Background on Bidding Strategy

Two of the most widely used automated bidding strategies on platforms like Google and Meta are target cost-per-action (tCPA) and target return on ad spend (tROAS). Both fall under the umbrella of “smart bidding” and follow a similar structure: you define a reward signal for the ad platform, whether it’s a conversion event in tCPA or revenue in tROAS, and then specify a target efficiency constraint. You can also include secondary constraints, like spend limits or frequency caps, to keep performance within your budget tolerances.

This setup simplifies your interaction with the ad platform. Where advertisers once had to submit thousands of granular bids across keywords and audience segments, smart bidding lets you focus on goals and inputs, while the platform manages impression-level bidding in real time.

But make no mistake, the underlying auctions are still happening; you’re just no longer managing them manually. Whether you’re using tCPA, tROAS, or newer products like Performance Max, the ad platform is now constructing bids on your behalf using each user’s data profile and context (such as search queries, demographics, or placement). Your success hinges on whether those bids are aimed at high-value users, which in turn depends on whether you’re setting the right goals and targets to win those impressions.

In this post, I’ll focus on how to combine these ML-driven signals with the right bidding strategy and how to choose between tCPA and tROAS based on your business model and goals.

Supercharging Your Bidding Agent

Your bidding strategy (e.g., tCPA vs tROAS), optimization goals (e.g., conversions, revenue, or profit), and the data you share with the ad platform collectively define a “bidding agent” a system that competes for ad impressions on your behalf. The more intelligently that agent is configured, the better it performs. Ideally, it should bid aggressively on high-value users in favorable contexts (like relevant keywords or strong audience segments), while ignoring low-intent traffic that wastes budget. Having worked on bidding systems at both Microsoft Ads and Booking.com, I’ve seen how much of an edge advertisers can gain by sending smarter signals.

Let’s explore how predictive modeling can shape this bidding behavior through real-world examples.

Predictive Bidding

Predictive bidding is an advanced approach to digital advertising where advertisers leverage predictive modeling techniques to enhance the signals they send to bidding platforms. Instead of relying solely on historical averages or simplistic conversions, predictive bidding uses machine learning to anticipate user value, intent, and profitability before bidding decisions are made. By adopting predictive bidding, advertisers significantly boost their campaign efficiency, reduce wasted ad spend, and maximize long-term returns.

In tROAS marketing, predictive bidding is particularly valuable. The tROAS strategy inherently rewards higher-value conversions, but without predictive modeling, the true potential of this approach often remains untapped. By integrating predictive signals, such as anticipated customer lifetime value, predicted profit margins, or likelihood of repeat purchases, advertisers provide platforms like Google and Meta with stronger, more precise guidance. This approach ensures your campaigns don’t simply chase short-term revenue but optimize toward meaningful long-term profitability.

Advertisers using predictive bidding in conjunction with tROAS marketing consistently outperform competitors who rely on traditional bidding methods. They gain the ability to dynamically adjust bids not just based on immediate sales but on anticipated long-term customer value, effectively balancing short-term ROI with strategic, growth-oriented objectives.

Predictive bidding isn’t limited to e-commerce; it also offers significant benefits for sales-led businesses. By predicting lead quality and future conversion probability, predictive bidding ensures ad spend is laser-focused on truly qualified leads, maximizing the productivity and morale of your sales teams.

Ultimately, predictive bidding provides advertisers with a strategic advantage, enabling smarter, more profitable decision-making that aligns closely with broader business objectives.

Example 1: Qualifying Leads to Improve tCPA Efficiency

Scenario: Two advertisers sell solar panel installations (~$10,000 per deal), closed by human sales reps. They acquire customers through an online quote estimation flow and rely heavily on search and social advertising.

• Advertiser A defines a conversion as submitting a quote form and uses a simple web pixel to report form submissions.
• Advertiser B uses ML to predict installation likelihood, based on factors like location, user intent signals, and budget. Leads scoring below 2% probability are zeroed out, so only qualified submissions are treated as conversions.

Both use tCPA bidding, targeting form submissions, but there are key differences:

Why it works: Advertiser B’s bidding agent operates with higher confidence per conversion, allowing a higher tCPA and enabling it to outbid A for high-intent impressions. Since the ad platform is financially incentivized to maximize return, it will prioritize Advertiser B, whose conversions yield greater downstream value.

Example 2: Using Fractional Conversions in tCPA Bidding

Follow-up question:
Why doesn’t Advertiser B just define the goal as a final installation and send a fractional conversion value (e.g., 0.05) for every qualified form lead? For example, if the average conversion rate for “qualified leads” in Example 1 was 5% and their bid was $20, this would correspond to a $20 x 1/0.05 = $400 bid per conversion. Wouldn’t this high target help them capture even more high-value users and further beat their competition?

This strategy has merit, but there’s a caveat. While the tCPA bidding supports fractional conversions, this treats the following scenarios as equal value:

• 10 leads with 1% conversion probability

• 1 lead with 10% conversion probability

For some businesses (e.g., e-commerce), that’s fine or even preferable. More exposure can build brand value when the cost to serve low-intent leads is effectively zero. But for sales-led organizations, low-intent leads waste sales resources and demotivate agents.

Recommendation: Sales-led organizations should start with a qualified-lead-based conversion strategy, then experiment with fractional conversions only after establishing a strong baseline lead quality. The merits of fractional conversion can be compared to a strong baseline, and lead quality can be closely monitored.

Example 3: Using Predicted Profit for tROAS Optimization

We can extend this example to consider tROAS bidding. Predictive modeling can be used to estimate the profit contribution of the user and send this as a signal back to the ad platform, as the “returns” in a tROAS bidding strategy. The advantage of this approach is that users with different profitability, for example, based on the expected size of purchase, will be treated differently. Many e-commerce firms have widely varying basket sizes and repeat purchase propensity. By predicting total profit contribution over a longer period, such as 90 days or longer horizons (customer lifetime value), differences across conversion values are properly accounted for.

My recommendation for e-commerce companies is to start with 30-90 day prediction windows, since actuals can be verified during testing periods, and pair this with a tROAS bidding strategy. For a sales-led organization, the caveat is similar to fractional conversions. Two users might contribute the same total predicted profit but differ in actual likelihood to convert. If purchase size varies significantly across customers, a more attractive alternative is to use the lead qualification strategy with a predicted profit contribution cutoff.

Key Takeaway

By pairing tCPA or tROAS strategies with predictive modeling, you improve signal quality, gain control over which users you reward, and empower the ad platform’s ML to bid smarter on your behalf. In our example, Advertiser B wins by giving the platform the confidence to bid higher, more accurately, and more efficiently than the competition, capturing the most valuable users and winning impressions in the most valuable contexts.

tCPA vs tROAS? Choosing the Right Bidding Strategy

Many advertisers eventually move from tCPA to tROAS as their data maturity improves. While tCPA is often the best starting point for simple acquisition goals, tROAS provides more flexibility when you have visibility into purchase value and long-term profitability. This transition is especially powerful when paired with predictive bidding, ensuring that campaigns don’t just optimize for volume but for sustainable ROI.

When to Use tCPA

tCPA bidding is a great fit for growth-focused advertisers who prioritize customer acquisition over immediate revenue or margin. If you’re just beginning to integrate ML-based predictive signals into your bidding strategy, tCPA is often the best starting point, especially if you’re transitioning from proxy-based conversion tracking (e.g., form submissions or shopping cart events). It’s simple to implement, easier to control, and gives you a clear metric to optimize toward.

E-commerce companies should lean towards fractional conversions or use predicted profit to qualify leads, especially when profitability per customer varies widely. Sales-led organizations should focus on cost per qualified lead, with qualification thresholds based on either predicted conversion or predicted profit, depending on business goal.

When to Use tROAS

tROAS provides more flexibility and allows you to directly account for variable revenue or predicted profit per user. It’s ideal when:

• You have good visibility into purchase value

• Your users vary widely in their long-term profitability

• You want to optimize for return on investment rather than conversion volume

• You do not have a sales-led conversion funnel

A key drawback is that tROAS can undervalue new customers, who tend to make smaller first-time purchases, unless you are using sophisticated customer lifetime value projections. In my experience, the solution is to combine predicted profit with a “new customer bounty” calibrated to balance short-run ROI with long-term customer building. Once you leverage predictive modeling to run your tCPA and tROAS bidding, you can make these sophisticated strategic tradeoffs.

Want to Start Using Predictive Signals This Week?

With Gencomm, you can:

• Deploy no-code predictive models aligned with your goals, delivered in less than a week, and connected directly to your CRM, database, and marketing platforms

• Use Gencomm’s tunable lead qualification score to send smarter conversion signals to platforms like Google and Meta

• We make it easy to get started. After a short onboarding consultation, our system will collect data and build your modes. You’ll be live within a week!

Want help applying predictive modeling to your bidding strategy? Start with a one-month free trial or book a demo with our experts to see how Gencomm can transform your marketing effectiveness.