PREDICTIVE AUDIENCE TARGETING GUIDE
Traditional prospect targeting often begins by describing the type of person a marketer believes is most likely to buy. Predictive targeting approaches the problem differently: it analyzes actual customers or responders to discover which characteristics are associated with conversion.
Those patterns can then be applied to a larger prospect universe to identify and rank people who more closely resemble the characteristics associated with successful outcomes.
This guide explains how predictive prospect targeting works, how it differs from demographic list selection, how predictive models are validated, and when marketers should consider using predictive audiences.
Published by Inbound Insight · Last Updated: September 2026
Predictive targeting estimates relative conversion potential. It does not guarantee that an individual prospect will convert.
Predictive prospect targeting uses statistical or machine-learning techniques to identify patterns associated with customers, responders or other desired outcomes and then applies those patterns to prospective customers.
Instead of selecting prospects because they meet a handful of predetermined demographic criteria, a predictive model can evaluate many available characteristics simultaneously and determine which combinations are most associated with the desired outcome.
The resulting model can then score a larger prospect population according to how strongly each prospect matches those conversion-associated patterns.
Traditional prospect lists often begin with a marketer's assumptions about the characteristics of likely customers.
For example, a marketer might request:
These selections can be useful when the characteristics have a clear relationship to the product or service.
But there is an important limitation:
The marketer must decide in advance which characteristics matter and where to set the thresholds.
A marketer may believe the best prospect is a 55-year-old homeowner earning more than $100,000. Actual customer data may reveal that conversion is influenced by a different combination of factors—or by interactions among characteristics that would be difficult to identify manually.
"Who do we think should be a good prospect?"
"What does the data tell us about the people who actually converted?"
Determine what the model should predict.
Examples might include:
The modeling process begins with a dataset representing the people who produced the desired outcome.
Available demographic, household, geographic, behavioral or other permitted attributes can be analyzed to determine which characteristics distinguish converters from the broader population.
Statistical or machine-learning methods identify combinations of characteristics associated with the desired outcome.
The model should be tested against data that was not simply used to train it, helping determine whether it can distinguish higher-propensity prospects from lower-propensity prospects.
The validated model is applied to prospective customers, producing scores or rankings that marketers can use to select an audience.
| ATTRIBUTE | DEMOGRAPHIC TARGETING | PREDICTIVE TARGETING |
|---|---|---|
| Starting point | Marketer-defined characteristics | Actual customer or response data |
| Primary question | Who fits our desired profile? | Who has characteristics associated with conversion? |
| Variables | Usually a limited set selected manually | Potentially many characteristics evaluated together |
| Weighting | Criteria generally chosen by marketer | Importance determined from observed data |
| Interactions | Difficult to evaluate manually | Model can detect combinations of characteristics |
| Output | Prospects meeting selection rules | Prospects ranked or scored by relative propensity |
| Validation | Often based on assumptions or campaign results | Can be tested against historical outcomes before deployment |
| Primary advantage | Simple, intuitive and easy to implement | Uses actual conversion patterns to prioritize prospects |
| Best suited for | Straightforward targeting or limited customer data | Organizations with sufficient historical customer/response data |
Demographic targeting is not inherently inferior.
If a company has little historical customer data, needs a simple audience or sells a product with obvious qualification criteria, demographic selection may be entirely appropriate.
Predictive targeting becomes especially useful when enough customer or response data exists to let the observed patterns help determine which prospects should be prioritized.
One advantage of predictive modeling is the ability to evaluate relationships that may not be obvious to a marketer.
Suppose two prospects both satisfy the basic demographic requirements for a campaign.
One may have a combination of household, property, geographic and other characteristics that occurs frequently among the company's best customers.
The other may meet the marketer's manually selected demographic criteria but share relatively few characteristics with actual converters.
A predictive model can distinguish between them.
Matches 4 manually selected demographic criteria
Predictive Score: Moderate
Does not look exactly like the assumed "ideal customer" but strongly matches patterns found among actual converters
Predictive Score: High
This is why predictive modeling can sometimes identify valuable prospects that would have been excluded by traditional list-selection rules.
A predictive score represents a prospect's relative relationship to the patterns identified by the model.
Higher-scoring prospects generally exhibit more of the characteristics associated with the desired outcome than lower-scoring prospects.
Scores allow marketers to rank a prospect universe rather than making only a yes/no selection.
This ranking can help marketers decide:
The objective is not necessarily to find every possible customer. It is to concentrate marketing resources on the portions of the prospect universe with the strongest modeled conversion potential.
Lift measures how much better a predictive model concentrates converters compared with selecting prospects without the benefit of the model.
Example: suppose a prospect population contains 100,000 people and historical data shows that 1% would normally convert.
If a predictive model successfully concentrates a larger share of likely converters near the top of its ranking, marketers may be able to reach a disproportionate number of potential converters by targeting a smaller portion of the audience.
ILLUSTRATIVE EXAMPLE — NOT ACTUAL CAMPAIGN PERFORMANCE
Prospects are often divided into groups such as deciles based on predictive score.
If conversion rates are substantially higher in the top deciles than in lower deciles, the model is successfully ranking prospects according to observed conversion propensity.
Lift does not mean every high-scoring prospect will convert. It indicates that converters are more concentrated in higher-scoring groups than they would be without the model.
One of the useful features of predictive modeling is that marketers may be able to evaluate a model against historical campaign results before using it for a new campaign.
A common approach is to separate data used for model development from data used for validation.
If actual historical converters are disproportionately concentrated among the model's highest-scoring records, that provides evidence that the model is identifying meaningful conversion patterns.
Predictive models need enough examples of the desired outcome to distinguish meaningful patterns from random variation.
The exact amount of data required depends on factors such as:
Not every marketer has enough first-party customer data to build a custom model.
There are two broad approaches to predictive prospect targeting.
Built specifically from an organization's own customers or responders.
Advantages:
Consider when:
The organization has sufficient customer or response data and wants a model tailored specifically to its business.
A predictive model developed for a defined purchase need, lifecycle event or category and applied across an appropriate prospect universe.
Advantages:
Consider when:
The organization lacks sufficient first-party data or when a proven category-specific predictive audience is available.
Inbound Insight offers both custom predictive modeling and selected pre-built Predicta Audiences™. The currently available Predicta models are:
Predictive models and lookalike audiences share a basic concept: use known customers or converters to find other people with similar characteristics.
However, implementations differ considerably.
Often created inside an advertising platform using the platform's own data, algorithms and available audience universe.
Typically optimized for activation within that platform.
Can use customer outcomes and a broader set of available attributes to score identifiable prospects outside a single advertising ecosystem.
Can create an audience that may be activated through channels such as direct mail, email or digital marketing.
The important distinction is not simply the label. Marketers should understand what data is being modeled, what outcome is being predicted, what prospect universe is being scored and where the resulting audience can be activated.
Predictive targeting can be particularly valuable for direct mail because every mailed piece carries a meaningful incremental cost.
When campaign volume is constrained by budget, marketers need to decide which prospects deserve a place in the mail file.
Instead of mailing every prospect who meets broad demographic criteria, marketers can use predictive scores to prioritize the prospects with stronger modeled conversion characteristics.
Predictive scoring can also support campaign testing. For example, a marketer could compare:
Existing demographic selection
Predictively selected audience
…and measure actual campaign response.
This makes predictive targeting particularly relevant for channels where reducing low-potential records can materially affect campaign economics.
Predictive scoring creates an important relationship between audience size and audience quality.
The highest-scoring portion of a prospect universe generally represents the model's strongest candidates.
As marketers expand farther down the ranking, they gain additional audience volume but may also include prospects with progressively lower modeled propensity.
Highest average modeled propensity
More prospects, lower average modeled propensity
This allows marketers to make informed tradeoffs between:
Predictive targeting and intent data answer different questions.
Who has characteristics associated with people who convert?
Predictive models emphasize In-Profile characteristics.
Who appears to be actively interested in the category now?
Intent data emphasizes In-Market behavior.
A prospect can be:
| Predictive score | LOW INTENT | HIGH INTENT |
|---|---|---|
| HIGH PREDICTIVE SCORE | Strong profile / timing unknown | Strong profile + current interest |
| LOW PREDICTIVE SCORE | Lower priority | Current interest but weaker profile |
Combining independent audience signals can provide marketers with a more complete view of conversion readiness.
Within the Inbound Insight Conversion-Ready Audience framework, predictive modeling primarily addresses the In-Profile dimension.
See how all four drivers work together in our guide to Building Conversion-Ready Audiences.
Are they interested now?
Online intent signals
Do they have characteristics associated with conversion?
Predictive modeling
Can they realistically buy from or visit you?
Geographic and drive-time targeting
Does the message fit the audience?
Audience insight and personas
Predictive targeting becomes even more useful when marketers consider it alongside timing, geography and messaging rather than treating audience selection as a single-variable decision.
Related: Online Intent Leads for timing and the Multi-Location / Drive-Time tool for geography.
What specific outcome is the model designed to predict?
What customer, responder or outcome data is being used?
What characteristics are available to distinguish prospects?
Is the model tested against data that was not simply used to build it?
Does historical validation demonstrate meaningful concentration of converters among higher scores?
Can the marketer understand how scores should be used to select an audience?
Can the scores be applied to identifiable prospects who can actually be marketed to?
Can the resulting audience be used for direct mail, email, digital or the channels required?
Can the model be updated as customer behavior or business conditions change?
Are the data, modeling and activation processes appropriate for the intended use?
The sophistication of the algorithm is less important than whether the model can reliably rank actionable prospects according to a clearly defined business outcome.
Inbound Insight builds predictive prospect models using an organization's best customers, responders or other conversion outcomes to identify characteristics associated with success and score larger prospect populations.
Build a predictive model around the organization's actual customers or campaign outcomes.
Use historical results where available to evaluate whether higher scores correspond with stronger conversion performance.
Apply the model to prospect populations and prioritize records according to modeled conversion potential.
Use predictive audiences in direct mail, email, digital and integrated direct marketing campaigns as appropriate.
For organizations without sufficient first-party data, Inbound Insight also offers selected Predicta Audiences™, pre-built predictive models for specific consumer needs and lifecycle opportunities.
A predictive prospect list is an audience selected or prioritized using a statistical or machine-learning model that scores prospects according to characteristics associated with known customers, responders or another desired outcome.
Demographic targeting generally selects prospects using characteristics chosen in advance by the marketer, such as age, income or homeownership. Predictive targeting analyzes actual customer or response data to determine which available characteristics and combinations are associated with conversion and uses those patterns to rank prospects.
A predictive marketing model analyzes characteristics associated with known outcomes, identifies patterns related to conversion, validates those patterns against historical data where possible, and then applies the model to a larger prospect population to produce relative propensity scores.
No. A predictive score represents relative modeled propensity, not certainty. Higher-scoring prospects have characteristics more strongly associated with the modeled outcome, but individual results cannot be guaranteed.
Lift describes how effectively a model concentrates converters within higher-scoring portions of a prospect population compared with an unmodeled or baseline selection.
The amount varies by objective and methodology. Inbound Insight generally looks for at least 5,000 customer records for a custom modeling project and ideally 500 or more responders when modeling campaign response. Individual datasets should be evaluated before determining whether predictive modeling is appropriate.
Often, yes. Historical customer or campaign data can be divided so that some records are used to develop the model and other records are used to evaluate whether higher model scores correspond with actual historical conversions.
Yes. Predictive scores can be used to rank identifiable prospects and determine which records should be included in a direct-mail campaign, making predictive targeting particularly useful when mailing costs require careful audience selection.
Predictive targeting primarily identifies people whose characteristics resemble those associated with conversion. Intent data identifies people whose recent behavior indicates potential current interest. Inbound Insight describes these concepts as In-Profile and In-Market, respectively.
A custom model is developed using an organization's own customers or responders. A pre-built predictive audience uses a model already developed for a particular purchase need, lifecycle event or category and does not necessarily require the marketer to supply a large customer dataset.
They are related concepts but are not necessarily the same. Lookalike audiences are often generated inside advertising platforms and activated within those ecosystems. Predictive prospect models can score identifiable prospects for activation across channels such as direct mail, email and digital marketing.
Predictive targeting is worth considering when an organization has sufficient historical customer or response data, needs to prioritize prospects, wants to test whether data-driven selection can outperform existing list criteria, or has meaningful campaign costs that make audience quality particularly important.
Learn how recent behavioral signals can identify consumers who appear to be actively in-market.
Read the GuideLearn how anonymous visitors to your own website can become identifiable, actionable prospects.
Read the GuideLearn how geographic targeting helps determine which prospects are realistically within reach of a location.
Read the GuideLearn how data-driven personas can be assigned to individual records and used to personalize campaigns.
Read the GuideLearn how In-Market, In-Profile, In-Territory and In-Sync signals can work together.
Read the GuideBrowse all Marketing Guides.
Inbound Insight helps marketers build better prospect audiences using predictive modeling, online intent signals, website visitor identification, geographic intelligence and data enrichment. Its solutions support direct mail, email, digital and integrated direct marketing programs.
Published by Inbound Insight · Last Updated: September 2026