Facebook Advertising Decoded in 15 Minutes

Principles I learned from over $1 billion of Instagram and Facebook campaign investment

James Chadwick
Jan 18 · 18 min read
Keeping the network working. Facebookers in the Odense Data Center in Denmark.

Over seven million businesses will invest over $80 billion across Facebook’s apps and feeds in 2020, and most of them will not do it well.

FB is the world’s most powerful marketing platform, but few make the effort to understand how it really works.

And that’s no surprise — it’s complex, disruptive, and it changes weekly.

I spent seven years working at FB between 2012 to 2019, first helping companies invest over $1 billion on our ad platforms, and then using FB myself as a marketer.

I learned some fundamental principles that can help decode how it works today, and also signal where the platforms are likely to head in the future.

To try to sum up this code in a line:

Use Facebook the way it was intended to be used, which means bringing real value to users with fast, fresh, and native content, and running signal-rich, liquid campaigns that allow the new Machine Learning capabilities to work their magic.

To caveat, these principles are written with larger advertisers in mind (e.g. $5m+ annual spend), but most will be applicable to smaller advertisers also.

To learn the principles is easy, but to have the organizational patience to follow and test them is often far more challenging.

For those that do, the rewards are worth it.

1. FB Is a Marketplace for Value

FB has built a thriving ‘attention’ market, where disruptor businesses of any size can compete fairly for over 2 billion potential customers.

All markets thrive on trust and liquidity, so FB tries to increase the ‘supply’ of attention, set clear rules and guidance, make data transparent, and remove bad actors.

The 1% of participants who study the market and act with discipline earn an immediate advantage over the 99% that don’t. Caveat emptor.

In most other auctions the highest bid wins, but FB also tries to balance both end-user experience and advertiser value.

More precisely, the auction considers 1) Bid price and type 2) Expected response, relative to the campaign objective 3) Relevance or value of the ad to the target audience.

This means strong creative and landing-pages, relevance and targeting should all drive down auction costs.

Also, FB selected a Vickrey–Clarke–Groves (VCG) auction type, which gives bidders an incentive to bid their true valuations.

2. Creative Systems Generate Value

Research by Nielsen and others suggests that creative impact explains up to 50% of ad results variance, and yet leaders rarely give it even 5% of their attention.

Facebook’s Creative Shop, 2019

Bay Area founders, in particular, tend to prefer reading code and charts to folders of creative ads.

Mobile creative is a ‘hits’ business like movies and music. For every 10 ads you test, expect one high-performance peach and 9 lemons.

Pluck and polish the peaches, and ditch the lemons.

It’s never been easier to build effective, mobile-first creative because you can literally use your iPhone to make ads, and we now know what works — static+video ads together, bold vertical formats, highlight the brand early, showcase the product, etc.

Here’s a great synthesis of many years of testing by Facebook’s Creative Shop team:

Creative Considerations for Direct Response Campaigns” by Facebook Business, Aug 2018

What else is important?

First, the more the ad ‘fits’ the screen and app natively the better it performs, so vertical video beats square or landscape.

Often it’s the raw, authentic content that beats out the polished, expensive ads. And the more seamless the creative handover from ad to landing page, the better.

Creative fatigue is a real issue, so a variety of lengths, sequencing and dynamic ad creation tools are all useful to stay fresh.

To make things even easier, you can now study all your competitors’ ads in the Ad Library, linked to their FB page, so new category entrants can benefit on Day 1 from all the optimization and work done by market leaders.

Through transparency, FB wants all marketers and ads to improve, to create more value for users and companies alike.

Just dive into the FB Ads Library and start decoding the ads that the category disruptors and leaders are testing in your region.

Here are the kinds of examples you’ll find:

One of my personal favorite accounts to track is Netflix, a company that consistently blends content and data to drive growth and brand love.

Speed to create and test is a competitive asset, because even an average chess player will beat a grandmaster if they make 2 moves for every 1 their opponent makes.

But craft is still essential. It takes deep expertise to translate brand insights into a modular “ABC” creative testing framework, rapidly generate high volumes of prototype ads using tools and shortcuts, and then focus deep production experience on the top-performing assets

To sum up, creative systems can create real business value, if they are fast, modular and native.

3. Start With Business Outcomes

Investing on FB before you commit to a real-world business outcome can be an expensive decision. The market punishes confused campaign setups, where the campaign objective, audience, creative, CTA, and landing-page experience are not aligned.

Everything in FB optimizes to the campaign objective, so start by translating your real-world business outcome into the most appropriate campaign objective, and build out logically from there, all the way through to measurement.

Never do it the other way around, force-fitting an existing ad into an objective.

There are now 14 different objectives to choose from and from this critical decision, FB will select everybody who sees it, based on their past responsiveness to ads with similar intent.

4. Work With Granular LTVs and CPA Truth

The auction rewards advertisers who can accurately estimate their Life Time Values (LTVs) and set Return-on-Adspend (ROAS) targets and bids based on true valuations.

High LTVs allow more freedom to test a range of ad placements, types, and retargeting loops up and down the funnel, and longer conversion windows, but exaggerated LTVs can be fatal.

Too many advertisers still fixate on generic benchmarks (e.g LTV: CAC ratio of 3) when they would be better focused on more granular LTV variability of different products, channels, and customer segments. Can you identify your highest LTV prospects, and customize loops to convert them, to hit your ROAS targets?

A common mistake is to bring unrealistic Cost-per-Acquisition (CPA) targets from search or affiliate campaigns and apply them to FB’s scaled, discovery platform. This often demoralizes teams, chasing impossible targets, while competitors are building a more realistic, robust machine.

I’ve even met contrarian growth marketers who take a very different approach to CPA targets — they deliberately use the FB marketplace to find and benchmark their true current value to customers, and then build from there.

5. Structure Creates Liquidity

Robust markets thrive on liquidity so that every media dollar can flow to the most valuable impression.

If you’ve given FB enough signal to find your best prospects with the right message, it’s important to create sufficient liquidity to let the Machine Learning do its job.

Consider increasing 4 types of liquidity — placement, audience, budget and creative.

Placement liquidity can be increased by opting into as many surfaces as possible, ranging from FB Newsfeed to Instagram stories, Messenger and even off FB and into the Audience Network.

Audience liquidity can be increased by targeting the broadest possible audience, including multiple countries.

Budget liquidity can be increased by removing restrictions on where the campaign budget can be spent.

Creative liquidity can be increased by allowing the automated system to test and choose the best-performing creative.

6. Machine Learning rules.

With classical programming and systems, you input rules and data, and you get answers. With ML you input data and answers, and it learns the rules.

FB is betting on ML to learn rules faster than humans and has built an integrated set of tools, including Campaign Budget Optimization (CBO), which will help level the playing field for less advanced advertisers.

ML uses signal on and off-FB, serial algorithms, and predictive analytics to find the right audience at the right time at the best price.

For example, you can start testing ML by using the so-called ‘Power 5’ suite of solutions:

  1. Account Simplification
  2. Advanced Matching
  3. CBO
  4. Automatic Placements
  5. Dynamic Ads.

CBO is essential to ML because it allows the budget to be set at the campaign level instead of at the ad set level, and automatically distributes budget to top-performing ad sets in real-time.

By running separate campaigns instead of CBO, you essentially put guard rails in place that restrict the ad delivery system’s opportunity to run ads in the most effective placement. This reduces the collection of signal data, and can drive up CPMs, lower reach and increase the cost per sale.

CBO currently works best with larger budgets, in categories where FB already has strong signal data to work with, but it will fast become the default. Advertisers who crack the ML code for their brands will gain an advantage.

It requires buyers to let go of their egos, and embrace a more patient and agnostic approach to placement, formats, creative and even audience. FB’s native Ads Manager tools are adequate for this approach and continue to improve.

In the future, advertisers will require smaller but more expert and vigilant teams to leverage the new ML, and aggressive ‘day-trader’ media buyers could even have a negative impact if they don’t give the ML time to settle and kick in.

The new ML era plays to the advantage of experienced, strategic marketers, as differentiation will be created more by the crafts of branding, messaging and creative breakthrough.

7. Bring Your Own Signal

The more signal you give, the better results you get. It’s like a Bring-Your-Own-Data party — FB provides some classic snacks, but you should also bring your own delicacies.

A signal is just data that an ML model can use to make predictions about future behavior. For example, if consumers add items to their carts, this may indicate they are likely to purchase in the future.

Signals can come from a variety of sources such as pixels (and the event-triggered rules you set for them), software development kit (SDK) data that tracks in-app behavior or offline events like in-store sales.

What happens off FB doesn’t have to stay off FB, so disruptors who study their consumers’ real-world journeys and find new signals to ingest into their campaigns will earn an edge.

FB has developed privacy-safe ways to protect sensitive data when companies onboard their data, and custom systems to ensure competitors don’t benefit from proprietary signal.

Signal management has become simpler, with native tools like manual and automatic Advanced Matching, which allow advertisers to report on more conversions, optimize ads against more conversion data, and reach more people on Facebook with their website custom audiences or dynamic ads.

8. Discovery Drives Growth

Nobody comes to FB or Instagram with the intention to buy something, but with the smartphone, we have all become partial and continuous shoppers.

On social media, users are often in discovery mode, open to new ideas, so for disruptive DTC brands especially, the apps offer opportunities for demand creation.

For example, on Instagram beautiful product and user images can be pulled directly from your catalog and dynamically targeted to the most relevant audiences through the Collection format. You can showcase your brand and multiple products in feed and customers can browse your products in a fullscreen Instant Experience called ‘Instant Storefront’.

As the ML prediction engine improves, the best DTC marketers will master the art of creating moments of serendipitous brand discovery. Partial shoppers come for the product, but they stay for the brand.

Businesses ‘born in search’ often struggle to break out of their intent mindset and seize the scaled acquisition potential of FB and IG.

9. Track a Logical Funnel of Events

Be wary of gurus declaring the death of the funnel — it remains the most logical language for founders, investors, growth and sales teams to debate and allocate budgets.

For example, by getting consensus and sustained focus on specific and measurable events up and down the so-called ‘pirate funnel’ (“AAARRR” = Awareness, Acquisition, Activation, Retention, Referral, Revenue) you will already be ahead of most competitors, and can sharpen your campaigns around outcomes and signal collection.

Most importantly, teams must be disciplined to only track events that 1) have business impact and 2) logically precede each other. If a user can skip a funnel stage, then that stage might not belong in your core funnel, or should be combined with another action, to retain the sequential logic.

Top/Mid/Lower funnel campaigns can be very different beasts with different expectations (e.g. ROAS targets breakeven/5X/8X) but keeping a strong, consistent brand thread down the funnel spine will help memorability and benefit every dollar.

Experiment up and down the funnel with the 3 very different flavors of creative — short, interactive, long-form.

There are no fixed rules, only hypotheses to test for your brand, so beware the gurus who tell you otherwise.

10. Irrelevance Is Expensive

As in life, on FB it’s expensive to be irrelevant. If your message is off, people will ignore you, and the auction will punish you, so use the revamped Relevance Diagnostics tool to understand this.

The tool helps you understand the wants and needs of your target audience instead of merely showing the success or failure of your ads. The diagnostics break down into 1) Quality 2) Engagement and 3) Conversion scores, and many factors load into each of these.

For example, a slow-loading or incomplete landing page or an inaccurate portrayal of your brand can lower your score.

Interestingly, it’s more impactful to move a ranking (e.g. your quality score) from low to average than it is to move a ranking from average to above average, so focus on improving low rankings rather than on improving average rankings.

Also, avoid baiting at all costs — clickbait, votebait, dealbait — or the algorithm police will hunt you down like a cornered rat, for crimes against relevance!

11. Targeting Is Like Salt — Too Much Ruins the Dish

Here’s the paradox — FB built the world’s most powerful targeting machine, created a $70B ad market around it, and now want you to use it less.

The targeting sweet-shop is still open of course — you can import your own datasets using Custom Audiences, use high-value users as seeds to generate scaled Lookalike Audiences, create fresh, dynamic remarketing target pools using pixels, and of course target by demographics, interests, behavior, jobs, city, etc. And even overlay all of the above.

But don’t overdo it in the sweet shop. More targeting means less liquidity, lower reach and often more creative fatigue, all poor conditions for ML to thrive.

Narrowing reach significantly increases the burden of effectiveness on the campaign. Keep asking yourself what is the minimum amount of personalization for the maximum impact.

Each advertiser will need to test and learn for herself. Smaller advertisers will still rely on targeting, and often the sharper the better. For instance, if you’re a new Direct-to-Consumer (DTC) custom tennis racket brand, forget fans of Roger Federer, target the fans of the players ranked #20 to #100. They will be the hardcore tennis superfans.

And of course, never stop testing classic, hard-working advanced Lookalike strategies, such as people who visited your website 3 times or more.

12. Think Global but Test Local

FB often knows where users are from, where they currently live, and even where they are right now.

This ability for advertisers to target at the city, town or even village level, has made it an essential tool for the long-tail of small businesses, competing with local classified ads and listings. Location targeting allows advertisers to target by Zip Code, DMAs, or even within 1-mile radius of a pin dropped on a map, although it comes at a cost and obviously throttles reach.

This allows a woodcarver in Indonesia to find global audiences, and global brands like KFC to act local, and drive footfall to individual stores and measure ROAS through POS data.

Creatively, some new adtech automation can dynamically switch local city names or images in ads, to increase relevance and results.

FB also gives you the option to scale out city-by-city as a strategy, so you can saturate a city for a short period, giving the impression you’re a trusted company.

Finally, some brands have used city targeting to run simple test frameworks to get a clean read on platform performance: Instagram-only in Seattle, Facebook-only in San Diego, test and scale.

13. Messenger Allows Conversational Commerce

It’s still early days, but businesses can now use ads to start talking directly with customers. More than 20 billion messages are exchanged between people and businesses every month, and 1.3 billion people use FB Messenger.

If you want to sell to younger customers, messaging beats email and phone. Higher funnel campaigns can use click-to-Messenger to reach audiences at scale and start individual conversations. At the bottom of the funnel, sponsored messages allow businesses to send highly targeted promos directly to people who have already talked to them.

Conversational commerce is simpler for consumers and is a step closer to the attention they might get from a sales associate in the store. Reducing the steps required, and the number of information sources consumers need to turn to, shortens the distance between buyer and purchase.

Thousands of companies are running tests to collapse their funnels using Messenger. For example, an insurance company recently found that 12% more people completed the quote questionnaire on Messenger compared to their website’s quote tool, reducing cost per leads by 23%.

Broadly, FB has used Messenger (1.3B users) to test new conversational commerce business models, and when things get traction, Whatsapp will likely follow (1.5B+ users).

To drive frictionless commerce, FB will increasingly integrate the checkout experience into all of its apps. For example, it’s testing a new native checkout feature for dynamic ads, making it possible for users to discover a product via a dynamic ad and purchase the product from the ad without ever leaving the Facebook app.

14. FB Was Made to Measure

The biggest myth still peddled about FB is that you can’t measure investment accurately.

In reality, the platform was built ground up with measurement in mind, from basic tools to measure the true value of a single channel, through to advanced models tackling LTV, Custom MTA and cross-channel.

For advanced marketers, one way to think about modeling the true picture of FB performance relative to other channels is by triangulating 3 different ‘ATM’ models: Attribution + Testing + Mix-Market Models (cross-channel).

Attribution is bottoms up, MMM is top-down, and testing/experimentation is the most pure and precise. All 3 approaches are valid and independent when optimizing spend across channels, so triangulation will offer a valuable output.

At the very least, FB marketers should keep a ‘test log’, like the science experiment book you kept at school, tracking everything they tried. The tests that don’t work can be just as valuable as the ones that do.

Two important tools are Split tests (to optimize) and Lift tests (to justify). Like an A/B test, split testing allows you to create multiple ad sets and test them against each other to see which strategies produce the best results. Split testing is good for comparing performance to decide between two strategies. Lift testing, on the other hand, shows the impact of an ad or campaign on business goals, compared to a world in which the campaign was not run.

Attribution is perhaps the most controversial measurement issue teams will face. The number of days between when someone viewed or clicked your ad and then took action on your website is referred to as an “attribution window.” The default Facebook attribution window settings show actions taken within 1 day of viewing your ad and within 28 days of clicking your ad, but you can change the settings for both the view and click windows to 1, 7, or 28 days.

This decision can make or break your ROAS, which will continue to increase throughout the entire 28-day window as purchases continue to happen.

You should ask yourself what type of products you have and how long the consideration process is for prospects, and if you’re not sure which conversion window to choose, use the one that’s most similar to the one you’d use for conversion optimization.

15. Plan for All Seasons and Expect Storms

The auction cost of FB advertising is both seasonal and volatile.

Interestingly, the costs tend to rise throughout the year for impressions, reach and clicks, as the overall volume of ad spend increases towards the holiday season. However, for DR outcomes like leads and conversions the prices go down as the year progresses, reflecting the greater volume of transactions.

Savvy marketers use this knowledge to build audiences when Top and Mid-funnel costs are cheaper earlier in the year, in order to convert them into revenue during holiday season when Bottom-funnel costs come down.

The US eCommerce calendar creates a drumbeat throughout the year — January sales, Valentine’s day, Tax Break April, Memorial Day, Labor Day, Halloween, Black Friday/Cyber Monday, Christmas and New Year. In the hot summer ‘beach mode’ months, mobile feed spikes and desktop dips.

Additionally, costs for top-funnel brand campaigns rise at the end of each quarter, when large advertisers rush to spend out brand budgets when operating in a ‘use it or lose it’ financial culture.

Besides this seasonality, you should plan for several ‘stormy’ periods in any given year, which you will need to ride out with patience. This systemic volatility might be caused by a bug, a privacy breach, an algorithm update, or (as rumored) an external factor such as a Safari privacy update. We rarely get an explicit explanation.

Always expect storms, diversify channels and buffer budgets. You only find out who is swimming naked when the tide goes out, as the Buffett saying goes.

A lesser-known type of volatility is A/A variance, or in other words, the persistent ‘background’ randomness on the platform. All of us, including FB, would prefer more predictability, but marketers should expect and tolerate persistent randomness. The exact same campaign that worked well today, might not work tomorrow.

As an aside, I recently completed a 10-day silent meditation retreat, where for 11 hours a day I learned to let things go, and that all bad things must come to an end, and I suspect this has prepared me to be a better FB marketer in the ML era! (I plan to write about those learnings about life separately later this year.)

My advice is to ensure everybody you work with understands that on FB A/A variance is a thing, and regularly quote Stoic philosophy to each other:

“There is never any need to get worked up or to trouble your soul about things you can’t control. These things are not asking to be judged by you. Leave them alone.” — Marcus Aurelius

In Summary…It’s All Just a Fishing Trip

I’ve shared many principles here, and I hope the level of detail has not put anybody off trying to master Facebook marketing.

To sum up

Use Facebook the way it was intended to be used, which means bringing real value to users with fast, fresh, and native content, and running signal-rich, liquid campaigns that allow the new Machine Learning capabilities to work their magic.

If you prefer metaphors, I like to compare FB advertising to a deep-sea fishing trip.

You are the Captain of your ‘boat’ of resources (budget, team). You must navigate the vast, seasonal, sometimes stormy oceans of FB.

You can set up and cast many rods (campaigns), testing different hooks (ad formats and placements) and bait (offers, deals, ABC creative) to catch different types of fish (customer segments).

But if you set up too many rods you will start missing the bites and losing opportunities.

So aim to set up and watch carefully your fewer, better rods and hooks, and test much tastier bait. And hopefully, you’ll start catching those Big Kahunas!

Happy fishing!

Advice & case studies

Thanks to Niklas Göke

James Chadwick

Written by

Father to 4 boys. 7 years@Facebook. Now building and writing again. I write because I don’t know what I think until I read what I say.

Better Marketing

Advice & case studies

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade