Exclusive in-depth exploration of oCPC optimization logic and breakthroughs

Optimizing oCPC is a necessary and likely long-term challenge on the road to smart delivery. In particular, the products and technologies of various platforms are far from mature, with rapid functional updates and frequent changes. If marketers cannot deeply understand the optimization logic behind them, they will not be able to grasp the value and limits of "human operations" and it will be difficult to make relatively good decisions in a market with so many uncertainties.

1. Non-zero-sum game

Taking Baidu search oCPC as an example, although it is far from being a mature intelligent delivery product, it allows us to see: the future of optimizing complex advertising conversion effects through AI.

The working principle of search oCPC is to anchor CPA (cost per action) to adjust the bid for each click, thereby maximizing CPM (cost per thousand impressions).

Did you notice it? The ultimate goal of machine learning is to maximize the platform's CPM, that is, to maximize the display revenue, rather than the CPA of a single ad being more cost-effective. But there is no need to be pessimistic, because this is a non-zero-sum game. Technological upgrades have made the advertising pie bigger, and advertisers can also benefit from it. The original motivation for this is of course that the platform can increase its revenue, but the means adopted are positive - the platform improves the conversion value of its traffic through investment in AI models.

This also reveals a fact: whether an oCPC plan is good or not essentially depends on whether the model can find the optimal CPA solution that can balance its own brand and the advertising market as soon as possible.

2. Best CPA

Our delivery goal is to "control CPA costs and maintain sufficient volume." The optimal CPA is definitely not equal to a lower CPA, but rather a CPA that finds a balance between cost and conversion volume. As shown in the figure below, I drew an equilibrium diagram to explain the logic here, hoping to help everyone understand.

The X-axis is the planned conversion cost, and the Y-axis is the planned conversion volume. The increasing curve represents the "launch price curve". The higher the bid, the stronger the competitiveness, and the conversion volume will increase (non-linear). The decreasing curve represents the "business cost curve". Taking into account the scale effect, as the conversion volume decreases, the input cost that the company needs to pay increases (non-linear). In this way, the intersection of the two curves is the optimal CPA value we want to find. Therefore, on the premise that there are no mistakes in the target audience, all delivery optimization strategies can follow the logic of this diagram.

In this way, our optimization scenarios are only the following two:

(1) CPA is to the right of the optimal point

The CPA is relatively high, and we need to find a way to lower it, but there is a high probability that some volume will be sacrificed. We start the analysis by breaking down CPA, the formula is as follows.

After turning on ocpc, advertisers' CPC will increase due to more accurate traffic. At the same time, CVR will also increase because it is allocated to audiences with better conversions - the growth rate ratio of the two determines the final level of CPA. Therefore, controllable CPA can be broken down into controllable CPC and higher CVR.

At this time, the point of human intervention is to control CPC and improve CVR. The main methods are:

a. Ensure a good data foundation in the first stage of CPC

b. Check whether the data sent back is accurate

c. Increase display volume-keyword optimization

d. Improve click-through rate-creative optimization

e. Improve conversion rate-landing page optimization

(2) CPA is on the left side of the optimal point

At this time, the CPA seems very low, but the conversion volume is not enough, and the conversion volume needs to be increased. The most direct idea of ​​manual intervention is to increase the target CPA appropriately, but be sure to control the amplitude. The maximum should not exceed 20% at a time, and try it little by little.

In addition, there are also external methods to increase conversion volume. We can use the "Traffic Problem Attribution Chart" below to mark out the main external causes of low traffic and defeat them one by one.

a. Insufficient budget

Re-estimate consumption. If it is due to low budget, you can reallocate the amount to the ocpc plan.

b. Difficulty in consumption due to low display volume

i. Add keywords manually or use mainstream AI keyword expansion tools to increase the range of available exposure. It is not recommended to turn on intelligent keyword expansion, as it is easy for junk words to come in.

ii. The matching mode is single, and the matching accuracy can be appropriately relaxed.

c. Difficulty in consumption due to low CTR

i. Increase CTR (click-through rate)

Creative optimization, more fully configured advanced creatives, information arrays, multiple image styles, dynamic titles, combined with segmented businesses and populations, personalized creative settings. Further improve conversion rate and optimize landing pages.

ii. Low ranking (you still need to use the method mentioned above to increase CPA, which will not be repeated here)

If you understand the above optimization logic, you will basically know which links require manual participation and which links can only be run by machines. We can also understand it this way: oCPC undertakes high-intensity algorithms that humans cannot complete, and captures higher-quality traffic according to the established framework; humans are responsible for feeding the oCPC model enough "feed" and observing "environmental changes" at any time.

1. The key value of human labor

From the perspective of human-machine collaboration, we can sort out the key points in Part 1 and see that the core value of humans is reflected in:

(1) In the first stage of the account, it is necessary to rely on people to provide sufficient and good data "feed". The efficiency of the first stage depends on the optimizer's ability in the CPC stage. From account structure, strategy adjustment, creative optimization, effect analysis, etc., controllable costs are needed to meet the conversion threshold and let the machine know what traffic is needed.

(2) In the second stage of the account, even if it is running steadily, we know that it is only temporary and requires continuous human monitoring and manual intervention when fluctuations occur. This requires people to have a clear understanding of changes in the market and customers, such as creative strategy optimization, landing page design updates, etc.

2. Mistakes that need to be avoided in manual delivery

Due to the unclear understanding of the principles of oCPC delivery, delivery personnel still make optimization operations based on experience during the model learning phase, resulting in poor results. For example, when you see that the click price is very high, you adjust the bid in a panic; when you find that the system automatically adds a lot of inexplicable keywords, you quickly delete them; when you find that the system adds the same keyword to different plans or units multiple times, is there a problem?

In the process of the model recognizing your conversion users, you can no longer stare at the account with inertia. At this time, many phenomena reflect the language of the machine. We can let the machine handle complex problems first, and then see whether the results meet expectations after it stabilizes. Be careful not to use too many intervention models, which will disrupt the learning rhythm.

3. When is it appropriate to intervene and when is it not appropriate to intervene?

(1) Less intervention in the initial budget

ocpc is an AI algorithm optimization. The more budget you have, the more traffic will be allocated. Therefore, the budget should generally be sufficient. When setting an upper limit, it is best to set it 20%-30% higher than your affordability. For example, if your daily affordability is 10,000, the budget should be set at least 12,000. The reality is that the more initial budget you have, the more traffic you will get. After setting the budget, keep a close eye on consumption, but do not adjust it frequently. When you are almost reaching your limit, you can balance it by adjusting the time period, etc.

(2) Basic directional late intervention

Many marketers will set up many secondary targeting strategies, driven by experience and inertia. But the actual results show that it is best to cast all the ads at the beginning, so that you can get enough exposure. Once the effects are achieved and a certain amount of data has been accumulated, the targeting can be gradually narrowed. At this time, industry and business experience can play a role.

(3) Bidding is only intervened in the stage of exploring the optimal CPA

This part has been introduced in detail in Part 1 of this article. Recall the "Finding the Optimal CPA" curve. When there are obvious left-biased or right-biased features, manual intervention is required.

(4) Strong intervention in creativity, but less intervention after optimization and stabilization

If you prepare enough creative materials in the early stage, the system learning can distinguish the differences. Various different pictures, titles, and landing pages will make the system's learning results more effective and more reliable when automatically optimizing. However, since the learning stage has been completed, we should reduce the number of major changes to the advertising creative and reduce interference with the model.

(5) If the final effect is not good, strong intervention is required

If the account performance is not good, consider adding new products. If the account performance is good, it is best not to make new adjustments. Because new products will affect the original plan and lead to the loss of well-performing plans, which is not worth the loss. In addition, the new plan should be different from the original plan. It is best if the price, pictures, titles, and even the landing page can be different.

(6) Landing pages should be intervened at any time

In fact, most brand companies now pay great attention to their pages, because landing pages carry conversions, which is the most important part of promotion. Revision and replacement of landing pages is the norm in advertising planning, but few companies can achieve scientific revision, high-frequency optimization, and effective analysis. The landing page optimization link is currently a relatively backward link in the intelligent field. This is why Lingqi needs to focus on this point, because companies lack the support of favorable tools.

The bonus period of search oCPC has passed. As more and more industries join in, the unstable results will only increase, not decrease. The breakthrough point at this time will be placed on the conversion capacity (acceptance capacity) of the landing page.

The platform’s increased intelligence enables it to identify the conversion intentions of different customers. However, accuracy cannot reduce competition. When competition for traffic resources is fierce, expensive prices are always an obstacle that cannot be avoided, and the optimization space for delivery will become increasingly limited. At this time, the main battlefield of competition will be postponed, and the ability to control ROI will be compressed to each small link starting from the landing page.

Not only that, the landing page belongs to the advertiser. If the advertiser can independently control the intelligence of the page, the conversion ability of the page can be greatly improved through dynamic functions, thereby increasing the actual conversion rate of people with low conversion intention (determined by the platform). In other words, the traffic purchased at a low price also has the opportunity to obtain a good conversion rate!

This involves an aspect that advertisers tend to overlook: the accumulation of volume expansion capabilities. In fact, data assets can be accumulated indirectly in this area.

Taking Baidu's current search for ocpc as an example, the expansion models can be divided into three main categories: conservative expansion, balanced expansion, and aggressive expansion.

When launching the campaign, you can create different plans, adopt different expansion modes, and connect to different landing pages. Through comprehensive conversion optimization/analysis tools, we help companies accumulate their own segmentation and identification capabilities, thereby exploring their own ROI algorithms for traffic of different prices/attributes.

Such operations can not only help us explore the rules for expanding the business, but also combine them with the landing page's acceptance process to conduct population segmentation matching and data accumulation, thereby making the conversion acceptance capability more hardcore.

Of course, we still have a lot of homework to do for possible new breakthroughs in OCPC. But one thing that makes us proud is that the optimization logic has been basically sorted out, and there is also a first batch of users who are willing to work with us, starting from the landing page and making breakthroughs step by step. If you also believe that "whenever you encounter a bottleneck, there will be a breakthrough", please contact us, or we may become your long-term partner on the road to transformation breakthrough.

Author: Lingqi

Source: LingXi (MartechtoolLingXi)