A Successful AI Innovation Sprint Requires the Right Roadmap

July 15, 2020

So you’ve decided that it’s time for your organization to step into AI innovation, after learning that other players in your industry are taking steps into AI themselves. It’s time to evolve or get left behind. However, you don’t know where to start or even if AI can help you at all. Well, have no fear because you are not alone, and today we will go over how to get started.

One good way to make the plunge is to first conduct an AI innovation sprint. Our COO and co-founder Carlos Melendez recently wrote an article about AI sprints. These are short term projects that allow companies to experiment with AI. An AI innovation sprint is used to develop a proof-of-concept and determine if AI is right for your organization. To start an AI innovation sprint you will need the following three things:

A problem

In general, problems that involve making sense or learning from lots of data are ideal for AI. For example, you might want to predict your customer’s behavior or simulate the decisions taken by a domain expert. Since there are many applications for AI, it’s okay if you are not sure if your problem is ideal because that is exactly what you will find out with the AI innovation sprint.

A data science team

For an AI innovation sprint, you want to keep your team small. Once you have proved the value of AI for your organization, then you can invest in growing your team as needed. These are the people you will need to get started:

  • Data Analyst – A data analyst is responsible for data collection and interpretation. The data analyst also converts the results of his analysis into tangible insights.
  • Business Analyst – A business analyst is responsible for converting business expectations into data analysis tasks. If your data analyst or data scientist lacks expertise in the problem domain, the business analyst closes the gap.
  • Data Scientist – A data scientist is responsible for data preparation, model training, and evaluation. He has coding, database, machine learning frameworks and models knowledge.

These professionals are in high demand right now and are hard to find. That is the reason it is common for organizations to find an AI outsourcing firm to help them get started. An AI outsourcing firm will have the right team to run the AI sprint and provide enough insights and value to justify a higher investment in the future.

Data

You probably think that you don’t have any data to get started with AI. Even if you have been collecting some data, you probably think it will not be enough. Well, that is exactly what you will find out with the AI sprint. Chances are that you already have more data than you think. Talk to your IT department and find out what information is being stored and make sure it is available to your data science team. It is okay if it turns out your dataset is not enough because the AI sprint will tell you what you need.

Now that you have a problem of interest, a data science team, and data to solve the problem it is time to start your AI innovation sprint. These are the five steps that you should take:

1. Identify the business challenge

Define the problem in clear and unambiguous terms. At this point, you should also define your success criteria and the approach to get it. One question to ask yourself is, how you can use data to solve your problem. The answer to this question can give you a path forward and define the right approach. Your approach might include statistical analysis, predictive models, classification models or other forms of machine learning models.

2. Conduct a data audit

Identify your dataset, where it is coming from, and who is responsible for collecting it. The dataset could come from various sources, so don’t be afraid to reach out to other departments within your organization for data. Once the dataset has been identified and collected, the data analyst gets to work.

The job of the data analyst is to validate that the data is representative of the problem to be solved. He uses descriptive statistics on the data like mean, median, mode, minimum, maximum, and standard deviation. He also visualizes the data and examines the correlation between variables. The data analyst also identifies and applies additional transformations to get the data into a state where it is easier to consume. He will consider things like missing data, invalid values, duplicated data, formatting, and feature engineering.

3. Build the algorithm

The job of the data scientist is to select and build the best model to solve the problem. He will build a descriptive or predictive model depending on the problem and the approach identified earlier and then train the model with the provided dataset.

4. Assess the results

Assess the quality of the model. There are two main attributes to evaluate a model. One is diagnostic measures that describe the actual performance of the model in a test dataset. The other is statistical significance that describes the confidence level of the model prediction. At this point, you will know if you need to adjust the dataset or the model to improve your results. If the results are not satisfactory, then the data science team will provide recommendations on necessary improvements.

5. Operationalize the algorithm

Once you get satisfactory performance out of the model, it is time to make it available to the organization. Define how the model will be consumed and how it will be maintained.

 

That’s it. You now have a roadmap of what is needed to run an AI innovation sprint and how to do it. In my experience working as a data scientist for Wovenware over the past two years, I have learned that a single AI sprint will usually take up to four weeks. However, a single AI sprint is generally not enough to get a good solution to operationalize. The first AI sprint will answer your question, is AI valuable to me? If the answer is no, then you might get enough insights to reframe your problem or collect better data. It usually takes a couple of additional sprints to refine both the dataset and the AI model to get good results that provide enough value to operationalize it. Above all, remember that this is an iterative process that takes not only knowledge but also creativity and passion to get right.

 

Leave a Reply

  • (will not be published)