Lityx is here to help. If you have any questions about this article or doing a POC let’s talk!
Conducting a productive Proof of Concept (POC) takes planning. The Lityx team has conducted numerous POC trials to validate the use of machine learning to enhance business intelligence tools and go the next step toward addressing core business problems. This is the second installment in our POC series where we share best practices based on our experience (see part 1 here).
To help stay on the path toward a successful POC our team has compiled the following checklist.
- Do you have the right problem to test?
- Is the data readily available and adequate?
- Are there clear metrics for success?
Choose the right problem
To be considered a success, your POC must be well-defined and highly relevant to your organization. Begin by compiling a short list of areas you would like to consider using ML and the approximate associated value. If any inspiration is needed, look at your competitors to see where they are investing.
The right problem is most likely one that is a priority for your organization and can be measured monetarily. For example, reducing churn has a measurable impact on your company’s bottom line and will gain the internal attention needed as you move through the POC process.
Here is a checklist to help you narrow the problem list and make a solid POC trial selection:
- Is the problem clearly defined?
- Will it impact your organization in a measurable way?
- What are the monetary benefits of success?
- Are the key stakeholders on board with a POC for this problem?
- Is it feasible to test within a reasonable time frame?
Inventory your data
Machine learning requires data of course, so be sure to examine what is available. As you narrow your shortlist of problems, understand the data requirements for a successful POC and if data quality or availability is unclear, remove the potential problem from the list.
Leverage team members in your group and in other parts of your organization for an understanding of the full extent of data availability, accessibility, and condition. Remember, the POC trial is a collaborative effort and getting involvement from the outset will make for a much smoother test.
For more information on data, refer to You’ve got data, now what? Three key data steps to maximize its value in AI
Define clear success metrics
Successful POC trials should generally be small in scale, conducted quickly, and the impact should be easily measured. When determining your success metrics focus on those that are largely understood by team members across your organization. If stakeholders do not understand what success looks like, the POC is likely to fail. Consider these examples when compiling your success metrics:
- For a manufacturer, a decrease in processing times or an increase in production
- For a retailer, an increase in customer return visits impacting LTV (lifetime value)
- For an online business, a decrease in cart abandonment or an increase in spend
- For a nonprofit, increase in giving and a decrease in cost per gift
Do you have an idea for a POC? Not certain how to move it forward or even where to start? We’ve helped numerous organizations across multiple industries determine POC parameters and conduct successful trials and welcome the opportunity to discuss your needs. Contact us here.
You may also visit our POC special offer page as an initial next step.
Read Part 3: Post POC Best Practices