Improve Paid Advertising Conversion Rates With Deep-Funnel Optimization

Introduction to Deep Funnel Optimization (DFO)
- Overview of the importance of lead quality over quantity in demand generation.
- Initial breakthrough with Frederik Hermann in reducing customer acquisition costs and improving lead quality and conversion rates.
Demonstration of DFO Across Various Dimensions
- Application of DFO across multiple companies, Ideal Customer Profiles (ICPs), geographies, industries, and target markets.
- Significant reduction in mid-market sales cycle duration highlighted.
Definition and Importance of DFO
- Explanation of DFO as a strategy focusing on high-quality lead generation and conversion rate maximization at every funnel stage.
- Emphasis on post form-fill activities as indicators of genuine interest.
Operationalizing DFO
- Description of practical steps for implementing DFO: identifying meaningful post form-fill conversion events and adjusting conversion reporting accordingly.
- The strategic delay of conversion acknowledgment until MQL status is reached.
DFO as a Catalyst for Efficiency and Insight
- Introduction to the OODA loop concept as it relates to DFO, highlighting the competitive advantage of faster insight and action.
- The role of DFO in driving efficient lead delivery, faster deal closures, and enhanced marketing ROI.
The Detriment of Pursuing Cheaper Leads
- Critique of the strategy to pursue ever-cheaper leads, highlighting its inefficiency and the superior strategy of optimizing for lower funnel events.
The Mechanism of DFO
- Discussion on how DFO works, leveraging advanced machine learning and changing the focus from a high volume of cheap leads to fewer, high-quality leads.
- Importance of sending meaningful events and values to advertising platforms.
Revenue Generation Processes Enhanced by DFO
- List of seven fundamental revenue generation processes improved by DFO, from messaging to customer engagement.
Stages of Company Growth and DFO Application
- Advice on implementing DFO at different company growth stages: early, mid, and late.
- Importance of both real and synthetic events for DFO success.
Technical Aspects of Implementing DFO
- Discussion on the use of machine learning for synthetic event generation and scoring.
- Insights into building vs. buying tools for DFO implementation, including the use of third-party tools like 6sense, RevSure, and Gradient Works Market-Map.
Implementing DFO – Tools and Systems
- Overview of tools and systems for DFO implementation: CRM systems, Marketing Automation Platforms (MAPs), and custom builds.
- Importance of data integration and continuous monitoring for effective DFO strategy execution.
Conclusion: The Value of DFO
- Emphasis on the critical role of calculating accurate values for DFO conversion events.
- Final thoughts on the transformative power of DFO for demand generation and the importance of continuous improvement and adaptation in marketing strategies.

Riding the Wave of Quality over Quantity: In the world of demand generation, all clicks are not equal — they’re magical only when they convert into closed won deals.

In 2019 Frederik Hermann and Harry Hawk teamed up to improve lead quality and conversion rates, and reduce customer acquisition costs. Our 1st breakthrough came quickly, cutting the cost of closed won deals by half. Based on 5 years of learning, we are here to share, “Why Deep Funnel Optimization Needs to Be Demand Gen’s ‘Ride or Die.'”

Over multiple companies, different ICPs, geographies, industries, and targets (SMB, Mid Market, and Enterprise) we have demonstrated that Deep Funnel Optimization (DFO) is the most effective tool for demand generation optimization. In Q4 2023 deep funnel optimization reduced our mid-market sales cycle from 140 days to just 65. A lot has changed since we saw our first success (2019)… here is what we have learned along the way.

What is the Deep Funnel?

Lower funnel optimization, also known as deep funnel optimization (DFO), is a strategic shift prioritizing high-quality leads and maximizing conversion rates at every stage of the funnel. Most of demand generation from conception, tracking, and reporting revolves around initial form-fills.

DFO shifts the focus to post form-fill events (real or synthetic — we will explain below) Two basic examples are attending a webinar or clicking a link within a downloaded eBook. These are events that show an increased or genuine interest in your product or service, thereby signaling increased likelihood of conversion.

There are many more examples, here are three:

Successfully scheduling a demo call
Completing a discovery call.
Reaching the next Opportunity Stage

Making Google & Meta ads work harder

Huge percentages of marketing budgets go to demand gen and if you are like most marketers you always have a ‘hunch’ those dollars could be working harder. DFO “makes” Google/Meta work harder so your smarketing team can focus on delivering value while helping ensure BDRs have time to “attack” leads within minutes not days.

The linchpin of DFO is upgrading or evolving how conversions get reported to Google, Meta, and other ad platforms. In some cases we have eliminated or delayed conversion reporting for “form-fills” (leads) shifting to acknowledging form-fill conversions only when a lead reaches MQL status. This tactic is also compatible with reduced use of gated content.

Is DPO a Flywheel?

Yes. DFO drives efficiency — you’ll deliver fewer but higher quality leads to the sales team. Deals will close faster, and your marketing team will be able to get “inside the OODA loop” of your competition. OODA stands for Observe, Orient, Decide, Act.

Observe, Orient, Decide, Act

The OODA loop concept was developed by military strategist John Boyd and describes the process and speed of decision making from combat operations to corporate strategy. In highly contested markets gaining insight faster than the competition is a competitive advantage.

Getting inside your competitors’s OODA loop means you can act and react to market conditions faster.

By the time your competition observes, orients, and decides on a course of action, your team is already several steps ahead.

By embracing DFO, demand generation experts can drive sustainable business growth, improve return on investment (ROI), and ensure that marketing efforts are focused on the prospects most likely to convert into loyal, high-value customers. We will cover how DPO and OODA work together later in this paper.

Reducing fully weighted costs

Especially for venture funded companies, the two crucial “Northstar” metrics are the fully weighted cost of customer acquisition (CAC) and actual Lifetime Value (LTV). For B2B, B2C, and critically for SaaS revenue generation teams must be laser focused on decreasing CAC and increasing LTV. DFO is the best tactic to efficiently deliver both profitability and scale.

Narrator: Many marketers trying to scale fall into the siren’s call of lower cost leads. Warning: It’s a death trap!

Faster Cheaper Leads

The Lead Gen fallacy is the belief that an ever increasing volume of ever cheaper leads will result in more opportunities, better opportunities, more closed won deals, faster deals, and a better CAC to LTV ratio! I’ve successfully increased spend, lowered CAC, and increased conversions all at the same time but not by chasing cheaper leads. Cheap leads are harmful — wasting marketing and sales resources, increasing payback time and creating diminishing returns. Scaling cheap leads is a negative flywheel.

Narrator: There is a better way. Optimize for lower funnel events!

Fewer more expensive leads?

Would you pay more for fewer better quality leads? Most marketers would, yet just spending more isn’t going to help. If you’re like us you worry if you are really getting better leads. The fear often comes from managing complex attribution, balancing arcane business rules, and perhaps suffering from a smidge of imposter syndrome. We’ve seen sales and marketing leadership double-down on cheaper leads.

Narrator: Fast cheap leads are always the wrong choice

It’s not magic, but DFO works!

It’s a truism or axiom that better quality leads are better leads.., but getting them is a complex multi-process motion. DFO is only one of several key processes and paradoxically by using DFO the cost per lead (CPL) may double or even triple, yet the cost per won deal will be reduced by up to 50%. On average, we saw the cost per deal drop by 33%. These results are achievable because they significantly leverage the huge investment Meta and Google have made in machine learning over the last 10 years.

Better signals mean better results!

We will dive into the nuts and bolts of what’s needed to make this work, but first we want to make sure that we’ve conveyed why DFO is mission critical, not just something “nice to have.” DFO is a catalyst that will improve your entire revenue generation process.

Revenue Generation

Here are what Frederik and I see as the 7 fundamental revenue generation processes:

Messaging potential customers in their own voice
Fresh creative, optimized for every buyer journey stage
Nurturing that delights while educating
Robust attribution and reporting
Rapid lead attack time and consultative selling
Effective onboarding
Engaging customers to learn why they love your product!

DFO is mission critical because it makes all of these processes more effective. DFO helps you continuously optimize each of those processes. In a contested marketplace, teams with faster deal cycles win!

Continuously better, not perfect

DFO doesn’t require perfection. Embracing DFO means giving up perfection for continuous improvement. The marginal benefit of continuous change means everything is always getting better! Getting all 7 revenue generation processes optimized is hard to achieve, DFO (as a catalyst) makes it easier!

Yours vs. Ours

Your top seven revenue gen processes may be different from ours… what matters is that your team is putting effort into improving them. Day-to-Day, your focus may shift among them, we know it’s a balancing act. If your team is doing the work, DFO will work for them.

Early Stage

If you’re at an early stage and a lot of these processes are underdeveloped it’s critical that you start using DFO right from the start. The real payoff however, comes when all the pieces are in place. You’ll need to maximize synthetic events.

Mid Stage

You have product market fit (PMF) but you still struggle to get good leads. You’ve raised funding, but your team would like to raise more. You are focused on getting better, faster, and more effective: DFO was built for this, but you are late to the party. Don’t delay your implementation. You have reasonable DFO conversation event frequency, you’ll need to focus on high-quality real and synthetic events.

Late Stage

You have product market fit (PMF), you have mature sales and marketing motions. You’ve had multiple funding rounds and you may be generating 1000s of leads a month. You are probably looking to gain further scale. DFO will work here because you already have a high volume of quality events. You’ll need to refine how you caculate value for them. Given the high volume, you can rely less on synthetic events (we promise we will explain that below)!

Putting the pieces together

Here are the DFO basics:

Identify your meaningful post form-fill events
Send those events (via API) to Google and Meta
Ensure each event has a meaningfully calculated value

Meaningfully? This means observing events that lead to new opportunities in sales. It means assigning values that are more variable or stochastic than fixed or deterministic. It also means sending conversion events at a high enough frequency to impact the algorithms deployed at Google and Facebook. Let’s break that down…

Fixed or Deterministic Values

Fixed or Deterministic: This approach involves assigning predefined values to events based on a specific formula or set criteria that do not change based on the situation or over time. The value of each event is predictable and does not vary from one instance to another.
Example: If every form submission that leads to a scheduled appointment is valued at $100, regardless of the service inquired about or the customer’s potential lifetime value, the assignment of this value is deterministic.

Variable and Stochastic Values

Variable and Stochastic: Contrary to fixed values, this approach acknowledges the inherent unpredictability and variation in the importance or value of each event. Values are assigned based on a range of possible outcomes, influenced by uncertainty and potentially changing factors.
Example: Instead of a flat rate for all scheduled appointments, the value assigned to a form submission leading to an appointment could vary based on the expected revenue from the customer (which could be influenced (“weighted”) by their interests, past purchases, firmographics, or demographics).
This could mean a form submission for a high-value service might be valued at $200, while one for a lower-value service might be valued at $50. The exact value might also be adjusted based on additional stochastic factors, such as the likelihood of the customer making a purchase based on their engagement level.

Machine learning needs rougher not smoother data

Given the same type of event the same value is a problem. It’s a problem especially because tools like HubSpot make it super easy to send fixed values, and difficult if not impossible to implement with variable or more stochastic values. Statistically, this works, but for DFO it fails. You can calculate the average value of an SQL or a discovery call as a fixed value, but it will “starve” the ML processes at Google/Meta of critical information.

From the perspective of machine learning, you’re saying that all of those “similar” events (an SQL or a discovery call) are worth exactly the same. As marketers, we know that they’re not equally worthy. Frederik and I will show you (below) our strategies for computing event values that incorporate critical signals like geography, industry, and firmographics.

Remember: oversimplification leads to models that are reacting to overly generalized, inaccurate representations of reality. This not only hampers the ML model’s ability to make precise predictions but also undermines the effectiveness of targeted marketing strategies.

Don’t Smooth Data

Data smoothing for the sake of simplicity can render a machine learning model less sensitive to the unique attributes that distinguish a high-value lead from a low-value one. Smoothing data in machine learning, especially when evaluating events for DFO, inherently introduces inaccuracies (or generalities) that can significantly distort the ground truth. By treating disparate events or leads as if they have similar or identical values—such as bucketing every lead into simplistic scores like 1, 2, or 3—we strip away the nuanced differences that are critical to understanding true customer behavior and value.

Add Noise

In refining our Deep Funnel Optimization (DFO) process, Frederik and I have strategically incorporated an element of “noise” into the valuation of leads, using nuanced signals such as geography, industry, technology usage, and firmographics. This deliberate addition of variability counters the risk of overly homogeneous data, which can lead to a flattening of the predictive power of ML models.

There are several ways we achieve this but a key method is assigning weights that reflect the complex (and changing) reality of each lead effectively broadening the data spectrum. This approach not only acknowledges the inherent diversity in lead quality but also enhances the robustness of Google’s and Meta’s machine learning algorithms.

The “noise” introduces a level of real-world unpredictability into the system, preventing overfitting and ensuring that our conversion signals are adaptable, resilient, and capable of capturing the subtleties of customer behavior. This nuanced and dynamic method of scoring “leads” fosters a more accurate and effective DFO strategy, driving meaningful engagements and conversions.

Meta and Google Conversions

Google calls deep funnel events “offline conversions” or “offline events” and the resulting conversions are “imported.” In the Meta ecosystem, deep funnel events are reported via the Conversion API (CAPI), originally the “server-side API.” To understand DFO, It’s important to understand that these techniques were originally developed for e-commerce, especially DTC. 99% of the time all eCommerce transactions occur online and the valuation of the actions taken are directly discernable.

Meta’s and Google’s algorithms were trained on high quality and high frequency signals. eCommerce signals flow every time somebody loads a page in eStore, looks at a product, watches a video, or dives into a product detail page (PDP). The algorithm also “see’s” the incremental value of each product added to a shopping cart. The culminating event are purchases and repeat purchases sent with the dollar value of those transactions, and often a taxonomic categorization of the purchase.

Signals outside of eCommerce

It’s easy to see that outside of eCommerce, there are a lot fewer signals/events available. It’s harder to calculate the value of an event as well. B2B/B2C events occur over a much longer period of time (from minutes to months). When a purchase event finally takes place it is often long after the attribution window has closed. When we say, “Send Meaningful Events” we mean the number of unique high quality events, the frequency they are sent, and sending high quality values as well.

Sending Meaningful Events: Real & Synthetic

When you are sending post form-fill events to Google or Meta you need to pick the best ones, and you need to assign a value for each event that is unique for the specific contact. As we detail below, both companies and contacts (within those companies) need to be uniquely scored. As noted above, sending the same value for the same event (MQL, SQL, Stage 3 Opp, etc.) isn’t going to help.

Real vs. Synthetic Events

A real event is one that occurs in a tangible way: A webinar sign-up, attending a webinar, or asking a question in a webinar are all real events. Other real events include scheduling a demo call, completing a discovery call, requesting a quote, a contract, or making a purchase.

Sythethetic events are based on intangible or intuited actions. For example, after a contact completes a number of real events, many marketing teams (based on scoring) will assign MQL status to the contact. Typically there isn’t one specific or discrete action that triggers the MQL status. Typically there are many paths/steps to MQL status. Some of those paths may only have 1 or 2 steps like requesting and scheduling a sales call. Reaching MQL status is a synthetic event.

All companies esp. early and middle stage start-ups, to get enough meaningful events will need to rely on some synthetic events.

Here are a few of the sources of signals that can create high volume high quality synthetic events. Note: Meta/Google recommend a minimum number of events per week.

You are using account based marketing (ABM). There are account level signals that can be attributed to individual contacts.
You are using tools like 6sense, RevSure, Gradient Works Market-Map, Influ2, etc that create synthetic signals for companies, contacts, or both. An example is 6sense’s 6QA score.
You are using intent signals from companies like Bombora, or “technology used” signals from companies like LiveRamp.
Changes within your company like completing a product roadmap milestone that will meaningfully improve PMF for a subset of leads is also a signal source.
Any change in lead score is by itself a synthetic event. Typically only positive events are sent.
Interpolated events. Creating events that didn’t occur because of how leads progress through the funnel. If a lead jumps directly from Stage 1 Opportunity to Stage 4 Opportunity you could logically create a synthetic Stage 2 and 3.

Using Machine Learning for Synthetic Events

Machine Learning tools represent the best way to create synthetic events because they can look at dozens if not 100s of data points to compute unique lead scores. We have used both internally developed machine learning lead scoring, as well as tools like 6sense.

Machine Learning tools are uniquely equipped to enhance the DFO process by generating and utilizing synthetic events and incorporating variable data. These tools have the capacity to analyze extensive datasets, consisting of hundreds of variables, to produce nuanced lead scores.

Generated lead scores dynamically reflect a lead’s potential based on a comprehensive view of their interactions, behavior, and external signals. This capability allows for the creation of synthetic events that are deeply informative and adaptive to changes over time “baking in” new or updated information every time they are sent.

A lead score changing (up/down) can be a synthetic event
We typically only send the up events
Below we will discuss how to categorize and send synthetic events

Advantages of Machine Learning in Synthetic Event Generation

Dynamic Lead Scoring: ML algorithms can continuously update lead scores by incorporating real-time data, ensuring that the scoring reflects the latest interactions and external signals. A lead’s score may increase due to positive engagement or new external insights, making this change a valuable synthetic event in itself.

Customization: Unlike static scoring systems, ML enables the customization of scoring models to fit specific business needs and objectives. This flexibility ensures that synthetic events are aligned with the unique markers of readiness or interest relevant to a company’s products or services.

Predictive Insights: By analyzing patterns in data, ML models can predict future customer behaviors, identifying leads with high conversion potential even before they complete traditional conversion actions. This predictive capability is critical for early and mid-stage startups seeking to maximize their engagement with leads showing early signs of interest.

Integration with ABM and Intent Data: For companies employing Account-Based Marketing (ABM) strategies or utilizing intent data from platforms like Bombora, ML models can attribute account-level signals to individual contacts, creating synthetic events that indicate a broader engagement or interest at the account level.

Implementing Machine Learning for Synthetic Events

Data Collection and Integration: Begin by integrating data from various sources, including CRM systems, marketing automation tools, and external intent data providers. This comprehensive dataset serves as the foundation for ML analysis. Data collection should encompass everything about the lead from their email address, to the time of day they filled in the form.

Model Development and Training: Develop ML models tailored to identify and score synthetic events based on the collected data. Training these models involves identifying patterns that correlate with successful conversions and engagement, allowing the model to learn over time.

Continuous Monitoring and Adjustment: Once deployed, continuously monitor the performance of ML models and update and adjust them as necessary to reflect changes in market dynamics, customer behavior, or business objectives.

Actionable Insights: Use the insights generated from synthetic events to inform marketing and sales strategies. For example, prioritizing leads with rising scores for immediate engagement or tailoring messaging to address the specific interests indicated by synthetic events.

Feedback Loop: Establish a feedback loop where the outcomes of actions taken based on synthetic events are used to further refine ML models, ensuring a cycle of continuous improvement.

By leveraging machine learning in the creation and application of synthetic events, companies can achieve a more granular and dynamic understanding of their leads, enhancing the effectiveness of their DFO strategies. This approach not only optimizes the allocation of marketing and sales resources but also contributes to a more personalized and timely engagement with potential customers.

Build vs Buy

We have done both. There isn’t a wrong choice here, as it’s really a question of time, resources, priority, and data. Tools like 6Sense, RevSure, and Market-Map from GradientWorks are three we have used. Multiple tools can be used at the same time.

These tools are looking at data from multiple companies they can help when your own sales data might not be extensive or detailed enough to provide clear insights. Third-party tools have access to a broader dataset that spans various industries and customer segments, offering a more comprehensive view of market trends, customer behaviors, and potential opportunities.

3rd Party Tools Help

The advantage of leveraging tools like 6Sense, RevSure, and Market-Map lies in their ability to aggregate and analyze data across a wide array of sources and companies. This is particularly beneficial for organizations that are either in their early stages and haven’t accumulated a significant amount of sales data, or for those exploring new markets where their existing data may not apply. These tools apply advanced analytics and machine learning algorithms to identify patterns, trends, and insights that would be difficult, if not impossible, to discern from a more limited dataset.

Building an in-house solution to collect, integrate, and analyze sales data across multiple dimensions can be a time-consuming and resource-intensive endeavor but you will have end-to-end control over it. In contrast, third-party tools are designed to deliver actionable insights quickly, allowing businesses to react and adapt to market changes more rapidly.

RevSure

RevSure takes a comprehensive approach integrating with a company’s entire go-to-market (GTM) data stack, including website analytics, CRM, marketing automation, sales automation, ad platforms, ABM software, and enrichment tools. This integration facilitates the creation of the RevSure GTM Data Graph. All of this can be used to pull data for both real and synthetic events.

RevSure also has built out tools to calculate and then send events (real and synthetic) to the Google API, including custom lead scoring, including various weighting factors (like firmographics or geography). Harry is an advisor to RevSure and has helped shape this system.

6sense

6sense offers a comprehensive ABM (Account-Based Marketing) platform powered by Revenue AI™, designed to uncover hidden demand, align sales and marketing efforts, and drive predictable revenue growth. It focuses on identifying in-market accounts through account identification, intent data, predictive analytics, and data enrichment.

All of these can be combined into generating high quality synthetic data. For example 6sense generates a synthetic signal called 6QA. It is an account level intent signal.

A 6sense Qualified Account (6QA) is an account that 6sense has identified as being ready for sales engagement. This qualification is based on increased intent, profile fit, and engagement, indicating the account’s progression from an early buying stage to a later stage, such as Decision or Purchase, making it ripe for sales activity.

When an account level signal is generated it can be used to send a synthetic event for every contact within that company, or selected contacts (i.e., a buying committee). We will address this below but to send the event to Google ads there must be a valid GCLID.

6Sense has a variety of signals that can be directly incorporated into an event value calculation. These includes signals at both the account and contact level:

GQA status
Contact intent score
Account score
Account reach
Account fitness

6sense Content Intent Grade & Numerical Scoring

Contact Grade	Contact Intent Score
A	≥86
B	70-85
C	50-69
D	≤49

6sense Content Account Buying Stages & Numerical Scoring

Account Buying Stage	Account Intent Score
Target (not-engaged)	≤19
Awareness	49-20
Consideration	69-50
Decision	85-70
Purchase	≥86

Gradient Works Market-Map

Sales teams use Gradient Words to identify and distribute books of business to sales reps. They offer a tool called Market-Map. “Market Map analyzes all the accounts in your CRM, using AI to research every single account. It then generates a map with clusters of similar accounts, with scores for each account that indicate how similar it is to a current customer.” We have used this signal for synthetic events. The February 2024 update to Market-Map now includes “enrichment attributes” for every account in your CRM.

Market-Map can now identify fit factors including custom data points. “Custom data points can be anything – roles companies are currently hiring for, whether or not they offer a free trial, how many customers they have, their integrations…” All of these factors can be used to generate/trigger synthetic and real events.

Reporting for Deep Funnel Optimization

If you have a robust reporting system in place it is unlikely you will need to make any major changes. However, it’s critical to socialize the changes that will impact the CPL (Cost Per Lead). We typically see the CPL for demand generated leads increase 2x-3x the cost… and there is no way to sugar coat this! It will scare a few folks! Our best advice is to remember the payoff:

Lower Cost Per Opportunity
Lower Customer Acquisition Costs
- Including associated marketing costs & sales costs
Faster Deal Cycle

Critical Metrics | DFO KPIs

Here are some of the KPIs we highlight when we implement deep funnel optimization.

In analyzing deal cycles by account, we focus on the complete journey from initial contact creation to the finalization of a deal. This involves several key metrics:

Days from Contact Creation to Deal Completion: This metric measures the length of time it takes for a lead that enters the marketing funnel to convert into a paying customer. Tracking this over time can help identify trends, bottlenecks, or improvements in the sales process.
Average Marketing Cost Per Deal: This includes the total marketing spend divided by the number of deals closed within a specific period. It’s crucial for understanding how much is being spent to acquire each customer.
Value/Cost of Their First Deal: This metric assesses the immediate return on investment from a new customer by comparing the value of their first deal against the cost to acquire them.
Marketing Cost ROI (Return on Ad Spend, RoAS): This is calculated by dividing the revenue generated from deals by the marketing costs associated with acquiring those deals. RoAS gives a direct insight into the profitability of marketing efforts.

By examining these metrics quarterly, we can observe patterns and shifts in deal cycles over time without getting bogged down into day-to-day changes. Breaking down the deal cycle data into quartiles—top 25%, 50%, and 75%—provides a more nuanced view. This quartile analysis helps in understanding the distribution, seasonality, and variability of deal cycles across all accounts.

Quartile Analysis For Deal Cycles

Why quartile analysis is helpful:

Identifies Performance Benchmarks: By categorizing deal cycles into quartiles, businesses can set benchmarks for performance. The top 25% quartile, for instance, highlights the fastest deal closures, serving as a model for optimizing the sales process.
Highlights Areas for Improvement: Conversely, deals falling into the bottom 25% can signal issues or delays in the sales funnel, prompting further investigation and improvement strategies.
Facilitates Customized Strategies: Understanding the distribution of deal cycles allows for more tailored marketing and sales strategies. For example, accounts that consistently fall into the top quartiles might benefit from upselling or cross-selling strategies sooner.
Quarter-over-Quarter Analysis: Comparing these metrics across quarters can reveal the effectiveness of strategic changes, seasonal trends in purchasing behavior, and the impact of external factors on sales cycles.

Fully Weighted Costs

Because DFO will positively impact all costs, it’s beneficial to track all of the fully weighted costs. While organic inbound leads may always have the biggest payoff in terms of costs/benefit by comparing demand generated leads using DFO to leads from other sources we have been able to demonstrate the unique value of each “flavor” of demand gen including:

Branded Search
Programmatic
Competitive targeting
Paid Social, etc.

Conversion Rates

You are already likely tracking the percent of conversions throughout your sales funnel (MQL to SQL, or Opp to Close). While DPO will generate fewer leads, you will observe far better conversion rates. We have seen this in both up and down markets. You’ll want to make sure you are segmenting out your funnel in meaningful ways. Typically this might be by industry segment (verticals), and/or geography. DFO will result in better conversion rates but there will be variation between them. It’s critical you have reporting in place to spot these. Let’s revisit OODA (Observe, Orient, Decide, Act).

OODA for Marketing Decision Making

During the early stages of the pandemic it became clear that some verticals like dentistry were going to slow down (as they were shut down) while other verticals would remain steady and some like healthcare would likely take-off. Fortunately we already had DFO in place.

It was a reasonable bet that healthcare verticals would accelerate but our faster deal cycles confirmed it. OODA posits that knowing what is happening “on the ground” faster than competitors means you can make changes before they even know what is happening. Because we had DFO in place well before the pandemic we were able to validate hunches around changes to the addressable market fast and effectively. We were able to quickly double down because we “knew it was working.”

Changes in the macro economic environment, as well as changes in your own and your competitors product offering will always impact sales. Knowing sooner than later gives you time to adjust your GTM effort including exploring new creative, positioning, adjusting targeting and affording you the time to develop new case studies, social proof, Etc.

Soft vs. Hard Benefits

The tangible benefits of DFO are easy to quantify and report. But, like improving your OODA, DFO will lead to a few other soft benefits. This includes being able to effectively target wider audiences with paid ads because Google and Meta will use the data from DFO to help refine the targeting. Another soft benefit is being able to leverage automated bidding, and automated campaign optimization strategies, and bidding objectives like Performance Max (PMAX). You can think of this as letting AI tools focus on the tiny details while us humans focus on the big picture inputs and the desired results (outputs).

Implementing Deep Funnel Optimization

We’ve explained what DFO is, the types of events that you can send (real & synthetic), and the importance of generating meaningful events with meaningful values, and the payoff (more deals at a lower cost).

Now it’s time to look at how to implement this. We will highlight the following options:

CRMs like Salesforce (SFDC)
Marketing Automation Platforms (MAPs), like HubSpot
RevSure
Zapier
Ontraport
Custom built

Salesforce: Can be a great tool for sending events to Google. For many organizations it’s their single point of truth.. and generally data from other systems are already rolled up into SFDC. There are also high quality integrations for Google ads and Meta. On the downside customizing SFDC for all of the weighted scoring will take some effort. I’ve seen it take many months for custom development tasks in SFDC to get prioritized and even longer to create requirements and commit developer resources.

MAP, HubSpot: The upside is there are existing integrations for Google ads and meta. This includes offline conversions for Google and conversion API (CAPI) for meta. Once restricted to only standard HubSpot lifecycle events (Subscriber, Lead, MQL, SQL, Opportunity, Customer) HubSpot can now handle some custom events but most of them are form-centric. Events can be sent with four options for the value: no value, a fixed value, predictive selling price (an enterprise feature), actual selling price. None of these are appropriate for DFO, although the predictive selling price may be beneficial in some situations, typically it’s too much value too soon.

The product team at HubSpot has been implementing new features on a continuous basis for the last several years and that is likely to continue. However, as currently engineered the system is not suitable for DFO.

RevSure: Is the best of all worlds. RevSure integrates with Salesforce, your marketing automation platform, and other tools like 6sense. They have fully implemented the Google offline conversion API and built-out lead scoring tools that can create real and synthetic events.

They have also implemented the tools needed to deprecate or repudiate fraudulent or spammy conversions which helps improve the signal to noise. Repudiation can be done manually by uploading a spreadsheet, but it’s far easier when it can happen automatically.

The primary business of RevSure is deep funnel optimization including robust attribution reporting, and determining which campaigns and ads are performing best. More importantly, for a given lead they can determine which is the “right” next campaign or ad that a specific contact should see. Their predictive tools can create synthetic events not possible with other systems while at the same time combining signals from multiple external sources like Salesforce, HubSpot, and 6sense.

Zapier: Zaps can be a great way to implement DFO. However, it’s generally going to require multiple Zaps and some custom JavaScript or Python code. Most users will be limited to the API coding that Zapier natively provides… Although it is possible to create your own custom API integration, that’s a fairly advanced task.

Zapier is also going to require field level knowledge of the data sources like Salesforce and HubSpot. When creating a Zap you’ll need to know exactly which fields hold the data that you need. Some fields are highly duplicative, for example two similarly named fields one of which returns the latest value and the other returning all prior values.

Specific to the Google ads offline API it’s critical to know the exact time of a conversion event down to the second plus the time zone. That can be tricky if you haven’t managed that previously. For synthetic events, you’ll have to create unique timestamps which also have to be managed.

If you detect a fraudulent conversion as stated above, it’s possible to repudiate that. However, to do that you must retain the exact time stamps for the conversions you wish to remove. That type of logging and data storage adds another layer of complexity to Zapier. Repudiating fraudulent conversions is not required, but it’s definitely a best practice.

Ontraport: A system that combines CRM + Marketing Automation, web forms and more. Ontraport supports a robust API and all of these can be combined to trigger and send deep funnel optimization signals for Google and Facebook. On this page is a video detailing how it works:

https://ontraport.com/university/Ontraport-for-/Marketing-performance-optimization/Deep-funnel-conversions

By combining their workflows and/or APIs you can craft a system that ranges from very easy to use to one that is highly customizable. This can work if you are using your own CRM and Map (like Salesforce and HubSpot) but this will work best if you are using Ontraport for both CRM and MAP.

Custom build: Creating your own custom code base gives you the maximum flexibility and control at the cost of managing a deployed server and maintaining your own code. A typical implementation would use Java, not JavaScript. You may still require custom programming in Salesforce and other programs, and/or the use of a data warehouse or data lake. A tool like Segment.io could be used to create the implementation but could easily cost 10k a month in fees.

I think custom code is the best approach If your team has the technical expertise, budget, server and developer infrastructure, and they understand the mission critical nature of DFO.

Our choices

Frederik and I have worked on multiple custom implementations, as well as built-out implementations in RevSure and Zapier. There are many tools out there similar to Zapier and there are other solutions that we have not used ourselves.

I think every organization is going to have to find their own “best option.” The size of the organization and budget will largely determine which way to go. With that said, the rest of this paper will focus on how we actually implement the real and synthetic events, including the value scoring that we’ve mentioned multiple times. We will discuss a few technical issues related to the various APIs like GCLID and FBCLID. We will also include a few mini case studies where we can share data.

Calculating Value Across the Buyer’s Journey

Most of this is outlined above but the process is simple:

Find high quality real and synthetic events
Calculate an unique value for each event
Send the event & value to Google or Meta

The hardest part may be identifying and selecting the real events, defining the synthetic ones and then working out the math for each of the values. Every company is going to be different, will call events different names, and it will require detailed knowledge of the full sales cycle and how contacts/leads move through the sales process.

Sales Cycle

Most teams will have a basic prospect lifecycle that starts with a lead and ends with closed won or closed lost. Of course folks will drop-off along the way especially early in the discovery process if there isn’t a fit. At some critical point a lead/contact will become a sales qualified and/or a sales accepted lead. Once “accepted” those leads can only become won or lost. Once accepted they are typically considered an Opportunity and there are usually multiple opportunity stages (Stage 1 – Stage 5). For example Stage 5 might review a contract and the next step is either “won” or “lost”.

Event Logic

No matter how your leads flow into and through the sales team not every stage is reached. Some leads will jump over a few stages going from SQL directly to Stage 2 or Stage 3 Opportunity. Or going from Stage 2 to Won or Lost. Some leads while qualified and even interested might be sidelined for timing (saved to re-engage later).

It’s also important logically to understand that some events may happen more than once (a lead may reach MQL or SQL status multiple times just as in eCommerce a shopper can add multiple times multiple times to their cart.

Rules of Thumb for Event Calculation

On the most simple level we recommend combining a lead score with a series of weights based on fitness, geographics, demographics. firmographics or other factors. Our preferred method is a ML calculated lead score, if not available a heuristically calculated lead score like HubSpots “HubSpot score” will work as well.

Where we have that ML score from an internal tool or from 6sense, etc., we also plan for a back-up score. For example 6sense may not have a score for every lead, so logically if it isn’t available we will defer a HubSpot score. For SMB targeting we are only focused on a few folks and we don’t often weight a “company or account” score, but for mid-market and enterprise we weave in account level scoring. This can be complex when you have large enterprise prospects with multiple divisions, separate smaller similarly named franchises like a hotel or car dealership.

Many systems determine company names from the domain of the contact’s email, which brings up the issue of open or public email servers like gmail, business.com, or hotmail. Leads with those public emails are often of poor quality but not always. Likewise, some folks who are seeking to bypass a content gate will use a fake email address from a real company (joe@abc.com). In these cases company scoring may be highly suspect. For example with multiple form fills over multiple years from gmail or business.com account scoring may heavily skew those leads as false positives.

Especially when using automated bidding and targeting systems we currently do not send conversion data for leads. It’s too easy to send a strong but false positive signal resulting in an influx of poor leads. Our recommendation is wait for a lead to hit MQL status and then and only then, send the lead conversion followed by the MQL conversion.

FBCLID and GCLID Explored

Google and Meta have a method to link clicks on ads to specific conversion events. They are respectively GCLIC and FBCLID (Google Click ID & Facebook Click ID). These IDs are appended to the referring URL when someone leaves Google, Facebook, or Instagram.

When sending DFO conversion events & values to Google you must include a valid GCLID. GCLID expires after 90 days. Meta currently doesn’t require a FBCLID ID, so you can send all conversion information to Meta but if you do not have a FBCLID, you need to include PII like an Email, First/Last Name, Etc. Since you can send all conversions to Meta, they are especially good at tracking view-through conversions. FBCLIDs expire after only 7 days.

Naming Conversion Events Custom or Standard

When sending DFO conversion events to Google or Meta you can either use custom events or standard events. Meta previously limited companies to 8 events and purchase events used up 4 of those. Some of those Meta restrictions (Event Manager) have been lifted. In your ads on Meta you can list a single event as the conversion goal if you are running conversion ads.

However, when sending conversion events to Meta it is a good practice to group events together. For example:

Put all marketing conversion events under “Marketing Conversions”
Sales events under “Opportunity Stage events”
Post sales events under “Customer Events”

Google doesn’t restrict the number of events that can be sent. Ads require a conversion event(s), or conversion action sets (a group of conversions). Because GCLIDs don’t expire for 90 days, there is a lot of flexibility. Depending on your ICPs it is possible you will be able to send a purchase event (when they happen within 90 days that a GCLID is valid). When you are selling to mid-market and enterprise customers and/or using an Account Based Marketing (ABM) strategy not every member of the buying committee will have converted on a Google Ad, so you can only send conversions for those who have. Because different members of the buying committee will convert over time, it is also possible that their associated GCLIDs will have expired before a purchase event.

Purchase Conversion Events

While a purchase event is the penultimate event, we need to send plenty of pre-purchase events since not every purchase will happen within 90 days that at GCLID is valid. In the case of Meta, since FBCLIDs are only valid for 7 days, it’s unlikely any purchases will happen within that time frame. However, because Meta doesn’t require a FBCLID, all purchase conversion events should be sent to Meta along with permissioned PII (Email, Phone Number, First/Last Name, Etc.).

Calculating Values for DFO Conversion Events

Setting values for events is the most critical aspect of DFO. It can be the most complex, and it will likely require some adjustments along the way. At the same time, you need ample time between any changes to give Google & Meta’s ML algorithms time to learn and react.

As noted above a key aspect of DFO is that perfection isn’t required. It’s important to send a range of data. Remember, as long as you are sending multiple events over time, each “new” event sent “bakes in” any learning (positive or negative). So while we don’t send negative events… when a lead’s score/value keeps increasing it will be noticed by Google/Meta. Leads whose score/value stays flat or decreases is also noticed.

Sending values is a bit like playing poker “all the information isn’t available.” You need to send values based on what is known at the time they are sent. For example, someone may initially convert using a gmail address to stay semi-anonymous; later you may learn they are a C-level executive at a company that fits perfectly into your ICP (Ideal Customer Profile). Over time the values sent for that lead will change as the information becomes known. That’s perfectly fine.

Conversely, someone may initially convert with an email address that is fake or fraudulent. You might assign a higher score to that lead because of their domain name. If it was truly fraudulent with Google you can remove any conversions (repudiate it). In the case of Meta the conversion remains but once the fraud is discovered no new events should be generated which will help Meta differentiate between Good/Bad leads.

Good, Bad & Fraudulent Leads

The same is true for Google. Leads that only have initial conversions are easily differentiated by the ML tools they use. However, with Google, within the 90 period conversions can be restated up or down. So while it is possible to go back and increase / decrease the values sent, this is not recommended unless there is an extreme difference. For example you think the customer is a small General Motors (GM) dealer, and it turns out the lead is for a Global GM opportunity worth millions, not thousands. However, even in that case, we feel this will largely get baked into the next events that are sent. We would only recommend restating the conversion value if the opposite is true (you thought it was a Global opportunity and it’s only for a small local dealer).

We have mentioned that conversions for leads that are outright fraudulent can be removed (repudiated). This is accomplished by restating the conversion value to 0. This is highly recommended for fraudulent leads as soon as it is learned. RevSure can manage this automatically, other systems may require this to be done manually.

However, you may be tempted to repudiate poor leads, or leads that don’t reach the opportunity stages, or opportunities that don’t close. Don’t do this, it would be harmful. Let us explain. The intention when sending DFO conversion events is to help Google/Meta ad systems to better understand which leads are valuable to you, and ultimately use that information to find better, more qualified leads. Thus while it is critical to remove conversions for fraudulent leads… all the others should remain. This is especially true for deals that are “closed lost.” Removing conversions for those leads will prevent similar companies from being targeted.

If a lead/account was good enough to reach MQL status, to be picked up by the sales team, etc. you will want more leads like that and if you removed the conversions your lead generation could suffer. If you find that too many leads aren’t closing, or aren’t reaching the opportunity stage, consider adjusting how you score them. This is one reason we don’t send a conversion event for a basic form-fill, we wait until it reaches MQL status. Ultimately, as long as you are sending enough meaningful events, at a high enough frequency Google/Meta will “learn” which are the best leads. Historically, Google wants 30+ conversions a week, and meta wants 50. One way to reach that goal is to increase your spending and the other is to send more synthetic events. We recommend both of those tactics.

Constructing a Value Matrix

For many businesses, esp. those with longer deal cycles since purchase events will be limited, we want them to overshadow all other events. We want the value of the purchase conversion events to be significantly greater than all of the prior conversion events for that contact, and larger than the values sent for leads that don’t reach “closed won” within the attribution window.

Here is an table showing for each stage of the buyer’s journey the relative value to be sent.

Align	10%
Acceptance	15%
Negotiate	25%
Finalize “purchase”	50%

Our recommendation for a Conversion Value Matrix is that it has to be in part based on your own insights. In some cases we have limited the pre-purchase value to no more than 20% of the Average Contract Value (ACV). Do recommend you think of these values as percentages of the purchase price, since they work if it’s a $100, or $100,000 product.

For higher value products there is more wiggle room. Also, later stage or mature stage businesses will have a high quantity of contacts to reach, negotiate and finalize. For those businesses (assuming a solid percentage convert to closed won), sending a larger percentage is meaningful since it is known to be valid. However, if you are an early or mid-stage business that only as a small percentage your opportunities close, it’s better to send smaller values keeping the total pre-purchase value to under 20% of the purchase price.

DFO CONVERSION VALUE FORMULAS

Let’s look at some actual formulas we have used to calculate conversion event values. We will show you a few examples: MQL status, engaged, qualification meeting set, 6QA event, and stage 2 opportunity. For these scenarios we are using data from HubSpot and 6sense. We have picked these as they are “typical events” that occur but it’s your choice which events to score and send.

Generic Calculation

(“Lead Score” x “Weight”) x .8

MQL Status Calculation

Calculate the initial score based on availability:
- If available, use “6sense Intentscore” x 0.8.
- If “6sense Intent score” is not available but “HubSpot Score” is, use “HubSpot Score” x 0.8.
- If neither is available, use a default score of 15.
If the account is in 6QA status, add 25 to the initial score.

Note: We recommend not sending a Lead conversion event, skipping it entirely or optionally sending it only after MQL status is reached. When we do send a lead conversion it would be scored similarly to the MQL event using a different weighting like .4 vs. .8. Timestamps: When MQL status is reached we would first send the “lead conversion event” with its own time stamp, and then the MQL event with its own later time stamp.

Engaged Status Calculation

Begin with the calculation of the total score using “6sense IntentScore” and “6sense ProfileScore”:
- Add “IntentScore” to “ProfileScore”.
Apply default values if necessary:
- If “IntentScore” is less than or equal to 0, default it to 25.
- If “ProfileScore” is less than or equal to 0, use “HubSpotScore” for the “ProfileScore”.
After adjusting for any defaults, sum the “IntentScore” and “ProfileScore” to get the initial total score.
If the account is in 6QA status:
- Add 50 to the initial total score.

Qualification Meeting Set (QMS) Status Calculation

Use the same calculation as Engage for QMS

Why? Not every lead achieves Engaged or QMS status, and some reach both. This keeps things simple while “rewarding” leads that “reach both” adding a bit of randomness

6QA Status Calculation

When 6QA status is reached, for every content in the account that has a non-expired GCLID send the following:

Calculate the Initial Sum:
- Start by adding “6sense IntentScore” to “6sense companyScore”.
- + 75 to this sum (since it’s 6QA status).
Apply the Fit Factor:
- Multiply the initial sum by “6sense fitFactor” to adjust for how well the company fits your ICP criteria. The “fitFactor” is determined by as follows:
  - ‘weak’: 0.75
  - ‘moderate’: 1.1
  - ‘strong’: 2
  - If not specified, use a default value of 1.
Apply the Stage Factor:
- Then, multiply the result from the previous step by “6sense stageFactor” to adjust for the stage of the company in the buying process. The “stageFactor” is determined by the current stage of the company, as follows:
  - ‘target’: 0.3
  - ‘awareness’: 0.5
  - ‘consideration’: 1
  - ‘decision’: 2
  - ‘purchase’: 2.1
  - If not specified, use a default value of 1.

Stage 2 Opportunity Set Calculation

Calculate the Initial Sum:
- Add “6sense IntentScore” to “6sense companyScore”.
- If the account is in 6QA status, add 100 to this sum.
Apply the Fit Factor:
- Multiply the initial sum by “6sense fitFactor” to adjust for the company’s fit with your ICP criteria. The “fitFactor” weights are as follows:
  - ‘weak’: 0.75,
  - ‘moderate’: 1.1,
  - ‘strong’: 2,
  - Use a default value of 1 if not otherwise specified.
Apply the Stage Factor:
- Multiply the result from the fit adjustment by “stageFactor” to account for the stage of the company in the buying process. The “stageFactor” weights are:
  - ‘target’: 0.3,
  - ‘awareness’: 0.5,
  - ‘consideration’: 1,
  - ‘decision’: 2,
  - ‘purchase’: 2.1,
  - Use a default value of 1 if not specified.

Value Calculation Notes

This structured approach ensures that each account/contact is evaluated comprehensively, factoring in both quantitative scores (IntentScore and companyScore), the qualitative aspects of fit and stage, and specific conditions like the 6QA status. This method allows for a nuanced assessment tailored to the characteristics and current status of each account. Use these as a “recipe” influenced by your own “ingredients” e.g., your own data sources.

As noted above, when you combine the values all of the events sent, and the total possible range of values calculated, we typically recommend the total value not exceed 20% of the final purchase price, but the total pre-purchase value

can run as high as 50% of the purchase price depending on your situation and the confidence in the data.

Brainstorming: Big List of Possible DFO Conversion Events

Your “list of events” will vary based on how your marketing, sales, and support teams are structured. It’s ok if you have or don’t have MQLs or if you have replaced SQLs with SALs, or you have both SQL & SALs. You can pick any real or synthetic event that you can track. That doesn’t mean 100% of your contacts/leads have to reach every event. For example, we have seen purchases where the account never reached 6QA status. Not every account signs-up for webinars, and not every webinar sign-up attends. Some accounts will attend trade-shows and some may get on site visits from their sales team. Consider including all of those as DFO conversion events even though not every account will get an on-site visit. The following is intended for ideation.

You aren’t required nor is it recommended to include all of these (although you can); what matters most is picking those conversion events that are meaningful to your company.

Big List of DFO Conversion Events:

Newsletter subscriber
Lead from a form-fill (recommended only after they reach MQL status)
MQL Status
Converted / SQL status
Pursuit stage
Engage stage
Qualification Meeting set status
Stage 1 Opportunity i.e., Discover Call
Stage 2 Opportunity i.e., Pain Point validated and acknowledged
Stage 3 Opportunity i.e., Product Demo to buying committee
Stage 4 Opportunity i.e., Price quote sent, champion identified
Stage 5 Opportunity i.e., Contract sent/negotiated
Closed Won
Product Led Growth conversion event i.e., started a free trial, clicked for info, etc.
Inbound phone call
Answered outbound call and spoke for > X minutes (real + synthetic event)
Downloaded eBook
Clicked within eBook
Registered for webinar
Attended webinar
Asked questions in webinar
Asked to speak to a sales rep
Scheduled a call with a sales rep
Phone call of more than 10 minutes completed
Scheduled call/visit completed without rescheduling
Competing a rescheduled call/visit
Buyer committee of X or more folks identified and reached (synthetic event)
Account reaches 6QA status (synthetic event)
Increase in account or lead score > X points/percentage (synthetic event)
Attended trade show
Attended hosted event like a dinner or seminar
Responded to a pre-purchase offer for SWAG or Gift Card
Positive reply to an email from an automated nurturing campaign
Frequent interaction with BDR/SDR/AE (synthetic event)
Started a limited free trial
Signed-up for a freemium product
Completed meaningful tasks in a trial or freemium product (synthetic event)
Reaching positive status within Gradient Works Market-map (synthetic event)
Reaching intent status with Influ2 (synthetic event)
Completing a self-paced training
Opened and clicked on a link in a newsletter
Opened and clicked on the last X newsletters (synthetic event)

Conclusion

In our collaborative effort, Frederik Hermann and I embarked on a mission to refine the quality of leads and enhance conversion rates, while also aiming to reduce the customer acquisition costs within the realm of demand generation. Our initial success was marked by a significant reduction, cutting the cost of closed won deals by half, a testament to our strategic approach centered on Deep Funnel Optimization (DFO). Over the span of five years, through experimenting across various industries, geographies, and company sizes, we have solidified our belief in DFO as the quintessential strategy for optimizing demand generation, showcasing its capability to drastically shorten the mid-market sales cycle from 140 days to a mere 65.

The “secret sauce” of our approach lies in the sophisticated integration of machine learning with DFO, enabling us to not only track but also to enhance the conversion pathways. By adapting the conversion reporting to platforms like Google and Meta, we shifted the focus from mere lead generation to the generation of high-quality leads that are significantly more likely to convert into paying customers.

This strategy involves tapping into Google and Meta’s use of machine learning algorithms that scrutinize and optimize for post form-fill activities, elevating the efficacy of our ad spend. The essence of our DFO methodology extends beyond the mere collection of leads; it’s about crafting a journey where each interaction is measured for quality and then sending the “quality signal” to Google & Meta.

This approach ensures that our marketing efforts are not diluted across a sea of unqualified leads but are sharply focused on those that exhibit a tangible interest in progressing through the sales funnel. We are not just capturing leads; we are enhancing the ROI from our marketing endeavors and securing a competitive advantage in the dynamic highly contested market landscape.

Cases

Performance marketers can significantly benefit from embracing machine learning (ML) optimization, as evidenced by Birdeye’s experience with reduced customer acquisition costs and improved close rates by 3x to 4x for their less performing campaigns.
Machine learning, when applied to paid digital lead generation, which is inherently complex, can yield astounding results. Birdeye witnessed improvements across both their best and worst campaigns through the utilization of offline conversions (as termed by Google) and deep-funnel conversions (as termed by Facebook).
A specific example given is a retargeting campaign run by Birdeye from July 1, 2019, to December 22, 2019. Upon switching to deep funnel optimization on December 23, significant improvements were observed: lead quality increased by 20%, cost per lead (CPL) nearly doubled, yet the customer acquisition cost (CAC) was reduced by 3x.
Further testing at 56 and 68 days demonstrated continuous improvement, with lead generation increasing, and CAC significantly lowering, despite varying degrees of lead quality and opportunity cost (CPO).

Frederik.