Home > Paid Traffic Sources > Facebook & Instagram

---

Campaign in Learning phase, high cost per conversion. What shoud I do? (47)

---

Hello all,

I have a FB running campaign that targets a look-alike audience built on 1000 past customers.

The campaign, after some days, is still in Learning phase, I've made some conversiosn but the cost-per-conversion is higher than the payout. ($22 cost vs $18 payout)

What should I do? Should I await for FB to complete the learning or this campaign has no hope?

thanks

---

$22 vs $18 is not a tragic result, I would let it run for sure to see if the performance improves once the learning phase is over... unless you're spending a lot. How much are you spending per day on this?

---

I'm spending 10 euro per day, really not much :-)

Also, in this campaign I have only 1 creative, the CPC is about 1.7 euro

thanks

---

Ok, that's a really low spend. So what do we have here: Very low spend, just 1 creative tested... and you're still making sales, the ROI is not THAT bad, that's a very good sign.

Personally, I would up the budget to see what the AI does once the learning phase is complete. Just to know how good the "first shot" was

---

Test aggressively if your budget allows it.

Add 3-4 more creatives. Duplicate the campaign. Try ABO with a whole bunch of ad sets inside, duplicated even. Cut duplicates really aggressively every couple of hours. Throw in videos vs images. Try double budgets vs boosting budgets in small increments. Make more audiences. Test broad too. Test interests. Scale using cost/bid caps & crafty audiences. And fuck the learning phase, just feed your pixel data, it is irrelevant for the most part and an annoying nuisance. But most importantly, try all of these and any other possible ad type/objective/variation or a combination that Facebook provides, at once. There's really not that much else to it and it is rather logical as long as you don't have cold feet and aren't scared of spending money.

Facebook is extremely simple to get to work – a trained monkey can run profitable ads with mediocre and bland creatives and angles as long as the product/offer is good – meaning unsaturated and having high market demand.

Then you can begin worry about the truly difficult advertising bits. Aligning a new product to a completely unfamiliar market - e.g. no existing demand – via great angles, copy, creatives and research, and a deep understanding of your market. Or cracking through an extremely saturated one in that same way. That's the hard shit and it's not at all about what any one traffic source, Facebook or any other, has as targeting options, objectives, ad types, flexibility, loopholes, bidding, momentum, algos or techincal/setup aspects -- which can work a lot of the time, yes, but that's besides my point. It's about being creative/innovative.

tl;dr Read Breakthrough Advertising or test a metric ton of stuff, quickly.

---

Thanks!

the offer is a product of mine that is selling well via a website that gets good traffic (SEO and a good mailing list)

So I think that demand and market are ok.

I'm testing Facebook ADS to accelerate growth, by reading your words and others it's a matter of testing as many as possible variables and let the AI decide what works best.

That means having a good budget but yes, it makes sense.

The only doubt I have is how could I leverage the assets already have (website, SEO, list and funnel) to improve Facebook ADS performance, if a way exists

thanks

---

Yep, this is not the classic AM campaign setup in which we test many things. I know I should expand testing, but for this campaign I'll follow your advice, I'll increase the budget and see the AI how far can go

thanks

---

I don't think I've ever nodded in agreement more while reading a paragraph... spot on x1000 man...

---

this is gold! thanks for the knowledge bomb

---

I would do it, because I would want to see what happens once the learning phase is over... so just like an experiment, to know whether it makes sense to use it in the future or not. But what shishev outlined above is what you should put focus on down the road. The truth really is, FB is quite easy to run if you have a solid and whitehat product. Many people struggle with it just because they are loosing accounts faster than old man loses his hair or they run products that everyone has seen a thousand times already... what you promote is your own offer and I think quite a unique one, so it should work pretty nicely.

---

This normally happens when i've set a low daily ad budget on an adset/ad, a minimum of $50/day on Financial Lead Gen offers, but on sweepstakes offers i don't see this issue running a minimum of $10/day daily budget. From your offers payout, i guess you're have to increase budget until the ad gets out the LP.

---

Thanks; I've raised the budget; I'll keep you updated :-)

---

Hi there, just my 2 cents to add to all the other folks said.
In my experience, raising the budget is not the best practice. I noticed that rising budget will not increase the quality of the users involved in the campaign. Fb just wants to spend all of your money.

Which is the country you are targeting? US?

In my opinion you should try different targeting settings and test at least 3-5 different creatives with your actual budget.
Another techniques which works often for me is to promote a post in roder to have a lot of interactions on that one. Than use that post as your Campaign so who sees the Ad, will see a lot of positive interactions and somehow Fb will lower the CPS because of the quality of the interactions.

But hey, we are in marketing so everything should be tested and nothing is forever true. Good luck.

---

Stop tempting me to run FB shishev style

---

I'm a bit excited and surprised.... I've doubled the budget and expanded the audience (targeting general interests related to my niche)..... now the ROAS is 1.6 with the Pixel still in the learning phase

@jaybot I feel like a well-trained monkey :-D

---

I personally haven't observed night-and-day difference between campaign performance during and after the learning phase.

@shishev's advice is solid.

Basically what FB is trying to "learn" here is how to refine its targeting (by figuring out who to show your ads to) to get you better results.

You can potentially do this so much faster - and more efficiently - by testing different audience targeting, and also by testing different audience POOLS within the same targeting using duplicate adsets.

Even performance between adsets that are identical, can be unpredictable. The best approach I've found is to throw up a bunch of tests to see what sticks, and pause quickly what doesn't work right off the bat.

More details can be found in the FB tutorial here: https://stmforum.com/forum/forumdisp...inner-Tutorial


Amy

---

Not bad, not bad at all So how many sales are we talking about here? Trying to figure out whether we are still in the "lucky/random conversions" territory or we're on the path to something "greater"

---

Yeah I know what you mean, statistical significance is crucial. We are talking about 10 sales (the payout is XX), let's see what will happen in the following days.

Anyway I should definitely test more things :-)

---

Thanks @Amy, this insight is helpful!

---

10 sales is not bad actually, the random stuff usually happens in lower quantities, as long as you enter 2 digit territory that usually means you're onto something

And yes, definitely test more things now! This small experiment definitely confirmed that the product has potential and can convert just fine.

---

Update: I did a few more sales, but the ROAS decreased to 1.2

Campaign still in learning phase

---

I highly suspect you won't get out of the learning phase. If i'm not mistaken, facebook needs 50 conversions in a 7 day window to get out of the learning phase.

With that spend and ur cpa, it's not getting there. Very soon it will become learning limited. Which doesn't matter in my experience.

Hey, but I've not been successful in this. I spent about 30k on facebook ads for this one gig last year and was only slightly profitable at the end.

Every time I raise the ad spend (10%/20%/100%) , the CPA would go through the roof. I didn't have the budget to get 50 conversions in a week considering the raising CPA. So I wouldn't know what happens when it does get "optimised".

I think with a low conversion rate product offer (like I had), facebook can do wonders if you are not aiming to scale. It really just gives you the highest quality traffic. I was spending $20-$30 a day, and was consistently getting a sale or two making $30-$40 profit.

---

Thank you that's really helpful! Is there some documentation that states "facebook needs 50 conversions in a 7 day window to get out of the learning phase" or is this based on your experience? Just trying to understand if this is a strict-rule

---

I read that in facebook documents I believe.
Googling 50 events optimisation facebook ads should yield something.

---

Indeed!
This page https://www.facebook.com/business/m/...learning-phase says:

"Typically, performance stabilizes after an ad set receives around 50 optimization events within a 7-day period."

thanks

---

So how far away from this limit are you? And since the ROAS still looks positive, is there anything that holds you back from raising the budget further?

---

I have this warning:

So I'm not sure on how to expand the audience

---

These warnings are "default" or "automatic" so to speak and they might or might not have any impact on the campaign delivery. I've seen those even with audience sizes in the millions so it really isn't necessarily accurate. You won't know for sure until you increase the budget

---

Thanks another hint really helpful ! Yesterday I've increased the budget (100 euro x day now) and overnight I've made more sales...ROAS 1.36... let's see what will happen today :-)

---

I just re-read your post.

You're getting conversion at $22. Yet your CPC is $1.7?

So your conversion rate is, 12%?

Sorry, my case is completely irrelevant. Please ignore what I said.

---

I remember when I was doing FB in 2019 this was baffling me big time. It didn't make any sense to me - an adset would be doing fine but then the performance would drop like a rock. Duplicating this adset didn't help either - the performance could be the same or even worse

I learned that the concept of "adset stickiness" (read - how well an adset performs) was a dogma in russian-speaking affiliate marketing community. A lot of people believed (and do believe now I think) that if you clone an adset and it doesn't "stick" - you drop it and duplicate again. You are basically running adsets as long as they "stick" and when performance drops - you stop these adsets immediately

No one tried to question why this stuff is happening. People just believed that there is such a concept and just went with it without trying to fin the root cause. It reminded me of a parable/joke:

A young lad comes to a rabbi and says: "Rabbi, you are so smart and wise. What is life?". Rabbi replied calmly and slowly without any hesitation: "Life is a river". The lad came back home and told his family what rabbi said. The word was spread across the village: Life is a river. Everyone was telling everyone that life is a river until the word came to a young boy who asked: "But why is life a river and not anything else?". The chain reaction started going backwards: everyone started asking everyone: "yeah, why is life a river?". Later the same lad comes to rabbi again and says: "Rabbi, you are so smart and wise. Everyone in the village is asking why life is a river and not something else". Rabbi frowned and was visibly confused. After taking some time he replied: "Maybe it is not a river......"

So, I started digging into the data and was terrorizing FB Blueprint courses and FB for business page. I was also running some tests to see the performance of adsets. I tried different creatives, adsets, budgets, etc. etc. The same thing was happening: some adsets were indeed not sticking. But. Even those adsets that were doing well started losing the steam and tuned to crap. So, back to the drawing board. Let me share an insight that I learned myself:

The first thing that I learned was how we perceive "lowest cost" bidding strategy. The name is hugely misleading because I learned from FB for business and blueprint courses that after the learning phase is finished, the lowest cost bidding strategy chooses the slice of an audience that will bring the most volume in the long run. Even if this slice is unprofitable for you. Then, FB will try to bring more volume and the lowest conversion cost with your creatives-LP-offer combination. As you can guess - in some (or many cases) this is a problem. Why? From my little experience with Car Loans lead gen - the biggest slice that was giving the most volume was 65+ age group (read - FB chose this age because he thought that it can provide higher volume), while the best slice was 35-44 age group in terms of cost per conversion and lead quality for the advertiser. 65+ was both shit quality and highly unprofitable segment for the offer I was running. So, what was the main reason why my adsets were turning to shit, even though they were performing well in the beginning? The adsets that were doing well "sticked" to 35-44 age group and the adsets that didn't do well at the very start sticked to 65+ age group. Can you guess why good adsets eventually turned to shit? Yes! They started prioritizing 65+ age group. Facebook noticed that old dogs have more volume than 35-44 and shifted his focus to them. So, the age groups are solely responsible for the fluctuations and stickiness of the adsets

What about interest targeting? As I read from FB itself - "unless you know your target audience to a T - let FB do the job of choosing interests". Yes, as far as I remember - I had better results with no selected interests rather than "Car Finance" and similar interests. The interests were definitely not responsible for adset performance - they didn't matter and I had a feeling that even when I chose interests - FB disregarded them and started targeting different interests. This is my guess because we can't see what interests FB is targeting

Sorry for hijacking your thread a little bit, but I thought this would be helpful

With all the yada-yada above, I would probably do this:

1. Create campaigns that are targeting different age groups.

2. Within those campaigns create adsets that reflect different angles.

For example:
Let's say I promote car loans. I created a campaign for 35-44 age group. No interests. I want to test 2 angles: "0 down payment, cheapest rates" and "stop using carsharing/subway because getting your own car has never been easier" (so, 2 adsets withing this campaign). Within "0 down payment" angle I would try images of different pickups, sedans, camera angles, etc. For "stop using subway" I would use images of people in subway during peak hours (nightmare ). You got the idea. If I remember correctly - FB allows you to upload few images and copies for your creative and it will auto-optimize them to choose the best combination

3. I am not sure about what to choose: CBO or ABO because I don't remember the exact difference between them. Probably CBO in this case

4. Let all of these age-separated campaigns run for a while and compare their performance to see what campaign gives better performance according to your goals: ROI, volume, etc.

But I think you shouldn't listen to me because I am not running FB these days and there are guys that know their ropes. I don't . But if this info provides any value to someone
- my pleasure

Also, I don't think that "learning is limited" impacts the performance in any way if you are isolating age groups or targeting LAL

---

Great observation and thanks for all the tips!

What you said illustrates the importance of two things:

1)Identifying profitable age+gender segments and excluding non-profitable ones.

And your suggestion of setting up adsets to target each age+gender combination isn't only good practice - it has almost become a must-do after IOS14.

2)Choosing audiences that are big enough.

The algo will do its best to bring the most-suitable prospects, but after that, if you continue to run the campaign, the algo will start bringing increasing less-suitable visitors.



The "duplicate adsets/campaigns and leaving the winners to run by killing off losers" approach DOES work, but the premise (as you've implied here) is that a)the audience is "good enough" to convert for the offer, and b)the audience size is large enough to sustain the duplicates.

Both conditions must be present - otherwise you can duplicate stuff up the yin yang and they'll all go to shit.

(And of course the creatives have to be strong in the first place.)



Targeting interests have worked for me. But again, it depends on how niche the offer is, and how big the audience size is for that interest.

I definitely agree with you though that we should target broad if possible. If the offer is remotely broad, definitely test broad targeting, i.e. gender and/or age only, as a maximum. FB has become pretty good at identifying suitable visitors to target, plus it's easier to scale when targeting broad vs. by interest.



For ABOs, we specify a budget per adset, and this budget is automatically distributed across the ads within the adset (with more/most of the ad spend going to what FB interprets as the "best-performing" ad).

For CBOs, we specify a budget per campaign, and the budget is automatically distributed across adsets within the campaign - again, with more budget allocated to what FB deems to be the better/best-performing adsets and ads.

The basic approach is to use ABOs for testing, and CBOs for scaling.

For testing, usually we'd want each adset (and the ads within it) to get enough traffic in order to give them a chance to prove itself. Using CBO may not be the best idea, as the algo may send more/most of the budget to the "wrong" adset - for example to an adset that makes a lottery conversion by luck, instead of to another adset that has would turn out to have better CR if only given a chance to convert over time.

The same "risk" exists for CBOs - if we put untested/unproven adsets/ads in a CBO, the algo may favor the "wrong" adsets/ads by sending most of the traffic to it. Therefore, it's safer and will yield better results to only put proven adsets and ads into a CBO, and then let the algo do its magic by giving more budget to the the winner adset on any given day.

Hope that helps the OP!




Amy

---

@vortex

Yes. I think this is the most important part since there is a conflict of interest between you and FB. FB recommends "lowest cost" because he wants to spend all of your money and we don't want to spend all of our money if it is not profitable for us. Target cost and cost cap can work but they are (or at least were in 2019) flawed and don't align with our goals as well. So, IMO, splitting the campaigns by ages is the best way to control how FB spends our money. Since we know that age groups are responsible for FB's decision on where to allocate your budget down the road - we have to deal with it. We want to spend our budget evenly across all age groups. This is not possible if we don't restrict the targeting by age because FB will favor one age group over others sooner or later (usually sooner) which leads to inefficient and uneven spend. Our job is to find profitable segments and angles/lps/creatives and try to scale them as much as we can, while FB's job is to spend all of your money


This is also a problem when people try to scale by increasing daily budget. FB is greedy and it will not tell you: "dude, your daily budget is too big and we will not be able to find a good audience for you". It will just start choosing colder and colder audiences because it wants to spend all of your money. So, our scaling is limited to diminishing returns of the combination of our daily budget+our funnel (age group-gender-creatives-landers-offers). If for example we see good performance with only 1 age group - we should try to max out this age group as much as possible by identifying diminishing returns of our daily budget by comparing the relative increase in traffic/conversions vs relative increase in budget

You are right. If your offer is broad (car loans, for example) - FB knows better than you what audience to choose. If you have a very niche product with small target audience (read - FB doesn't have enough data for your niche/offer) - you know better than him what audience might be interested in your product

I see. I think for my suggested strategy it doesn't really matter what you choose (but ABO could be slightly better I think) - we should still manually analyze the results and choose campaigns and adsets that bring the best results for us, not for FB (the best result for FB is the volume, obviously. More volume in segment - more money for you to burn)

Anyway, I do hope too that this helps the OP and other members. Sharing is caring, you know

---

Absolutely! That's the whole foundation STM is built on.

Definitely resonate with everything else in your post as well!

Thank you for sharing!



Amy

---

@maestro and @vortex, thank you so much! These insights are gems.

I've increased the budget and ROAS is stabilized to 1.25, I really need to plan a new campaign with the right demo segments and some creatives to test; I'll post in the Follow Along section soon

Thanks again!

---

The concept is intriguing - I took a brief look - hoping to find time to read up on the subject.

One of the causes of the randomness of results on FB is that different adsets would hit on different "audience pools" within the same target audience. It would be synonymous to going fishing and dropping the line at a random location in the pond/lake. And of course different audience pools can result in different performance. This can at least partly explain why some adsets will succeed and some will fail, even if they have the exact same ads and targeting and bid etc.

I can certainly see similarities between the randomness in selecting audience pools, and the randomness in the initialization point on your graph.

Thanks for sharing - interesting stuff!



Amy

---

@matuloo @vortex @maestro I've launched another campaign, with proper testing, creative splitting and audience targeting; I've followed the advice on this thread and now the ROAS is stabilized to 2.1

I suspect there is a lot of room for improvements, which means more testing testing testing

---

Ah very nice @johncarlof ! 2.1 ROAS is already very nice and I'm sure you can improve further

What types of ads are you using now, static images or some videos too?

---

For now only static images, but video will come

---

I'm sure you have heard this already, but TikTok style videos work great for Ads on multiple platforms. So when planning to do some, try to do it in such a way.

I can totally imagine a video where you play the guitar and go over some part of your course, explaining how to improve a specific passage by following the particular lesson. Something like before/after would work well too. I'm sure you know what I mean.

---

@johncarlof Congrats on the 2.1 RoAS!

Next, focus your efforts on your creatives. That's THE thing that has the most impact on a campaign.

These days, videos are all the rage. Not saying images don't work (they do, as you've reported), but there's a preference for videos.

And what can work extremely well is UCG - user-generated content. Since you have 1000 past customers, consider asking them for video testimonials, in exchange for some type of perk.

If done right, this can be a never-ending source of video creatives for your ads.

Another thing you can try - now that you have good creatives and good audience targeting - would be to scale using CBO.

Make sure you have a big enough audience size though.

Best of luck on scaling! And please do keep us posted on further progress!



Amy

---

@matuloo @vortex thanks a lot for these other suggestions...they are really helpful again

I'm a quite skilled guitar player and I'm sure I can shoot some cool videos...but I'm a bit lazy and the process is time consuming...but I'll do soon :-)

now I have to do some math accounting to check cashflow and the like, if ROAS keeps stay good I have to pump money into the Facebook ADS blackhole ah ah

---

How is it going? How are the results? Is there an age group that is doing better for you than others? Is there an angle that's doing better?

---

Yes I identified an age range that performs well, it's perfectly aligned with the demo on the website.

---

Good. Gather data and dig into it to see if there is any difference in performance for genders/cities/states and test more angles for this age group

---

As long as you see positive ROI, this is a very pleasant experience Pumping money into a platform that generates profits, that's the best thing about AM

And definitely try to shoot some videos, this CAN be a game changer! And nice videos can also bring in organic traffic, which is a huge plus.

---

Yes I'm sure that video will skyrocket the effectiveness of the campaigns, in this niche videos play a big role!

---

Home > Paid Traffic Sources > Facebook & Instagram