The Importance of Lead Scoring for Automated Bidding Strategies

When Google launched their automated bidding strategies years ago, they were terrible. We tested them frequently, and any competent managers on Manual CPC could handily outperform the automated bidding models. Every so often, we would test them again. It wouldn’t work, so we’d give it some time and then test again. Over the next few years, the gap started to close. And over the past year or two, these automated bidding models have become much more viable strategies.

The core principle of automated bidding strategies is that you’re allowing Google (or another platform – bid management software far predates most of Google’s automated tools) to optimize for a certain outcome on your behalf. If we start to accept the idea that they can accomplish that, it’s vital we turn our attention from granular bid management to the data we’re providing on the outcomes themselves. It’s imperative we give Google the best data possible so they can optimize campaigns to maximize the value it drives.

This post is dedicated to the importance of providing additional data for companies focused on generating leads with Google Ads campaigns (or any acquisition channel for that matter).

Lead Scoring & Value Estimation

Measuring value on an e-commerce site is relatively straightforward. If a user spends $100 on the website, you can read in $100 worth of value to your Google Ads campaigns. That functionality has been there a long time, and it’s easy to use. Value for e-commerce companies is certainly more nuanced than that, considering ideas like different product-level profit margins, attribution modeling, cross-device conversions, and lifetime value. But on a basic level, feeding in revenue is a simple and useful way to understand the value ads are generating.

For companies that rely on generating leads, it’s not so simple.

Anyone who fields leads with a “comment” section knows exactly what I’m talking about. The value of a lead varies substantially. Let’s take LDM’s situation as an example: a digital marketing agency offering paid search management services. Which of these leads do you think I’m going to get more excited about:

Lead #1: “We’re a B2B company with a $50k monthly Google Ads budget interested in finding a new partner to manage our campaigns. Can we set up a time to discuss your approach and review relevant case studies? Thanks!”

or

Lead #2: “I’m in my google adwords account and i cant find average position. it was there before but i cant find it can you help?”

I’m a nice guy so I’ll send Lead #2 a link to the announcement about Google Ads removing average position, and maybe a link to a recommendation on how to deal with it. And who knows, maybe Lead #2 will appreciate the help and become a client at some point. Probably not though. Lead #1 clearly seems to have much more potential.

Enter Lead Scoring. Not a new concept. All it means is that you’re assigning a relative numerical score to your leads to measure their potential. Rank things 0-5, 0-10, 0-100, estimate the lifetime value of the lead… score it however you want to most accurately estimate the value of that lead. That score, or value estimation, should become an integral part of your optimization algorithm.

As a marketer or salesperson, you probably have been doing something like this anyways. Most major CRMs (e.g. Salesforce, Hubspot) have some kind of lead scoring functionality, although whether it reports them back to Google Ads seems to be more miss than hit for most small and many medium-sized businesses. That’s something we want to change, whether it means integrating the CRMs or manually uploading lead scores to your ad interfaces.

If we’re relying on algorithmic optimization more heavily, we have less ability to intuitively adjust for lead quality and must force the automated models to make those decisions for us. That’s why adequate scoring is important now more than ever. Automated models have a chance to optimize more effectively as long as we give them the right data.

Bidding Models

For the purpose of this post, I’m separating different bidding models into three different groups:

• Click-Based Models
• Non Value-Based Conversion Models
• Value-Based (Lead Scored) Conversion Models

An analogy for comparing these three takes me back to a fairy tale you’re probably familiar with: The Three Little Pigs.

Using a click-based optimization model is like building your house out of straw. If there are wolves around, you’ll get eaten alive.

Using a non value-based conversion model is better. Especially with good craftsmanship, a house of sticks can keep you pretty competitive. But up against a formidable opponent, its weaknesses can be exploited and it too will fail.

Lastly, we have the value-based conversion models. Built properly, you can relax on your balcony and watch your competitors get eaten alive.

The Problem With Click Models

Click-based models include Manual CPC and Maximize Clicks. The “straw house” analogy probably isn’t entirely fair, because managers who use Manual CPC (including us frequently) are likely optimizing for leads and value manually, so really only consider the analogy apt if they’re using these models without any kind of conversion or value tracking.

The core problem with click models is they make no effort to optimize based on the value of those clicks. It’s simply about traffic, which I regard as nearly meaningless. You shouldn’t be putting much stock in the number of clicks your campaign is generating. It’s a performance indicator to watch, but it isn’t a KPI.

“Maximize” models are scary unless the calculation includes some kind of efficiency correction. Maximizing Clicks (or Value for that matter) does so based on a target total cost, so any Maximize models are susceptible to dramatic overspends. It’s most dangerous when you’re targeting a niche audience and have a maximum daily budget set significantly higher than what you typically spend. In this situation, Google tries as hard as it can to spend too much money on a limited inventory. If you’re familiar with supply and demand, you can guess which direction your CPCs are going to go (hint: it’s way up).

This is an actual account we audited for a client who switched from a Manual CPC strategy to Maximize Clicks.

Results when switching to a “Maximize Clicks” model

Comparing a 2-month period before and after the switch to Maximize Clicks, Average CPC rose 103%. Cost/conversion increased 104%. Clicks were down 1%, and conversions were down 2%. Obviously this is an extreme example, playing out like this due to the limited keyword inventory relative to the available campaign budget. You’d think whoever was in charge of the account would notice the direction of this trend and make corrections, but that didn’t happen here.

Trying to optimize an account based purely on CPC without tracking leads or value is a very bad idea. Using a “Maximize” model adds to the danger.

Value-Based vs. Non Value-Based Conversion Models

If you’re running an automated bidding model like Target CPA (target cost per acquisition) or Maximize Conversions, you’re telling Google’s (or Facebook’s, or Bing’s, etc) algorithm that all leads are worth the same, even though that often isn’t the case (see my example above). Over time, this could cause you to lose ground to your competition.

Here’s how: Let’s say you and I are both using automated bidding models, but I’m using a lead scoring-based model and you’re just using an automated conversion-focused strategy without values. Let’s also say half the leads are like #1 above (high quality) and half are more like #2 (low quality). To start, we don’t have much data, so Google will be distributing them relatively evenly to both of us.

I’m scoring the leads on a 1-5 scale, let’s say 5’s for the lead #1 bucket and 1’s for the lead #2 bucket. You’re not scoring them, which is equivalent to just saying they’re all 3’s.

My automated model is trying to maximize my scored value, and since I’m telling them I’d rather have the 4-5’s than the 1-2’s, they’re going to give them to me as long as it’s possible without bidding irresponsibly higher. Since all you want are leads, there’s nothing stopping Google from just gradually trying to give you all the leads that are 1-2’s and me all the leads that are 4-5’s. As the optimization model gets more data and more accurately can predict whether a user is a 1 or a 5, it’s going to funnel all the 5’s my way and give you all the 1’s, because they’re all 3’s as far as you’re concerned. And that will become a problem for you – I’m going to get all the high-quality leads while you’re left with the scraps.

Bidding models that account for value should gradually gain value against those that don’t

The graph above theoretically holds true when comparing any two bidding models where one uses a lead valuation metric while the other doesn’t. The model not using lead scoring will think the low-quality leads are under-priced and the high-quality leads are overpriced, thus shifting more and more budget to those less efficient leads over time. Failure to score leads (at all, or as effectively as competitors) leads to the systematic overvaluing of low-quality leads and the systematic under-valuing of high-quality leads.

More accurately, the value-based bidding model will win on a “lead score per cost” basis. It isn’t necessarily focused on just sending the best leads. It’s focused on sending the most lead value relative to their costs.

What If Everyone Uses Automated Lead Score Strategies?

In a world where the entire competitive landscape has lead scoring in place, has the same lead values, and is using similar automated bidding methods, the winner will be the company that can most accurately represent the value of the leads to their company. That’s also assuming identical ad effectiveness and quality scores, which is a nonsense assumption, but regardless it underscores the importance of properly valuing leads if all else were equal. Accurate value estimation can give you a real competitive advantage.

The entire idea of Google trying to put everyone on an automated bidding models is an odd one. It’s an obvious and direct conflict of interest, as they’re more interested in increasing their profitability than they are your efficiency. Ideally, they’d like to do both, but when they can’t whose side do you think they’re going to be on?

Google Reps actually recommend you start target CPA models higher and then lower them as you collect data to force your CPA down. Excuse me, force it down? What they’re essentially saying is that it’s giving you a $60 CPA because that’s what you ask for even though they could be giving you a lower one, and you need to lower your target CPA if you want it lower. Or in other words, they’re going to waste your spend as they see fit to drive up their revenue as long as they can hit your target CPA. And we’ve seen it in action! If we increase the target CPA, cost/conversion simply increases without changing any other variables. But that’s a rant/conspiracy theory for another day.

The theory behind usefully implementing automated models for multiple companies at the same time is that each company is different and is able to derive different values from that traffic. There are probably enough ifs and assumptions in the first paragraph of this section to make it work for Google (they’re off the hook if companies are using value-based tracking). Multiple companies being able to run effective automated value-based bidding strategies and the marketplace still working relies on two things:

1. Customers being worth different amounts to different companies
2. Google’s data being able to predict those value differences

How might that be doable?

Specialization. If Company A specializes in paid search marketing for dentists, they might be able to better monetize a paid search lead from a dentist better than Company B, a general paid search marketing agency. Company A has an efficient system in place and cross-sell other services, so that lead may be worth $500 to them while it’s only worth $300 to Company B. That satisfies the first condition. To satisfy the second, Google has to know that the person searching is a dentist. Which is reasonable. So, if we’re both scoring based on customer value properly, there’s nothing wrong with us both running automated bidding models because the value of the same lead is inherently different.

Convenience. That value difference isn’t just limited to the data we’re providing. Google uses proprietary data to make these optimizations too. It’s all very secretive. But let’s say in their mounds and mounds of data that they’ve determined there’s a 70% chance someone searching for a seafood restaurant is willing to drive 15 minutes, but only a 35% chance they’ll drive 40 minutes. All else equal, that click is worth half as much to the restaurant that is 40 minutes away. And Google must use data like this; there’s a reason it features close restaurant listing when you search for one (people are happier with results if they’re closer).

Personalization. It your dating profile says you’re violently allergic to cats, their matching algorithm probably won’t set you up with someone whose profile constantly talks about his or her cats. And Google probably won’t recommend cat cafes to you. Similarly, if you’re vegan, Google knows (everyone knows, Karen, you can stop mentioning it) and probably won’t advertise steakhouses to you. “Personalization” is admittedly broad, and could even include the first two cases here. It’s all Personalization, arguably.

Those are just 3 quick examples. It’s data points like these, and who knows how many others, that Google must use to differentiate the automated settings to deliver unique results for advertisers. Done efficiently, it’s a process to properly quantify competitive advantages across all the companies participating in the market.

As a marketer, you shouldn’t actually need to identify these inefficiencies or competitive advantages. All you need to do is score the data properly and then put your trust in the algorithms. The data should do it for you. Combining Google’s use of Big Data with sufficient quantities of scored leads, the algorithm could discover Personalization anomalies and competitive advantages without us even being able to identify them, or even understand why.

Setting Up Lead Scoring

Effectively using lead scoring to optimize acquisition campaigns requires a collaboration between your sales and marketing teams. Sales software like Salesforce, Hubspot, and Pipedrive have this functionality built in, and you can easily import lead values to Google Ads. If you operate a smaller business that doesn’t utilize a CRM, or has a CRM without this data, the process will need to be a bit more manual. How to accomplish this is a different discussion, suffice to say you can capture gclid parameters from Google Ads and then use Offline Imports (more info here) to import lead scores/values.

Doing that will require some technical knowledge and support, but it’s relatively simple once you have it set up. Keep your lead data and scores in a Google Sheet, and it can be automatically imported to Google Ads regularly.

Need help setting up an effective lead scoring system for your business? Learn more about our web analytics services or contact us today!