to solve a problem, they’ll be a long and bumpy ride.
2. Is this problem worth the hassle? and the iterations?
- What are the business metrics related to the problem and which of those are most critical for you? In general:
o $ metrics are always the best to handle. They impact cost or revenue directly.
o One-step removed, like Market Share & Customer Experiences, are still very relevant
o Avoid those that are too strategic to be feasible. - A project champion once mentioned passionately “Data is the new oil and…”. I had to revert, “yes, agree. But do you want to sell petrol? or make French fries & sell?”. I hope you get the drift towards measurable financial metrics.
- What is the current state of the specific baseline metrics related to the problem?
- If this problem can be addressed through AI, what is the maximum impact we can make?
- What are the business metrics related to the problem and which of those are most critical for you? In general:
3. Is the customer journey, current and expected, known?
- The word “customer” is a personification of any user, entity, or process being targeted. Not just the real customers who make/break any business.
- What is the customer group being pursued?
- What does a good/bad customer look like, which customers are better left alone (sleeping dogs)?
- What is the intended user experience in the context of the problem?
- What are the key customer characteristics that we plan/like to focus on?
4. How do you intend to influence the change in customer behaviour?
- Interventions here vary a lot, from rewards & incentives to penalties, commercial clauses, regulatory needs and more. For eg., if you’re running a marketing campaign to retain users, an effective intervention could be to give freebies or offers.
- What interventions are available and what are out of bounds?
- Is there a willingness and ability to change the incentives? If not, how do we influence the change in behaviour?
- How will customers be targeted and reached?
5. What are the known operational bottlenecks and a plan in place for addressing those?
- If the operational bottlenecks are too big and cannot be addressed, better to agree a testing model for the project (like a Prototype) and a potential pausing-point.
- If there are no bottlenecks, very likely, it is too good to be true. OR we need to expect to run into a few of those as we go along the journey
- Ideally, you should have an understanding of which challenges are contributing the most to the business results (or what is causing the problem?
6. Do we have enough data, in any shape/format, to solve the business problem?
- We don’t need ALL data, but just enough to start. If there is no data / too limited data, we have a dead-start problem on hand. Effective data collection can be a major distraction for the Champion Team that is better avoided in early stages.
- Have all relevant data sources been identified?
- What data sources available and usable? Is this sufficient?
- What are known technical and data limitations?
You may be wondering if these 6 questions are ‘too much’ of an ask to start a new initiative. You are right, they are. The effort it takes to get these right is usually longer than the effort to execute the first AI model. Driving value from AI is an iterative and continuous process. Hence, at this stage of its evolution, it is best to set it up for success on as many dimensions as possible.