Running Ads

How to run ads on social networks

This page teaches you the strategy behind running high-performing social ads. These lessons apply to Facebook, Instagram, Snapchat, Twitter, LinkedIn, and more.

We'll be using Facebook & Instagram for the examples in this guide.

We'll cover:

• Targeting audiences
• Assessing metrics
• Optimizing ads
• Retargeting

Why Facebook/Instagram ads?

Facebook and Instagram ads perform remarkably well because:

• They command attention — Ads are injected into the Facebook Newsfeed. They're impossible to ignore—especially when they're video-based.
• Rich targeting — Facebook's targeting breadth and specificity is unparalleled. You have a good chance at narrowing into the audiences most likely to buy your product.
• High volume — Facebook is the world's largest social ad network, and audience volume is important for the longevity of a campaign.
• The potential to work for every business — I've never had a client who couldn't at least find an audience on Facebook. (Perhaps not an affordable audience, but a viable audience nonetheless—if they were willing to pay.)
  

The reason Facebook Ads work for all businesses, including enterprise, is simple: People who work at businesses are humans. Humans use Facebook.

It doesn’t matter if you advertise to a human on a work-related (e.g. LinkedIn) or non-work-related website (e.g. Facebook). A lack of business context has never been a deal-breaker for ad conversion.

Before we start

Three housekeeping items before we start:

• Before you create your first Facebook campaign, you’ll need to make a Facebook Business page. It’s free.
• You'll then need to create a campaign from your Business page. When you do so, Facebook asks for the campaign's marketing objective. Choose Conversions (or App Installs — if you’re advertising for mobile).

Finally, here's our acronym cheat sheet:

• CPM — Cost-per-thousand. How much you pay per 1,000 ad impressions.
• CPC — Cost-per-click. How much you pay per ad click.
• CTR — Click-through-rate. The percentage of people who click your ad after seeing it.
• CPA — Cost-per-action. The average amount spent to reach a conversion goal (sign up, demo, free trial, purchase, etc).
• LTV — Lifetime value. How much revenue the average user makes you.
• SaaS — Software-as-a-service. A company whose product exists in the cloud. Often monetize with recurring subscriptions.

Lastly, this stuff is pretty complicated. And doing it wrong, okay, and exceptionally well can make huge differences in the profitability.

We've driven over $1B in revenue from ads for clients, we've created the ad agency for startups, Ad Labs.

Ad campaign structure

Here's the hierarchy of an ad campaign:

Campaign → Ad set → Ad

Keep this hierarchy in mind throughout the coming sections.

Steps to setting up a campaign

1) Set up the pixel

Working with an ad channel begins by setting up the channel’s conversion pixel (linked is Facebook’s). This is the JavaScript code that reports conversions on your site back to the channel.

This is how channels know which ads are performing best—since cost-per-click is just an intermediary metric that isn't good enough to base ad spend on.

2) Determine budget

Set budgets for each ad set high enough that ads can individually reach ~3,000-5,000 impressions. You need such a significant sample size for your cost-per-click (CPC) to stabilize.

CPC is how you determine in the short term which ads perform best.

As you see consistent CPC or CPA numbers that you can afford, incrementally increase your budget over time until the CPAs drop significantly.

3) Create initial ad sets and their ads

For each audience segment (e.g. mothers, software engineers, men under 40 with any job title), create an ad set for every value prop you want to pitch them.

In each, create ads presenting the same value prop in distinct ways—via distinct ad copy and creative.

The more distinct ads you have, the less frequently users on profile-targeted channels like Facebook see the same ads on repeat, so the longer it takes for them to tire of your product.

4) Monitor

Check the ad channel's dashboard once daily to see which ad sets, ad value props, and copy/imagery combinations perform best.

If you've reached a significant sample size of 3,000-5,000 impressions, turn off all the ads with much lower click-through-rates (CTR’s) and/or cost-per-customer-acquisitions (CPA’s) relative to the others in the same ad set.

If some ads perform much worse than their siblings and they've only reached, say, 1,500 impressions, you can turn those off too instead of wasting money on them.

5) Optimize

Begin tweaking your highest-performing ads by duplicating them and making changes to their copy and creative.

(Build the habit of archiving all your work so you can revisit past costly-to-acquire advertising data when setting up future channels.)

When tweaking ads and ad sets, you're looking to lower your CPA and increase total conversion volume.

At some point, depending on your audience size and budget, audiences will tire of  your ads and will increasingly ignore them. Your CTR’s will steadily drop.

At this point, a more realistic goal is not to drastically improve CPA's, but to stabilize them: keep your ads fresh by regularly creating new ones and including multiple in each ad set.

If your optimizations cannot sustain acceptable CPA’s, it’s time to pause the campaign. In a few weeks' time, you can resume the campaign as-is. Audiences will be less tired by then. Surprisingly, this works fairly often.

6) Run ongoing experiments

While optimizing the campaigns that work, simultaneously test radically different ones: new audiences, value props, copy, and imagery.

Running ad channels is a process of relentlessly pursuing improvements—not finding a few small wins then setting and forgetting it.

If you don't continually experiment, you're leaving better CPA's on the table. I recommend setting aside 15% of your budget for bi-weekly experiments.

7) Retarget

Segment out the people who visited your site but didn’t convert into a separate ad audience. (Using conversion pixels, every major channel allows you to do this. These are called "custom audiences.")

When retargeting a custom audience, you can push them a unique set of ads that play into their newfound awareness of your product and its benefits: Consider, which new value props can you pitch to tip them over the edge and purchase?

(I talk a lot more about retargeting later on this page.)

Ad sets: Segmentation

Ad sets are the level at which you group ads by a value prop and an audience.

On Facebook, it's critical you don't group multiple value props into a single ad set. Because, when you launch your ads, Facebook assesses the CTR for every ad in a set then stops serving the worst-performing ones.

So, if you have ads pitching multiple value props in the same ad set, even the slightly lesser-performing value props may never be seen by your audience.

This isn't acceptable: Audiences need to be pitched a product from multiple angles so that your messaging stays fresh. Even if some value props perform worse, it'll boost performance in the long run to not have audiences see the same value prop on repeat.

Ad sets: Targeting

Let's run through the criteria via which Facebook allows you to target audiences.

Targeting option: Geography

Geography is the biggest demographic determinant of cost-per-impression.

First, note you'll probably want to select the "People who live in this location"  option. This is in contrast to people who "are in" or "were recently in" your list of locations. Usually, the purpose of targeting by geography is to show your ad to people who have homes in that location and live there most of the year.

Second, know that countries split into three unofficial geographic "tiers:"

Begin by targeting countries you believe are the most likely to convert. This keeps ad spend efficient while you test your initial audience, value prop, and ad hypotheses.

Targeting option: Language

Explicitly target English speakers if you're targeting countries where the majority doesn't speak English.

If you are targeting an English-speaking country such as the US or the UK, leave this field blank—otherwise you're missing out on Facebook users who have their UI language set to something other than English but may in fact speak English.

Targeting option: Gender and age

Unless you have a specific reason to, avoid restricting by gender for your initial ad set tests. Instead, keep your targeting broad then parse your ads' performance by the gender and age of the people who actually converted into customers.

You always want to let the data tell you who to target. Don't prematurely constrain your audience unless you have a strong reason to.

That said, there is a rule of thumb for age: If you're not explicitly targeting a younger demographic, the key buying age range is 22 to 50. Run ads to this range for better performance CPAs on average.

Targeting option: Device type

Here's what device targeting looks like in the Facebook Ads Manager:

If you're advertising a mobile app, obviously only advertise to mobile devices. Facebook allows you to target just Android or iPhone devices, which is critical to spending efficiently if your app is in only one of those app stores.

For web-based products and services that are also accessible via mobile, it's worth testing mobile targeting too because, if the CPA's are acceptable, it'll expand your audience volume. Many people only use Facebook on their phone. And Instagram is almost exclusively accessed via mobile.

Targeting performance breakdowns

The targeting criteria we've covered so far can all be segmented and assessed within the Facebook Ads Manager dashboard. Meaning, Facebook will let you "breakdown" your CPC and CPA metrics by each of these criteria so you can compare how they individually perform.

But, for the upcoming targeting criteria I'm covering, Facebook will not allow you to perform breakdowns on them. This means that each of the upcoming decisions would result in a new ad set so you can track and compare that criteria's performance.

Targeting strategy

With your demographic and device targeting chosen, your next step is determining your primary means of profile targeting:

Because the Facebook/Instagram dashboard doesn't let you breakdown by interest and lookalike audience, you must create separate ad sets for each of these.

Let's begin with profile targeting then we'll explore lookalikes.

Targeting: Profile data

Here's an incomplete review of the profile data Facebook allows you to target:

1) Profile data: Interests

Facebook determines user interests based on Pages Liked and links shared.

Examples of interests include:

• Activities and initiatives — Fishing, drawing, video games, veterans' support, etc.
• People and entertainment — Tom Cruise, Barack Obama, Rick and Morty, etc.
• Companies and organizations — Nike, Nintendo, Red Cross, etc.

A common mistake novice Facebook advertisers make is targeting high-level interest categories, such as Food, Web Design, or Marketing—instead of their niches such as BBQ, Responsive Web Design, and Digital Marketing, respectively.

Consider this: If someone once Liked "Food" on Facebook, or some other broad topic, that's a very weak indicator that they're food fanatics. Which is probably what you actually want to target to increase conversions for your cooking ads.

The other problem with broad targeting is that it makes it hard to know which niches are performing best. As I mentioned, Facebook won't breakdown interests for you. So unless you're targeting Indian food and Japanese food in separate ad sets, you won't know which cuisine-based interest is converting best.

If you're targeting via Lookalike audiences, which I cover shortly, starting broad is okay. Facebook is good at algorithmically narrowing into the best audience subset.

2) Profile data: Job titles

You can also target people based on their job title or job category. Examples of job titles include Digital Marketing, Software Engineer, CTO, etc. Examples of job categories include Sales, Marketing, Business Operations, Computers and Technology, etc.

Unfortunately, there are endless variations to how people title their jobs on Facebook profiles, e.g. Salesperson, Sales Associate, Sales Manager, and so on. Fortunately, after we enter just one of these titles into the Facebook ad set creator, related job titles appear. Spend a few minutes identifying all the relevant ones.

Job targeting is typically reserved for B2B companies who know the exact job types that most benefit from their product.

Otherwise, if you're a B2C company, it’s generally more efficient to rely on interest targeting. It leads to bigger audiences—because Facebook's job data is very thin and will limit you to a subsection of the total Facebook user base.

You know how Facebook keeps nagging you to fill out your profile data? It's not because they want to help your friends know more about you. It's because they want more data advertisers can target you by.

3) Profile data: Behaviors

You can also target Facebook users based on their "behaviors," which are categories Facebook lumps users into based on their online activity.

Examples include:

• Users who recently purchased a car
• Users who donate to charities
• Small business owners

Facebook regularly improves the quality of these behavioral audiences. So they're worth testing.

Targeting: Lookalikes

Separate to profile targeting, Facebook and Instagram also support Lookalike targeting.

This consists of having Facebook dynamically generate an audience for you.

• You provide Facebook with a list of email addresses of people who have engaged with your business.
• Then Facebook algorithmically analyzes them for commonalities, e.g. demographics and interests.
• Facebook then searches for these commonalities among their total user base to create a tailored list of 1 million or more Facebook users who would be ideal for your ads to target.

The initial email list you provide Facebook is called a "seed" audience. It's often your existing userbase exported through your email marketing tool.

You want your seed to be the most down-funnel audience you have that's at least 1,000 people. Down-funnel means closer to the purchase conversion event. So if you don't yet have 1,000 purchasers, back one step out to registered users. If you don't yet have 1,000 registered users, back out to an earlier event such as add-to-cart (if you're an ecommerce company). And if you still don't have that, you can use your existing website visitors.

After you've uploaded your seed audience, you're presented with a slider to choose how broad you want Facebook's audience extrapolation to be. Stick with the default value of 1%, which matches 2M people (for U.S. audiences). You should consider going up to 5M only if your product appeals very broadly, e.g. food and apparel. Otherwise, your lookalike's matching will be diluted.

There are two things to keep in mind when uploading your seed list.

Seed size

If your seed audience is smaller than 1,000 people, you won't be giving Facebook enough data to significantly identify the commonalities among them.

Conversely, if your list is much bigger than 5,000, you're probably not segmenting out your best users (those with the highest LTV). It's much better to have a lookalike generated from your best, highest-paying users than from the average of your users.

Customer segmentation

Don't lump all your users into one seed audience if their purchase behavior varies significantly.

For example, if you're an apparel company, don't generate a seed audience from a combined list of men and women customers. They are vastly different apparel purchasers and should be targeted separately.

Instead, generate a seed for each major product category and create independent lookalikes for them.

Facebook Newsfeed ads

Let's run through each of the major components within a Facebook Newsfeed ad.

Newsfeed ad: Text and Media

The Text component of the Newsfeed ad unit appears above the creative.

Here's what to do with it: In two clear and concise sentences, describe the problem you solve or the value of the product you offer. You can be a bit wordy in service of contextualizing the problem if it's a non-obvious one, but don't ramble on about all your features and value props.

Examples:

• For a sales prospecting tool — "Need to track someone down? Enter any company, and we'll show you the employees' email addresses."
• For a visual site design tool — "Design sites visually. Get all the power of coding without any of the engineering hassle."
• For a therapy service — "We've found the best 14 therapists in San Francisco. If you're facing mental obstacles, talk through them with an expert who deeply cares."
• For a bicycle manufacturer — "Our tech-enabled bike raised $1,000,000 on Kickstarter. Craziest feature: We built lights into the handlebars that light up to indicate when your GPS wants you to turn."

Facebook Newsfeed Ads then contain an image or a video as their creative. Your choice of multimedia serves the same two purposes: grab the visitor's attention then get them to read your Text.

For more advice on ad copywriting and creative, read the Making Ads page.

Videos

Facebook videos often out-perform image ads. And, even if they only perform on-par, they're at least an alternative ad type to keep your ads fresh.

Here's what you need to know about running video ads—on any ad channel:

• Get to the point ASAP — Remember, unlike TV advertising, audiences aren't being forced to sit and watch your video ads. Your ad needs to hook them in the first 5 seconds. There is no time for preamble.
• Prioritize demos over lifestyle marketing — When aiming for conversion, prioritize demoing the product over broad lifestyle marketing (e.g. drinking a Corona on the beach). For SaaS software, this could mean showing a screen recording of the tool in action. For physical products, this could mean a 360 of the product followed by close-up shots of it in action.
• Use closed captioning — Never design ads that only work with audio. Assume users don't have headphones and that they'll be relying on in-video text overlays or closed captioning to summarize any non-self-evident parts of the video. (Facebook provides a handy closed captioning feature.)

Newsfeed ad: Headline

Your ad's headline appears below your ad's creative.

If there's no text in your ad's image, enter your primary ad copy here.

If you do have copy in your image, consider using this component to provide social proof, such as how many customers you have, who your marquee clients are, how many app store reviews you've received, and so on.

Examples of social proof:

• "Used by Zendesk, Salesforce, and Optimizely."
• "5/5 stars in the Google Chrome store!"
• "Over 5,000 photographers love us."

If you have none of these metrics to show off, pitch a secondary value prop:

• Clearbit: "The most up-to-date email address database in the world."
• Kip: "Vetted therapists with five-star ratings."
• Vanhawks: "A carbon bike frame for more speed."

Newsfeed ad: Real URL

This is your real landing page URL.

To track ad performance in your web analytics tool, each ad should have its own URL parameters ("UTM tags"). These will help you associate individual ad visitors with their long-term on-site or in-app behavior.

Newsfeed ad: Link description (optional)

This is the final piece of text you can include in your ad. Of all the text components, this is the least read by audiences. So don't put your most important value props here.

My recommendation is to use this field to address a significant product objection you know hurts conversion rates. Otherwise, if you have nothing critical to say, leave this section blank so your ad's other text gets more attention.

Newsfeed ad: CTA button

This is the button at the bottom of the ad that people click to visit your site. (They could alternatively click the ad's image, which is more common.)

Generally, you should only use the "Learn More" and "Shop Now" CTA texts. And only use Shop Now if you're an ecommerce site. All other CTA texts convert worse.

Monitoring ad performance

We're done with ad setup.

With your ads running, you'll want to keep track of several key metrics.

Monitoring ads: Metrics

You monitor ad performance in the Facebook Ads Manager. It's the default screen you arrive at on when you log into Facebook Ads.

The default reporting columns in the Facebook Ads Manager are not ideal. I recommend using this process to customize them as follows:

• Ad Set Name
• Amount Spent
• Delivery — The status of your ads. This warns you of ad rejections.
• Reach — This is how many distinct people saw your ad (any number of times).
• [Your conversion event(s)] — Once you setup conversion tracking, monitor it here.
• Cost per [conversion event(s)]
• Unique Outbound Clicks — "Outbound" means clicks that lead to your landing page (as opposed to clicks that lead to your Facebook Page).
• Cost per Unique Outbound Click (All) — This is your CPC.
• Unique Outbound CTR (All)
• Frequency — How many times each Reached person has seen your ad.
• CPM — I covered this in the Bid amount section. It's how much you're paying per thousand ad impressions.

Once you've configured these columns, click the Columns button once more and choose "Set as Default" to save your selections.

Feel free to further customize your columns however you wish. (Study the Facebook Ads documentation to learn the various columns.) Ultimately, what matters most is that you're tracking your cost per conversion. Your conversion event is usually acquiring a customer, which is your CPA.

Videos

When assessing the performance of your Facebook video ads, there are unique metrics to monitor, including Video Watches at 50% and Video Percentage Watched.

Video Watches at 50% is your key metric for determining how many engaged watches your video received. This is a reflection of how enticing the thumbnail image and first 15 seconds of the video are.

Optimize for this metric to increase the volume of landing page clicks that'll then be taken by people watching your video.

Video Percentage Watched is the metric that indicates the quality of your video: The more people who watch to the end, the more engaging the video is.

Monitoring ads: Breakdown analysis

To meaningfully assess your ads in the Ads Manager, you need to wait until you have at least 3,000 Reach per ad, which is the number of people who've seen your ad. That tends to be a sufficient sample size for comparing one ad's CTR to its siblings in the same ad set.

When this Reach is achieved, perform a breakdown analysis by clicking the Breakdown button next to the Columns button in the dashboard. Then select the demographic criteria you want to segment your ad data by.

If you find very bad CTR's for certain genders, age ranges, devices, or regions, consider removing them from your ad set's targeting. You can later return to targeting these demographics with ads better tailored to them.

Until then, focus on what works in the short-term so you can prove to yourself that Facebook is going to be a positive-ROI acquisition channel for you. And that you're successfully creating compelling ads. Try long-tail experiments later.

Keep in mind Facebook does not allow you to perform breakdowns by interests, job titles, or behaviors. Therefore, to assess the performance of sub-audiences within those targeting criteria, you should have created different ad sets for each.

Monitoring ads: CTR versus CPA

When tracking ad performance, we study two sets of metrics:

• Ad metrics — An ad's impressions, video views, and clicks.
• Site metrics — A landing page's page views, time on site, and conversions.

Site metrics are ultimately more important than ad metrics because they're further down the growth funnel. They're closer to the conversion event we care about.

So we must not overvalue great ad metrics without factoring in that weak site metrics could be neutralizing them.

For example, you could write incredibly enticing ad copy that generates an enormous amount of clicks—by promising something fantastic that your landing page doesn't deliver on. Great CTR, bad CPA. But all that matters to your revenue is CPA, so what's the point?

Conversely, you may have ads with low CTR because you're targeting micro-niches, but the few people who do click are highly likely to convert at high basket sizes.

Therefore, we only concern ourselves with ad CTR when we're determining which ads within an ad set are performing relatively worse. Shut those off and keep the top performers on. If the top performers lead to financially acceptable CPAs, then leave them running.

Monitoring ads: When to make new ads

Even if your ads perform well, whenever the following occur, create new ads:

• Product changes — When your product's value props change, create new ads to pitch what's newly offered and to target audiences that newly fit.
• Cultural events — Be on alert for events that are important to your audiences (e.g. holidays, seasons). Piggyback on them with promotions.
• New channels — Be on the lookout for social networks newly releasing ad platforms. When dating sites and messaging apps, for example, gain traction, they often release ad platforms to further monetize themselves.
• New ad units — Be on the lookout for existing ad platforms releasing new ad units. Facebook and Instagram release new ad units regularly. Test every new ad unit to see if you can find a high-converting fit.
• New multimedia — Across all channels, video ads tend to outperform image ads. So if your company generates new video marketing materials (e.g. case studies, promo videos, launch videos), it's worth testing ads with them.
• Market and LPA migration — If you're in a hot market where consumer savviness is rising, more people may be on a higher step of the ladder of product (LPA) awareness today than when you first ran your ads. So you'll want to test brand new ads to more accurately target this LPA-climbing audience. For example, if you're selling a virtual reality headset and the Oculus Rift has been out for either a month or a year, there's a significant difference in how much the public now knows about virtual reality.
• Competitors become popular — If your market becomes saturated with competitors, you may want to double down on copy that highlights what differentiates your product—even if it means spending fewer words explaining the problem you solve. (Plus, thanks to your competitors, now more people know about the problem to begin with.)

Facebook retargeting ads

Most people who visit your site will not convert.

These non-converters aren't a lost cause, however. In fact, they're the best possible audience to target with ads because they already know who you are. So now your ads can narrowly focus on explaining why they should use you instead of pulling double duty also explaining what you do.

Advertising to site visitors in this fashion is called retargeting.

As long as these visitors were well-fitting audiences to begin with, retargeting is always worth your time: Retargeted conversions tend to cost much less than prospecting conversions (which is the term for first-time targeting).

In fact, to not setup retargeting is to lose out on a single digit boost in your monthly conversion volume. Be prepared to spend perhaps 20% of your ad budget on retargeting.

On Facebook, a retargeting campaign is created similarly to a prospecting campaign. There are a few differences in your audience setup, value props, and landing pages.

Retargeting: Audience setup

To maximize the efficacy of our retargeting, we should create up to four custom audiences using the Facebook conversion pixel.

Retargeting: Ad set targeting

Here's where these 3-4 new custom audiences come together.

You're going to create a new ad set that:

• Includes A) site visitors B) 50%+ video viewers (if you run video ads).
• Excludes C) 3 day visitors and D) converted visitors.

Got it? If not, take a minute to re-read that logic: our current goal for retargeting is showing ads to people who engaged but didn't convert.

Retargeting: Show new ads and landing pages

Since your ads failed to convert retargeted visitors the first time around, don't show them the exact same ads again!

Instead, pitch them new and complementary value props.

Further, if you want to go above and beyond for retargeting, not only should you show new ads, you should also show new landing pages. Specifically, test pages that separately focus on each value prop you're newly pitching: If you couldn't get visitors to convert with a broad focus, try going narrow with your pitch.

Putting it all together

Here's what everything on this page looks like in practice.

Create campaigns, ad sets, and ads

• Campaigns — Create one prospecting campaign and one retargeting campaign for each product.
• Ad sets — Within every campaign, create one ad set per combination of value prop and audience segment. Keep your initial targeting broad to leave room for surprises. And use a 3 day exclusion audience to avoid inefficiently over-exposing people to your ads.
• Ads — Within every ad set, create ads that pitch the value prop with a unique combination of copy and creative. Create as many ads as you think are good. If you can't brainstorm at least 5 ads per set, you're probably not giving them enough thought. Refer to the Ad Copywriting page.

Monitor performance

• Ads — At around ~3,000 Reach per ad is when you have a sufficient sample  to determine whether you should turn off an ad with low CTR relative to its sibling ads in the same ad set. Ads with CTRs 30-40% less than their siblings should be turned off.
• Ad sets — Use Breakdown analyses in the Facebook Ads Manager to inform you of which gender, geographies, and age ranges perform relatively poorly. Do this once each ad set has at least one ad (preferably more) that has attained over ~3000 Reach. Consider removing the weak demographics from your targeting.

Optimize

Retarget

• Setup your retargeting campaigns with the custom audience combinations.
• Retargeting campaigns should have unique ads that pitch new value props.
• If you feel like going above and beyond, you can also design new landing pages that focus on pitching these new value props.

If this was a lot, we understand. We've been running ads for years.

If you want us to run them for you, check out Ad Labs.