Mediafly
MediaflyKnowledge BaseAdvanced InsightsMachine Learning
FAQs Knowledge Base
Mediafly Admin Settings Admin Guide: Workspace Management Admin Guide: Group and User Management Admin Guide: Content Management Engagement Content Getting Started with Mediafly Working with Content and Workspaces Sharing, Personalization, Preferences and Integrations Digital Sales Rooms Content Troubleshooting Apps and Requirements Value Getting Started with Value Using the Design Studio Items Library Output Reports Marketing Tools Value Integrations Additional Value Resources Advanced Insights Getting Started with Advanced Insights Setup and Configuration Advanced Insights Integrations User Management / Access and Visibility Value Data Analytics Salesforce Specific Hierarchy Report Management Creating and Editing Reports Reporting Types (Visualizations) Dashboards and Cards Creating and Editing Dashboards Out of the Box Dashboards Creating and Editing Cards Goals Machine Learning Available Machine Learning Models Administration: Machine Learning Activity by Day Energy Score Content Properties and Types Inspection Coach Settings: Onboarding Settings: Basics Call Card Overview Settings: Smart Alerts Coach Integrations Revenue Intelligence Opportunities / Pipeline Classic Insights Admin Guide to Insights & Reporting Product Updates Resources Genius Drive
Related Articles
Properties
Tags imported-from-zendesk, Intelligence360

Do I have enough data for machine learning models to be predictive?

Intelligence provides robust and flexible machine learning models. However, machine learning requires a data set to be trained against-again, machine learning is measuring what characteristics of deals are predictive and, based on that history, what's likely coming.

Machine Learning can struggle in the following cases:

  • There are not enough records. (i.e. Number of Won Opportunities Last Year is too low)

  • Something happened that makes what happened in the past irrelevant to where we are now. (i.e. Covid)

Intelligence's approach to machine learning is highly robust to market and process changes, so we'll cover some case studies as well. 

So what does this mean, and do I have enough data for machine learning?

Unfortunately, it depends. This document lays out high-level guidelines on when you might (or might not) have enough data for machine learning to work for you.

Machine learning predictiveness is based on three things; volume of data, consistency of data, and completeness of data.

  • Volume - you just need enough records to profile against, more is always better

  • Consistency - does your data follow set profiles, or is it more or less random?

    • Note: Volume helps this, small data sets seem random but with higher volume patterns emerge

  • Completeness - does every record have data to measure against? Or do only a handful of opportunities have values for certain fields?

    • Note: Intelligence activity capture and context sourced properties help this, we have 200+ data points we can measure consistently and completely across customers.

Intelligence has created resources to help review this, one of them is the CRM Hygiene Dashboard - check out the guide here.

Evaluating Volume

Here are the key questions you should ask, and evaluate for machine learning fit. Review the "opportunity hygiene" row in your hygiene dashboard.

  • Is opportunity type my best split?

    • ML model splits should follow significant sales process differences

      • Ex. New Business vs. Renewal

      • Ex. Services vs. SaaS

      • Ex. Commercial vs. Enterprise 

  • Is activity capture enabled for all customer-facing resources?

    • Yes

      • Do I have at least 100 opportunities a quarter per opportunity type/split, and do I win at least 10% of them?

        • If you have more opportunities and a lower win rate, that's ok.

        • If you have fewer opportunities and a higher win rate, you will still have issues.

    • No

      • Do I have activity capture only account execs, or no one? If activity capture isn't enabled, you are in effect relying on manual data entry by reps for your predictive model, this will significantly increase the number of opportunities required for accurate predictions

        • Note: AE's only typically only provides 60% of customer-facing activity, significantly reducing the VOLUME of data that can be measured for model predictiveness. We highly recommend all customer-facing resources participate in activity capture

        • Do I have at least 300 opportunities a quarter per opportunity type/split, and do I win at least 10% of them?

          • If you have more opportunities and a lower win rate, that's ok.

          • If you have fewer opportunities and a higher win rate, you will still have issues.

Evaluating Consistency / Completeness

  • For a given field in the model, do the majority of records have a value?

  • Does the value change after closed-won? (remove these fields)

  • Do my reps follow consistent processes for basic opportunity management?

    • Do they consistently create opportunities when they start the sales process?

    • Do they consistently update stages or forecast categories?

    • Do they consistently set close dates and maintain them through the sales cycle?

If consistency is low, the best approach is increasing volume, which would increase our guidance for minimum opportunities from 100 to 300 per quarter or more.

Intelligence context properties, activity capture, account contact enrichment, and other features were built to create a complete data set for machine learning. Because of our approach, we are highly resilient to changes to processes or environments. "Platform" fields will always be complete across all your opportunities.

Learn more about what data we create for opportunities here.

Case Studies:

  • COVID Sales Impact

    • Intelligence measures all data related to activity, we identified that the overall sales effort increased, as well as deal cycles were slowing. Our models adapted to the new reality as it was incremental to the complete data we were already measuring, and we stayed predictive!

  • Sales Process Change

    • A client changed the sales process and stages. Intelligence was able to carry comparative metrics of activity levels across the sales process change and continue to remain predictive, even while traditional stage conversion rate metrics failed.


Is this article helpful?
0 0 0
Leave a Comment
 
Attach a file
Powered by: UseResponse - Knowledge Management Software