Four Types of Bidding Strategies for the Open Web

Performance advertising on the open web has changed more in the last five years than in the previous 15.

The technology driving that change is bidding: specifically, the shift from manual bid management to automated models that can evaluate millions of impressions in real time and optimize toward outcomes a human team could never hit at that speed or scale.

This guide covers how automated bidding strategies work, which ones fit which campaign goals, and how to get through the setup process without burning budget while your algorithms learn.

What Is a Bidding Strategy?

A bidding strategy is the set of rules or algorithms that controls how much you pay for a given ad placement, click, or conversion in a real-time auction.

Every time a user loads a page, a programmatic auction runs in milliseconds to decide which ad wins that impression. Your bidding strategy is what determines how you show up in that auction — what you’re willing to pay, which impressions you compete for, and how those decisions get made.

Some strategies put a human in charge of those decisions. Others hand them to an algorithm. Either way, the underlying goal is the same: win the right impressions at a price that delivers a return, and avoid paying more than an impression is worth.

Get that balance right, and your budget works harder. Get it wrong, and you’re either overpaying for placements that don’t convert, or under-bidding on ones that would have.

4 Core Types of Bidding Strategies for Performance Marketers

Different campaign objectives call for different bidding approaches. Here’s a breakdown of the four strategies most relevant to performance advertisers on the open web.

1. Target Cost per Acquisition (tCPA)

Target CPA (tCPA) is the conversion-focused default for most performance marketers. You set an average cost you’re willing to pay per acquisition — a lead, a sale, a form fill — and the algorithm does the work of hitting that target across a campaign.

The key word is average. TCPA doesn’t mean every conversion will cost exactly your target number. The algorithm bids aggressively when it predicts a high probability of conversion and conservatively when it doesn’t. Some days will run above target, some below, and the goal is to hit your average across the campaign’s lifetime.

Taboola’s target CPA guide covers this well: TCPA delivers the strongest results for advertisers with clearly defined cost-per-lead or cost-per-sale goals and the data volume to support consistent optimization. In practice, it requires a minimum conversion threshold — typically 30 to 50 conversions per month — for the model to have enough signal to optimize reliably.

TCPA works best for lead generation, direct-response campaigns, subscription services, and any vertical where acquisition cost is a hard business constraint.

2. Target Return on Ad Spend (tROAS) / Value-Based Bidding

TROAS is tCPA’s more sophisticated sibling. Rather than optimizing for the number of conversions, it optimizes for the value of those conversions.

Instead of trying to acquire any type of customer, tROAS focuses on acquiring the most valuable ones.

This requires passing revenue or value data back to the platform. When you tell the algorithm that a $200 purchase happened on a given click, it learns to find more users likely to generate $200-plus transactions and bid accordingly. The result is a shift away from volume-based thinking toward value-based bidding that accounts for the actual revenue each customer generates.

For e-commerce, this is especially powerful. A campaign optimizing for conversion volume might chase lower-order transactions at a lower CPA. A tROAS campaign, properly configured, pursues high-order-value customers even at a higher individual acquisition cost — because the math works out better at the campaign level.

The trade-off is that tROAS requires more data and a more complex conversion-tracking setup. You need to pass reliable revenue values to the platform so the model has something to learn from.

TROAS works best for e-commerce brands with variable order values, subscription services with known lifetime value benchmarks, and any advertiser able to pass post-click revenue data back to the platform.

3. Maximize Conversions

Maximize Conversions is the volume-first strategy. You set a budget, and the algorithm spends it as efficiently as possible to maximize conversions. There’s no CPA guardrail as the goal is pure volume within the budget constraint.

This makes it a useful launch strategy. When you’re starting a new campaign and don’t have enough historical data to calibrate a tCPA target reliably, Max Conversions gives the model room to explore. It bids broadly, gathers conversion data across a range of placements and audiences, and surfaces the patterns that tCPA can then exploit.

As Nadim Batista-Kuttab of Xevio, one of the largest native advertisers in the world, explained in Taboola’s comparison of max conversions vs. target CPA: Max Conversions is the strategy that unlocks growth and fuels platform learning, while tCPA is the precision tool to use once campaigns have matured or hit a plateau.

The practical approach is to run Max Conversions until you have a stable conversion baseline, then transition to tCPA once the algorithm has enough data to optimize toward a specific cost target.

Max Conversions works best for campaign launches, budget-flush periods where volume matters more than efficiency, and any situation where you need to generate conversion data before tightening toward a CPA goal.

4. Viewable CPM (vCPM) for Awareness

VCPM is the odd one out in this group since it’s not a conversion-focused strategy. While tCPA, tROAS, and Max Conversions are lower-funnel tools, vCPM is an awareness-and-consideration play.

The difference from standard CPM is that you’re not paying for impressions that nobody saw. VCPM charges only when an ad meets a viewability threshold — typically 50% of the ad is visible for at least 1 second for display, 2 seconds for video. This filters out the buried placements that inflate impression counts without generating any actual exposure.

For brands running upper-funnel campaigns alongside lower-funnel performance activity, vCPM can be a cost-effective way to build reach among audiences not yet ready to convert. The goal is to move users through the consideration phase so that later retargeting and performance campaigns find a warmer audience.

VCPM works best for brand awareness campaigns, new product launches, reaching audiences early in a longer consideration cycle, and situations where visibility confirmation matters more than click volume.

How to Choose the Right Bidding Strategy for Your Campaign Goals

The choice usually comes down to three questions:

1. Where are you in the campaign lifecycle?
2. Do you have enough conversion data?
3. What’s your business objective?

If you’re launching a new campaign and don’t have historical data on your open web audience, Max Conversions is often the right entry point. It generates the data you need to make better decisions later.

Jumping straight to tCPA without adequate data leads to erratic performance as the model guesses rather than learns.

Once you’ve accumulated enough conversions to establish a reliable cost baseline, tCPA becomes the right tool. You have a cost constraint, the data to calibrate it, and the algorithm can now optimize specifically toward that number.

If you’re operating an e-commerce business and can pass revenue data back to your platform, tROAS is worth the additional setup complexity for bid optimization. The efficiency gains from value-based bidding at scale routinely outpace what tCPA can achieve on its own.

For the best performance marketing channels question, performance advertisers should default to outcome-based strategies — tCPA, tROAS, or Max Conversions — rather than awareness-based models like vCPM, unless you’re running a distinct upper-funnel campaign with brand reach as the stated objective.

Mixing optimization goals within the same campaign is a common source of performance confusion.

Here’s a practical framework to help you choose the best bidding strategy for your campaign. If:

• the campaign is new with limited conversion history, start with Max Conversions to build data.
• your campaign has 30-plus monthly conversions and a defined CPA target, move to tCPA.
• it’s e-commerce with variable order values and revenue tracking in place, test tROAS.
• your goal is upper-funnel brand awareness with confirmed viewability as a priority, use vCPM as a distinct campaign.

Manual vs. Automated: The Evolution of Bidding Strategies

Search and social built the performance advertising playbook. But rising costs per mille (CPMs), shrinking audience pools, and creative fatigue have made those channels increasingly expensive to scale.

Advertisers competing in the same auctions, for the same users, with the same formats are finding that efficiency gains are harder to come by.

The open web is where that growth ceiling doesn’t exist yet. Billions of daily impressions across premium publisher environments, reaching audiences that search and social simply can’t access. The inventory is there. The challenge is using it effectively — and that’s where manual bidding falls apart.

Manual bidding made sense when digital advertising was simpler.

A handful of keywords, a clear audience, and a maximum cost per click (CPC) set at the ad group level. That felt like control.

The open web broke that model. Millions of publisher properties, each with distinct audiences, content environments, and conversion probabilities. No human team can evaluate every impression at that scale. By the time someone identifies that certain placements are underperforming and adjusts accordingly, a machine learning model has already made that adjustment thousands of times over.

Automated bidding removes that friction.

Instead of setting a static maximum bid, you define an outcome — a target cost per acquisition (CPA), a return on ad spend (ROAS) goal, a conversion volume objective — and the algorithm works backward from that goal to determine the optimal bid for every auction it enters.

It also filters automatically, deprioritizing low-probability impressions and flagging fraudulent traffic before budget gets wasted on placements that were never going to convert.

The result is bidding that’s faster, more responsive, and — once it’s had enough data to learn from — often more efficient.

The Key Data Signals That Power AI Bidding Models

The reason AI bidding outperforms manual bidding lies in its ability to process data at a scale humans can’t replicate.

When an AI bidding model evaluates an impression, it ingests and weighs dozens of contextual signals in real time. Some of the most significant ones:

• Device type and browser: A user on a desktop Chrome browser completing a purchase-intent search behaves differently than a user casually scrolling on mobile. AI models learn these patterns at the publisher and placement level, not just the audience level.
• Geolocation: Conversion rates vary by region, city, and even neighborhood. A bidding model that accounts for location can reduce wasted spend on geographies that consistently underperform.
• Time of day and day of week: User intent and purchase behavior shift throughout the day. AI bidding adjusts in real time rather than waiting for a scheduled campaign review.
• Publisher context: The content a user is reading when they encounter your ad matters. A reader on a financial news site is in a different mindset than a reader on a movie review site. Contextual signals help real-time bidding models weigh the value of each impression more accurately.
• Historical performance data: Every conversion or non-conversion feeds back into the model. Over time, it builds a detailed picture of which impression characteristics predict success for your specific campaign.
• First-party behavioral signals: Platforms like Realize use proprietary data built over years of network activity to identify patterns that go well beyond what any individual advertiser could observe from their own campaigns alone.

These signals combine in real time to produce a single bid figure. A good impression for a high-intent user in a relevant context on a proven publisher gets a higher bid. A marginal impression gets a lower one. A consistently underperforming placement might get no bid at all.

Navigating the AI Learning Phase and Data Requirements

Every automated bidding system requires a period of learning before it performs reliably. This isn’t a flaw in the technology — it’s how machine learning works. The algorithm needs enough conversion data to identify patterns, and gathering that data costs time and money.

The learning phase typically lasts one to three weeks for most campaigns. During this time, machine learning optimization and performance will often look worse than expected. CPAs may run higher, ROAS may be lower, and the temptation to intervene is strong. Resisting that temptation is one of the harder disciplines in automated bidding.

A few things to avoid during the learning phase:

• Changing the CPA target significantly. When you shift your target, the algorithm has to re-learn what “good” looks like. Frequent target changes reset the learning process and extend the timeline before you see stable performance.
• Pausing and restarting the campaign. Every pause interrupts the data accumulation that the algorithm depends on. If the campaign doesn’t have continuity, neither does its learning.
• Making large budget changes. A sudden budget increase can force the algorithm to compete for inventory it hasn’t yet learned to evaluate, which often temporarily drives up costs.

Automated bidding strategies maintain sufficient daily budget for at least two weeks without major changes and consistently outperform more reactive campaign management. So, commit to the learning period and evaluate performance on a two- to four-week average rather than daily for the best results in smart bidding.

Best Practices for Optimizing Your Open Web Bidding Strategy

Here is how to get the most out of your bidding strategy.

Set up conversion tracking before the campaign launches

The algorithm is only as good as the signal it receives. If your conversion events are misconfigured, delayed, or incomplete, the model is optimizing toward noise. Installing the Taboola Pixel correctly and verifying that events fire reliably is non-negotiable before switching to any automated bidding strategy.

Use clean, complete first-party data

Predictive targeting and audience optimization improve significantly when algorithms have access to high-quality first-party behavioral data. Customer relationship management system (CRM) audiences, pixel audiences, and server-side event data all help the model identify users who resemble your best customers. Feed the algorithm better inputs, and it returns better outputs.

Calibrate your CPA target carefully

Setting a target that’s too aggressive relative to what your campaign can realistically achieve leads to under-delivery. Setting it too loose leads to inefficient spending. The right starting point is usually close to your historical average CPA from other channels, then adjusted based on actual open web performance data as it accumulates.

Don’t judge performance by day-one numbers

Data-driven campaign management means evaluating performance over a meaningful window, not a 24-hour snapshot. Automated bidding strategies typically require two to four weeks of data before performance stabilizes. Building that evaluation period into your reporting cadence prevents premature campaign changes.

Test creative continuously

The bidding strategy is not the only lever. Motion ads, in particular, give AI bidding models richer engagement data to optimize against. Creative fatigue is a real performance drag, and algorithms that work with fresh, varied creative consistently outperform those running on exhausted, static assets.

Measure true performance with multi-touch attribution

Post-click conversions don’t capture the full impact of open web campaigns. Controlled experiments, blended metrics such as the Marketing Efficiency Ratio, and third-party attribution tools provide a more accurate picture of what your campaigns are actually driving. Walled-garden attribution consistently overstates performance — measuring open web campaigns with the same rigor prevents that distortion.

The Future of AI in Media Buying

The shift from manual to automated bidding is the first act. The second act is already underway.

AI agents are beginning to handle tasks that still require human input today: budget reallocation across channels, creative generation and testing, audience expansion decisions, and bid strategy selection itself.

Platforms like Realize are already moving in this direction, using predictive models trained on years of proprietary first-party data to surface recommendations and automate decisions that previously required an analyst.

Predictive targeting represents one of the clearest near-term applications. Rather than building audiences based on demographic similarity to existing customers, predictive models identify users who are actively demonstrating intent-consistent behavior — reaching high-value prospects earlier in their decision cycle, before they’ve entered the saturated search and social auctions where every competitor is vying for the same clicks.

The open web’s interoperability challenge is real, but platforms investing in unified data layers and privacy-compliant signal management are building infrastructure that makes advanced AI optimization increasingly viable outside the walled gardens.

As that infrastructure matures, the performance gap between open web campaigns and search/social will continue to narrow for advertisers willing to invest in the technology stack.

The long-term trajectory looks like automated bidding becomes table stakes, and the competitive advantage shifts to the quality of the data feeding those models and the sophistication of the AI systems interpreting it.

Key Takeaways

The core lesson from the shift to AI bidding strategies is simple: human decision-making doesn’t scale to the volume and complexity of modern open web advertising.

Automated and AI-powered bidding strategies — tCPA, tROAS, Max Conversions — give performance advertisers a way to compete at scale across the open web without requiring a team of analysts to manage every placement and adjustment manually.

The practical path forward:

• Match your bidding strategy to your campaign stage and objective.
• Commit to the learning phase without premature intervention.
• Feed the algorithm clean, complete conversion data.
• Evaluate performance over meaningful time windows, not daily snapshots.
• Layer in value-based optimization as your data matures.

Stop treating the web as a secondary channel. You already have the tools needed to compete at scale, so use them.

Frequently Asked Questions (FAQs)

What is a bidding strategy?

A bidding strategy is a specific set of rules or algorithms that determines how much an advertiser is willing to pay for an ad placement, click, or conversion. It governs how your budget is allocated across real-time ad auctions, and whether those decisions are made manually by a human or automatically by a machine learning model.

How does an AI bidding strategy differ from manual bidding?

Manual bidding requires an advertiser to set and periodically adjust a maximum bid at the campaign or ad group level.

An AI bidding strategy uses machine learning to analyze signals — device type, geolocation, browser, time of day, publisher context, and historical conversion data — in real time, adjusting each bid for bid optimization based on the predicted likelihood of a conversion.

The result is faster, more granular optimization than any manual process can achieve at scale.

Why should performance advertisers expand to the open web?

Search and social audiences are finite and increasingly expensive. The open web offers access to a vastly larger audience across premium publisher environments, at CPMs that are often more efficient than those in saturated walled-garden auctions.

Paired with AI bidding technology like Realize, the open web can deliver comparable or better customer acquisition costs — with the added benefit of reaching users in content-engaged, high-intent contexts that search and social can’t replicate.

What is the difference between target CPA and target ROAS?

Target CPA optimizes your campaign to acquire conversions at a specific average cost. The algorithm bids based on conversion probability without differentiating between the value of different conversions.

Target ROAS is a value-based strategy that optimizes toward the predicted revenue a given user will generate — prioritizing high-value transactions over raw conversion volume.

TROAS requires that revenue data be passed back to the platform and generally performs best in e-commerce and subscription contexts where conversion values vary meaningfully.

How long does the AI learning phase take?

Most AI bidding algorithms require one to three weeks to accumulate enough conversion data to optimize reliably. The exact duration depends on campaign budget and conversion volume — higher-spend campaigns with more frequent conversions learn faster.

During this period, performance will often look inconsistent. Avoiding major changes to bids, budgets, or targeting during the learning phase is important for giving the model enough continuity to stabilize.