Software and Service companies alike have been collecting data for years and sometimes decades. Innovators in these companies are looking to spin up new features or services to differentiate and enhance their offerings. Valuable insights leading to these innovations could exist in the mountain of data, but the question always is: How do we start looking for knowledge in that data mountain?
Time to put your hard hat on and go mining for business insights in that mountain of data. Here’s how we help SaaS companies implement AI into their products:
- Start with the end: The best place to start is to document high value insights that might be able to be produced from your current data set. Target things your customers would want to know or things you would like to know about your customers. Examples:
- Manufacturing: Predict if a part on a machine will fail soon. #predictivemaintenance
- Banking: Predict if a customer is going to switch banks. #customerretention
- Supply Chain: Predict production and consumption quantities. #demandforecasting
- Catalogue Data: Now that we have the end goal of the effort, we work backwards to identify the expected data we will need to predict the insight chosen. We catalog our target data sources and data types for ingestion into our Data Lake or other data store. Many times, a valuable business insight occurs at this stage. It’s possible that the expected data is incomplete or we discover another data source that improves the outcome of the effort.
- Minimal Viable Product is a Must: It is unlikely for any company to go from data wrangling to prediction in one quick step. Begin by targeting a minimum viable product. Descriptive analytics is a good MVP to start. Something as simple as basic clustering of data to discover linked attributes could produce business value. It’s not prediction, but it could establish an early win for the business to be able to categorize customers, machines, or geographical regions.
- Iterate, Iterate, Iterate: It’s important to go into these efforts knowing that you aren’t going to get a perfect model on the first try. Expect iteration to improve models, outcomes, and findings. We suggest the CRISP-DM methodology. It has been around for decades and it’s difficult to beat when it comes to data science process and we suggest it is followed.
- Deploy: Deployment means different things to different people. Having a model that produces insightful information and deploying that to the business can mean implementing it into reports, creating an API accessible to software, or be as simple as sending spreadsheets around. Once your iterative CRISP-DM method produces a finding that deserves deployment into the business, consider what deployment means to the solution. Can you implement this feature into your software? Can you charge for it? Or should this be implemented into your data enterprise data visualization strategy? How you deploy your findings, and to whom, matters for the overall success of your data science practice. Only then will you realize the benefits of your efforts.
Contact us: If you are a software or services company looking to harness the power of your data, contact us for more information about how we can help you find valuable business insights in your data mountain.