Modeling Your ICP

While data profiling describes who your clients are today, modeling learns from complex combinations of attributes in your client profiles to predict new accounts to prospect against. Defining your prospects by ICP filters and by model are two different but valid ways to uncover new accounts. Modeling may show you accounts that you wouldn't have discovered on your own using strict ICP-based filters.

The SalesIntel ICP model is a logistic regression model that considers thousands of attributes before identifying the 60 that are most important to your input data set as the features of your model. In this way, a bespoke model is created for you that best matches your business. Once the model is created, it is applied to the entire SalesIntel data set, returning you the best 10,000 matches. 

• Each attribute/feature selected has a different weight or importance
• Companies at the top of your fit list have more of these features, or have several of the most heavily-weighted features

To run a model, start by profiling your input companies, and then click "Launch Model" in the upper right. Your model will take a few minutes to run. In order to run a model, you will be prompted to save your profile first.

Viewing Model Results

Once your model completes, you will see search results with up to 10,000 companies, plus a score that indicates how well they match the input set. You can also filter this score using the slider at the top of your results. 

Scores can be exported to your CRM if you map the "ICP Fit Score" field in your CRM exports. The score will be shown on the company/account record in your CRM. Scores are also available in CSV exports.  

If certain attributes are critical to your business (you will only sell in certain regions, you will only sell to companies with certain revenue, you are headed to an event, you want to add a buyer persona, you are running an offer/campaign, etc), you can apply SalesIntel's standard search filters on top of the companies that the model returns, to get a more focused result. 

Save your model results as you would save any search, so that you can retrieve the results again. Your model is represented by a filter on your search so that you can always retrieve it with your saved search or include it with searches tied to GTMCanvas workflows. Selecting this filter will return your top fit companies. While the model will not be re-run, this outcome will be persisted for you. Models should be re-run quarterly to account for new clients in your input set and updated company attributes, or more frequently if your client base is growing quickly.

Once your model is saved, you can also set up workflows that use this saved search, such as:

• ICP Agents that will automatically push new buying center contacts for your best fit accounts into your CRM as they are added to the database
• VisitorIntel Agents that will tell you when someone from your best fit accounts visits your website
• Intent Agents that will tell you when your best fit accounts are exhibiting in market purchasing signals

Learning More About Your Model

Each model created is unique to the companies used to generate it. To understand what drives the power or energy of the model, pull up your model visualization using the model network icon on the ICP page or next to the score column in your search. This visualization will also load automatically the first time you run your model.

The "power" of your model comes from feature weights by category. "Features" are the attributes that have the predictive power that determines if companies are a good fit based on your input companies.

• Features are selected based on over- or under-representation in your seed companies as compared to the general population
• Features are given weights based on their impact on the outcome - think of it like importance
• Features can be qualifying (positive) or disqualifying (negative)
• By comparing the weights for the features across categories, each category’s impact is determined

In the initial waffle chart view, you see a representation of the power of each category of data. The categories always add up to 100%, with each square representing about 1% of the total energy of the model. On the right, you can see this numerically.

By drilling into one of the categories on the right, you can see further insights about how many features are in the category (in total, there are around 60-70 features in your final model), and whether these features are largely qualifying (help include companies by boosting scores) or disqualifying (help exclude bad fits by decreasing scores).

Saving Your Profile or Model

Once you run a profile, you can save that profile to retrieve again. If you are modeling your profile, you will be required to save when launching the model. Saved profiles can be retrieved from the main ICPIntel page.

On the main page, you will see several options:

• View your profile: this will load a snapshot of your profiled companies
• Rerun the profile: this will pull the latest data and rerun the profile. This action is recommended after several months, as data changes over time, but you can rerun as often as you wish. The new profile will replace the last one.
• View your model: this will load the top 10,000 best fit companies for your current model
• Rerun the model: this will pull the latest data and rebuild your model with fresh feature selection. This action is recommended after several months, as data changes over time, but you can rerun as often as you wish. If there was no existing model, this step will run the model for the first time.
• Visualize your model: this option pulls in model insights to help you understand what's driving the custom model created for your input

The recommender (lightbulb) icon will help you understand if there are further steps you should take to keep your profiles and models fresh.