One day, a major source changed its algorithms, but we couldn't stop the campaign because passing the learning phase again would be more expensive, and as it was a top channel, we had to adapt. Eventually, we managed to adjust, and things got even better than before. However, at the moment, you can lose hundreds of thousands of dollars. For a small company without a budget reserve, this can be critical. Let's explore what could happen and how to prepare for it.

So, the traffic source is about to change its ad optimization algorithm. We learned about this a couple of weeks in advance, but the exact launch date was unknown.

According to the logic of the new algorithm, it was necessary to reduce the target ROAS goal in advertising campaigns to reach the KPI ROAS business goal according to the Payment Profile. We needed to lower it by multiples because previously, if you needed a ROAS target of 15%, you set it to 18%, and it somehow lived within the maximum 20% difference. For us, a percentage discrepancy of up to 15-20% was acceptable.

Then we learned that soon, the increased goals would most likely not work. To get impressions, we needed to lower our target.

Tip #1. Maintain good relationships with your partners, even if the spend isn’t high. Perhaps, not everyone knew about the algorithm change in advance, especially among small companies. Therefore, it is always better to be on good terms to improve your chances.

We set up emergency Slack channels, discussed our options, and considered different scenarios. But it was still unclear how to react. Perhaps it was possible to switch without consequences, but the point is that our optimization, tailored for the previous version of the ad optimization algorithm, stopped working on the day the algorithm changed. That meant we had to change our previous approach completely.

Tip #2. Consider the possible workflow in case of an emergency in advance. Some should shift their budgets to other sources, while others may find it cheaper not to stop advertising — more on that later.

So, our advertising campaigns stopped paying off. We significantly reduced the ROAS goals, and it took about two months for the algorithm to be retrained on new data.

For a large publisher, this could easily mean a six- or seven-figure loss. For small publishers or independent studios, this can be critical. Firstly, they may not know about the algorithm change in advance, and secondly, not everyone has enough budget to wait out the storm.

You again have to invest in the source for it to pass the learning phase, where you have long laid out the spend costs, set up the creative pipeline, dimensional conversion along funnels, and built a model. But the model collapsed because of an external update.

Passing a new learning phase is still expensive because you need to collect a minimum threshold of payers and installs for the ROAS algorithm to work correctly.

If you’re in the hyper-casual niche and have AdROAS optimization, things are not so grim — you only need a couple of hundred installs for a closed D7 cohort on average. It's very cheap. But if you have a mid-core game, where your cost for a unique payer for optimization is $400-600, then you need dozens of such users. Therefore, stopping is not an option, as this leads to additional expenses due to relearning and costs tens of thousands of dollars.

Tip #3. If possible, keep a cash cushion. Otherwise, it might hurt.

The first thing we did was assess the discrepancy between the actual ROAS and the campaign ROAS goals. Then, we referred to our ROAS KPI Dash, an internal tool that indicates our ROAS KPI for the cohort day, taking into account the different ratio of in-app purchases revenue. This tool helps determine the target needed to break even based on the accumulation, curve, and ratio of in-app purchases revenue. We took this as our real reference target and lowered the rates by 25%+, although our usual optimization steps are 5-10%, with a maximum of 20%.

Due to the algorithm change, different optimizations perform differently. What didn't perform before may start to pay off better — the task of UA managers is to find a growth point and shift the effective advertising budget to this optimization. But in the end, when everything settled down, the effectiveness of our advertising channels and spans increased.

Tip #4. From a UA perspective, quickly respond to updates and make targets according to the network's recommendations, but rely on your experience.

For example, the ROAS of the first days may become bad, but if we attract paying players at the same prices as before, we can continue without fear.

At critical moments, I tend to trust the networking. We constantly meet with partners and communicate — it’s not only pleasant, but you can also learn a lot of insider information.

In moments of crisis, you need to believe in something. You can trust an advertising network that has more internal data, for example, about the operation of the algorithm, or you can trust your established metrics, flow optimizations, reference metrics, etc.

I don’t know how we would have survived that situation if we had stayed at the old targets. Perhaps the campaigns would break even, and we would not be competitive at all in the auction. Staying on the old flow, reducing targets, and lowering conservatively by 10-20%, we would have reached targets for effective spend and volume in 2-3 weeks, and, in an empty auction, when they rush again to a new algorithm, and you can miss the opportunity to grow due to having a honeymoon place in the auction.

A few more words about networking. Not only personal meetings work, but also participation in various online communities. There are actually working chats where practical information often appears, and participants answer constructive questions.

Tip #5. Small studios should take resource diversification seriously and shift budgets in the event of an emergency.

Let's take three common resources — AppLovin, Google, and Facebook. If one of them sags, then shift part of the budget to the remaining two while you are training on the first one. But if you have only one source, this can become a big problem, and I have actually seen such cases. It's cliché, but don't put all your eggs in one basket.

Tip #6. You can always ask for help.

You don’t have to do it alone. Independent and small studios do have a tougher time in the market. And large publishers have not only resources, but also established relationships with partners. Nothing to add here.

Sometimes, there is no way to prepare at all. We had a case when the payback on Blended ROAS optimization on one of the sources suddenly dropped by one and a half times with the same targets.

We were looking for a problem in our accumulation of ARPU Growth due to product updates, in setups, and creatives, because before that we were breaking even. We started rebuilding the optimization mix, rearranged some things, but still didn’t pay off, although not that much. As a result, after some time we learned in retrospect that it was a bug on the resource side.

What could be done here? Without the so-called “air” in the form of money, it would not be easy. We were also saved by diversification across campaign setups, which smoothed out the drawdown. And here, you can’t really advise anything.

In the end

It all depends on the business goal. If you have a small budget and it is vital to spend everything to turn around, it may make sense to stop campaigns in the event of an emergency.

Then transfer it to other sources, but don’t leave the auction completely, set yourself a period — a week or a month — to find creativity, a new approach to problematic sources.

For some, it’s more profitable to stop and go to retraining, or cruise on a low spend, but not for everyone. Someone may have a commitment to the developers, or there is simply a run-rate monthly goal you can’t fall below. If you constantly attract users, you should not leave some networks, because training will break, and restarting will be much more expensive.