Gartners predicted that within 2021, the number of artificial intelligence (AI) projects in place will be doubled, and 40% of leading organizations would have developed AI projects by the end of 2020. Though with a surge in the number of AI projects, many of them have also predicted to be a failure. According to Gartners again, 85% of the data science project will fail.

In this article, I would like to reflect on a few lessons based on our consultation experience from SMEs to large enterprises — to significantly improve their chances of success. To set the stage, this is a general consultation approach we took in the past.

Let us use a use case on a jewelry image classification engagement we have done in the past for a company where we started everything from scratch. The company is tremendously satisfied with our consultation and is always ready to serve as our reference. Following are some of the key lessons learned:

Lesson 1: Find a right AI consultant

There is no existing work on jewelry image classification, and the most we can get is a few research papers. But having the right expertise in image classification will help in navigating this problem, and significantly improve the chance for the project to be successful. An AI consultant would be someone who has relevant experience such as usage of deep learning framework and has good knowledge of related AI domain, in our case computer vision. Even better, he or she has read a significant amount of research papers related to the area, as that would help to develop intuition, as well as awareness of a specific toolset or methodology to use to solve challenges when they arise. Research papers are extremely helpful, as most of the challenges that we faced are most likely faced and addressed by other researchers previously. Standing on the shoulder of the giant enables us to leverage their experience and insight to resolve our problem efficiently and effectively.

Lesson 2: Communicate with clients closely and frequently

There are several benefits of communicating with the client closely and frequently:

1. Pivoting earliest the possible if we are not doing the right thing

Sometimes it easily falls into the trap that we fully understand the customer requirements. We should gauge the satisfaction of customers by having frequent meetings with them. For example, in our six weeks of engagement, we divided it into three sprints, which are for different breakdown objectives of the project. We also have a mid-sprint review at the middle of each sprint, which helps us to make sure what’s in the sprint review is the focus on the client- if it is not we will start pivoting to make sure we got it right.

Aside from having review sessions, we also setup slack channels and box folders to facilitate our communication with the clients so that we get timely feedback from the client when we have a problem, and at the same time, the client gets timely guidance from us whenever their journey to AI hit a roadblock.

2. Educate clients and handle misconceptions earlier

Sometimes a customer might have hidden misconceptions or unrealistic expectations towards a project. By communicating with them and seeking their feedback frequently, we can help to address their misconceptions timely and to manage their expectations to a reasonable level — often this involves educating the clients, esp. on technical details, to enable them to have a realistic expectation and make an informed decision.

3. Exceeding the expectations

What differentiates a good engagement and an excellent engagement is that an excellent engagement exceeds the clients’ expectations. In our project, through our conversation, we discover that the clients do not have the right technical expertise on the project and feel insecure. We, therefore, proactively place an additional emphasis on the documentation at an earlier phase, so to make sure we get a happy customer at the end, which is indeed the case.

Lesson 3: Follow up

Upon successful completion of an engagement, that should not be the end of the story — it should be the start of other amazing stories. We should try to follow up with our clients to ensure our solution serves them the best when time moves. This is to make sure our approach will bring long-term benefit to our clients, and we want an always happy customer.

All in all, I have covered the lessons learned that I think it’s generalizable to other AI projects.

In Visual Tensor, we have streamlined and simplified it to a 3-step process by building a framework that allows our Visual AI service to be readily consumable:

1. Contact us – Contact us and we will get in touch with you in 1 business day.

2. Our AI expert – Our AI expert will get in touch to propose the right offering based on your business.

3. Visual AI journey – We will hold your hand on your Visual AI journey, and make sure our Visual AI technology helps you in creating significant business value.