Why Test ?
A good model backtests.
Testing helps you prove and improve your ICP model. This doc outlines what to test, how to test in Keyplay, and best practices for testing your ICP Model.
2-minute video walk-through:
What to Test
The key to backtesting is contrast. In Keyplay you can upload "test lists" to analyze your ICP model results on historical business data.
We recommend adding 3 test lists to start:
Your "best" customers. Make a list of your highest value customers (ideally based on CLV or NRR). This list should be accounts you expect to score high.
"Not a fit" accounts. Make a list of DQ'd accounts (excluding timing reasons) or Stage 0 Closed Lost opportunities. This list should be accounts you expect to score lower.
A control list. Use a random sample from your SAM that you'd expect to represent your overall market score distribution. You can generate this in Keyplay with 1 click.
Upload these lists in the Keyplay tester by following the steps below and you'll see report like this:
This is a "good" backtest because the higher ACV accounts are scoring higher in my model.
To gain deeper insights, analyze your customers by ACV, recency, or other key variables that impact your model’s performance.
Getting a "Good" Backtest
In general you want your best customers to score high (~60-70% should be A or B tier) and your DQ'd or bad accounts to score low (leaning toward C/D). Your Control (SAM Sample) will also act as a good anchor point to understand how your model applies across your entire serviceable market.
The exact ratios can vary, but the key is to see a decreasing score distribution as you go from "best" to "worst" accounts.
How to Set-up Test Lists in Keyplay
1.) Upload Your Test Lists
To upload a list, you need a CSV with a "Domain" column. The Domain column should include the primary website domain for each account (e.g., example.com). Once your lists are ready, click "Add a Test List" and follow the steps to import your CSV, name the list, and see the results.
You can upload up to 10 lists with up to 1,000 account each. The first time you load a list might take a couple minutes to process.
2.) Create a Control List from your SAM
You can now create a Control List from your serviceable market with just one click. This provides another point of comparison for testing.
If you change your SAM, you can refresh your Control list.
3.) Review Results & Adjust Your Model
Once you've setup your test lists, review the results and adjust your model accordingly. As you make changes to your model and "Save," your test lists will automatically refresh.
Switch between the model builder and the 'Test' tab to view updated results.
ICP Testing as a Program
Treat your ICP as a program, not a one-time project.
After your initial roll-out, set a cadence to review and re-test your ICP model. Each time you revisit your ICP, use the backtest to gain confidence and build trust with your team.
Need help? Contact your CSM for guidance on best practices or help with the setup process.
Happy testing!