How Predictive Optimization works

Lead generation suffers from a serious problem that doesn't exist in ecommerce advertising: the delay between when a lead is generated and the sale is made.

In ecommerce, purchase value can be immediately sent back to the ad networks for optimization. That enables their machine-learning algorithms to decide who should see each ad, how aggressively to bid and where to deliver it, maximizing the advertiser's net return. This is known as value-based optimization.

With lead generation, on the other hand, it often takes weeks or months for leads to mature and final sales numbers to come in. Until then, the ad networks don't know what those leads are worth. The result is that leadgen advertisers are largely shut out of value-based optimization.

That limitation can have a substantial impact on net return. AdEspresso ran a Facebook ecommerce test that held the ads and targeting constant and changed only the optimization method.

Value-based optimization increased one retailer's ROAS (return on ad spend, or the revenue generated for every dollar spent) from 1.08x to 3.06x. The exact same ads. The exact same targeting. And nearly 3x the revenue from the same spend. The only difference was the optimization strategy.

It's no surprise that ecommerce advertisers have adopted value-based optimization almost universally. When Smarter Ecommerce audited 3,000 retail ad campaigns on Google, it found that 95% were using a value-based bidding strategy. For ecommerce, it's standard practice. For lead generation, it's been out of reach.

Predictive Optimization changes that. I'll explain how in a moment, but first let's look at the problem with the proxy metrics that most companies use today and why they fall short.

Why CPL is a poor indicator of advertising performance

Cost per lead is one of the most common, and also one of the worst, metrics used to measure leadgen performance. It tells you what you paid to generate a lead, but nothing about how likely that lead is to become a client or how much revenue it may ultimately produce.

In the example above, Ad A looks better based on CPL alone: $30 compared with $44 for Ad B. But Ad B converted at about 11% versus 6%, produced a lower CAC (customer acquisition cost), generated a higher ROAS, and delivered nearly $20,000 more net return.

Judged by CPL, Ad A appears to perform better. Judged by actual economic performance, Ad B is the clear winner. Cutting Ad B and scaling Ad A based on CPL alone would be a costly mistake.

When a campaign is optimized to generate leads at the lowest possible cost, the ad networks do a great job of getting you more leads for your budget, but at the expense of everything that happens after the lead is submitted.

CAC and conversion rate are better than CPL, but still incomplete

Conversion rate and CAC (customer acquisition cost) improve on CPL because they account for whether leads actually become clients. But they treat every conversion as equally valuable.

In the example above, Ad C and Ad D look about the same by these metrics. They both convert at about 8%, and their CAC differs by only $7. So you might conclude that these ads perform about the same.

But let's dig deeper. Despite having nearly identical conversion rate and CAC figures, Ad C generated roughly $54,000 more net return and a considerably higher ROAS.

How could this be? The hidden difference is in average client value. The clients generated by Ad C were worth over $1,200 more on average than those generated by Ad D.

CAC tells you what it costs to acquire a client. Conversion rate tells you what percentage of leads become clients. Neither tells you what those clients are worth.

Importantly, they also don't change what the ad network itself is optimizing for. If the campaign is optimized for leads, the ad network is still trying to generate more leads for the same budget, regardless of their economic value.

CAC and conversion rate can help you decide which ads to scale or cut, but they can't help the algorithm decide which people to put your ads in front of.

Mid-pipeline signals unlock algorithmic optimization, but miss client value

A more advanced approach is to send mid-pipeline CRM events such as Contacted or Qualified back to the ad networks. Now they can see which individual leads are progressing through your pipeline and use their algorithms to find more people likely to reach those stages.

This solves an important part of the problem by unlocking algorithmic optimization, where the ad networks automatically adjust targeting, bidding and delivery based on downstream outcomes.

Instead of optimizing only for people likely to become leads, the ad networks can now target people more likely to progress through your pipeline and ultimately convert. This aligns the ad network algorithms more closely with the results you actually care about.

But we're still left with the same problem as before: the ad networks don't know what those leads are worth to your business.

Two leads can reach a status of Qualified in your CRM and end up with very different sale values. Lead A may be expected to generate $1,500 in revenue while Lead B is worth $5,000.

To the ad networks, however, they're two identical Qualified conversions. That means the algorithms have no reason to favor the types of leads that go on to become higher-value clients.

Predictive Optimization adds the missing economic signal

Predictive Optimization solves this by assigning each lead its own projected value before the sale occurs. Instead of treating every lead at the same stage as equal, the ad networks receive a value based on each lead's expected economic outcome, using the following formula:

Projected sales value = conversion probability × expected client value

Conversion probability estimates how likely the lead is to become a client based on its current pipeline status and historical conversion rate. Expected client value estimates how much that client will be worth based on industry-specific lead attributes.

The expected value part of the equation will look different for every industry. For a debt firm, unsecured debt is the primary driver of client value. A mortgage broker might use loan size, while an insurance agency could use coverage amount or policy type.

Going back to our previous example, the two Qualified leads are no longer identical to the ad networks. Assuming a historical conversion rate of 20% for leads that hit the Qualified status in your CRM, the ad networks would now receive these projected values for leads A and B:

Lead B now carries more than three times the economic weight of Lead A, giving the ad networks a reason to favor people more likely to produce leads like B.

Instead of optimizing for the lowest cost per lead or the most qualified leads, the ad networks can now seek out the leads expected to drive the highest ROAS and net return. This is how Leadworks unlocks value-based optimization for leadgen advertisers.

Choose the pipeline stage that gives the ad networks the best signal

Leadworks lets you choose the preferred CRM status that triggers the projected value signal, whether that's Contacted, Qualified, Appointment Booked or any other stage in your pipeline.

Earlier stages give the ad networks more data, faster, which can help the algorithms learn more quickly. The downside is that greater uncertainty around the final outcome makes the projected values less accurate than later stage signals.

Later stages provide a stronger, more accurate signal because there is greater certainty about the eventual outcome. The downside is that fewer leads reach those stages, and the signal arrives later.

As a rule, we recommend using the deepest pipeline stage that still gives the ad networks enough data to learn from. Advertisers with shorter sales cycles and higher lead volumes can use later stages, while those with longer sales cycles or lower lead volumes may get better results from an earlier trigger.

Switch your campaigns to value-based bidding

To put those projected values to work, simply switch your campaigns to a value-based bidding strategy. On Meta, use Maximize value of conversions; on Google and YouTube, Maximize conversion value or Target ROAS; and on TikTok, Highest Value or Minimum ROAS.

This tells the ad networks to use the projected values when making targeting, bidding and delivery decisions. Higher-value leads carry more weight, allowing the algorithms to seek out the people most likely to drive greater ROAS and net return.

Predictive Optimization starts with a better signal and improves over time

Compared with optimizing for CPL, conversion rate, CAC or mid-pipeline events, Predictive Optimization gives the ad networks a more accurate signal of each lead's expected economic value from day one.

As more leads and sales come in, Leadworks learns from the results, adjusts its algorithm, and continually improves its projections. Over time, the signal becomes increasingly accurate and better calibrated to your business.

Align the ad network algorithms with your bottom line

Value-based optimization is no longer limited to ecommerce advertisers. Predictive Optimization gives leadgen advertisers the same advantage by feeding projected sales values to the ad networks before their sales cycle is complete.

Instead of optimizing for metrics that don't align with your business goals, you can now use the ad networks to maximize what matters most: your net return.