To configure contextual advertising on your own is not so difficult, in a couple of evenings you can get it to work at grade "C". And if it works at grade "C", then it makes sense to invest and get it to an "A". And this is just the job for an agency, because the complexity between grade "C" and grade "A" grows exponentially. Conversely, if it does not work at grade "C", then most likely it will not work even if you bring it up to "A" either, since the cause may be not just contextual advertising, but some other factors like competition, prices, product range, delivery, etc.

The client had previously worked with freelancers, tried managing campaigns on their own and looking for a full-time specialist. So they came to us with an understanding of the task at hand and our role in their business. This enabled us to speak a common language and speed up all the processes.

The task was initially formulated simply as “context advertising seems to work, but I don’t understand how to manage it, let's do it properly.” This is just what we need, when the business model is tested, we need to systematize and scale everything up.

What will be discussed below?

• Conversions from contextual advertising were attributed to emails confirming registration and payment services, which prevented us from correctly assessing the efficiency.
• Users come to the website from context ads, leave to think for a week, and then they come back directly and make a purchase, and such conversions were being attributed to direct traffic, whereas in fact they were from contextual advertising.
• What categories of goods to launch first of all, when there is a 100 of them? We made a matrix of demand coverage by brands and categories of goods.
• How to calculate LTV with breakdown by several years, cities, product categories, and more?
• How to increase revenue twofold? No secrets, we just worked a lot :)

Audit

Before the start of a project, we conduct an audit to get a substantive understanding of how we can help the client and whether we can at all, for the following items:

• The client does not upload revenues to Google Analytics. Therefore, we investigated the number of confirmed orders and their cost by product category, brand, geography, display sites and other parameters, and there may be a hundred of them.
  
  

That is, our task is to find anomalies. For example, with the media-contextual banners the cost per order is twice higher than the average. The same for Omsa brand and for CIS geography.

Next, we think together with the client how to increase sales in the CIS and why Omsa sells poorly (the range was not wide enough).

• The client's previous contractor did not use brand queries in contextual advertising, BUT instead used an affiliated network to direct traffic to the website receiving a reward for orders. Which is quite a peculiar scheme! :)
  
  

• Contextual advertising did not use product model queries, which have the highest conversion rate, where the user is looking, for example, for a specific product model or make.
  
  

• The entire Yandex Advertising Network was not optimized. For example, there were sites bringing over RUB 10k in costs and not a single conversion
  
  

• Different cities had different conversion rates, but the prices were all set at the same level.
  
  

• No adjustments were made for device types or age and gender segments, although conversion rates were different and prices must also be set differently.
  
  

• We checked all links for 404 responses and found some errors.
  
  

• The client was analyzing the advertising campaigns in terms of the order quantity and cost per order, although each product category had its own average revenue and margin, which is the same as treating everyone with band-aids.

The story is quite trite: it is not so difficult to assemble and run advertising campaigns — just watch a couple of video lessons. The work begins when campaigns need to be managed and scaled up, for example, to manage payback:

Let's take car bumpers as an example of a product category. Yandex Direct auction works in such a way that the more clicks you get, the more expensive each next click gets. Hence, the more orders we receive, the more expensive each next one gets (see the blue line on the chart).

Since the cost per lead (see the chart) grows with the growth of order volume, at a certain point it starts to eat up your margin and you need to find the optimum. It's a school task to find the optimal production volume.

That is, for each product category you need to look for the best coverage, at which the profit will be maximized. And now, imagine you have 100 categories and for each you need to find the optimum — imagine that? And then each category has its own optimum by region, gender, age, devices and other parameters.

Let's make it easier. Imagine you have a chain of coffee houses, some of them make profit, and some make losses, but you invest equally in the development of each, so you grow very slowly because you sponsor unprofitable coffee houses the same way you sponsor the profitable ones.

Marketing analysis

Often, clients come to us with problems in contextual advertising whereas in fact there is a problem in marketing. For example, prices that are double the market level or the website is not user friendly. To eliminate these risks, we do a minimal marketing analysis. Some conclusions below.

We compared our offer with competitors in Excel format

• We made sure that our prices are in the middle of the market range, the delivery and payment terms are the same, the range is wide
• We studied the competition structure
• indirect (offline shops, car markets)
• regional stores with offline outlets
• specialized shops for certain types of parts (for example, towbars at farcopoff.ru , thresholds at http://всепороги.рф )
• large online stores https://www.ozon.ru/catalog/1175980/
• chinese shops (aliexpress, alibaba)
• major suppliers of spare parts exist.ru
• and so on
• We identified the largest competitors and investigated where they get their traffic from
  
  

We launch contextual advertising

Where to start if you have 100 categories of goods? For this, we have compiled a matrix of product category coverage by brand, where in blue cells we get untapped demand. Based on the matrix and other studies, together with the client we made an expansion plan and allocated a budget for testing.

Problem with conversion attribution

Challenges of growing advertising budget

When you make a landing page for the first time and launch contextual advertising, the analytics is simple: You spent RUB 10k, got RUB 100k, all the numbers are clear. You do not need Google Analytics, ROIStat, CRM and other tools.

Now imagine your business is growing and your advertising budget is already 1 million for various traffic channels. You are also looking at the overall numbers and it seems everything makes profit. However, if you looked “under the hood”, Yandex Direct takes up RUB 300k and brings only RUB 100k in revenue while SEO brings RUB 500k in revenue and takes up RUB 50k.

You disable Yandex Direct, and your revenue drops by 40%. You don’t understand how this happened, because based on the charts, Direct brought in only 5%.

Then you find that some users did not immediately make a purchase. Instead, after some thinking, they returned to the website from search and made the purchase. Therefore, conversions were not attributed to Yandex Direct, but to search, although Direct was the first point of contact. And the list of such reasons could run indefinitely.

Conversions are attributed to email and Yandex Checkout

When we launched our advertising campaigns, the number of conversions, contrary to our expectations, did not grow. However, we saw an abnormal growth in conversions from direct transitions and referral sites.

In some cases during the ordering process, the user is asked to confirm their email and, when you click on the link in email, Google Analytics attributes conversions not to contextual advertising, but to email distribution

We added the utm_nooverride, tag to the confirmation link so that this transition was not counted as a source of conversion.

The second problem had to do with the payment services. During the online checkout, the client switched to a payment service, for example, Yandex Checkout, and when returning to the online store, the session would be terminated and Google Analytics counted users as new visitors. Therefore, the conversion was attributed to payment services and not contextual advertising, which is funny. The problem was solved by technical corrections.

After debugging the analytics system, we turned off advertising campaigns that did not bring conversions and only wasted money, thereby reducing the CPA by 30%.

Dynamic remarketing showed good results. In our experience, this is a must for any online store, It's a method where the content of the ads corresponds to the flypages the user had visited.

Similarly, dynamic ads in Google and Yandex (DSA) showed good performance, but they need to be closely monitored as the traffic they generate is often up to 30% untargeted.

In 6 months since we started our work, we have reduced the cost per lead and almost doubled the number of leads, and these numbers are from correct analytics. It will be more evident through the cost per lead, since the client did not start measuring the revenue from the start.

At the start, we made an online report with breakdown by the main metrics,

and a number of reports by product categories, geography and other.

How to calculate LTV for an online store

The client needs to measure LTV to measure the inflow and outflow of customers for different cities or product categories, but this feature and a number of others is not available within Google Analytics due to its limitations. That is why we configured import of Google Analytics raw data into Google BigQuery

Similarly, we send the order data there as well.

It sounds complicated, but it virtually removes the boundaries on the kind of data analysis you can do. As of writing about the case, we have data for 3 months on contextual advertising revenue for one of the traffic segments. The screen below shows a cohort analysis.

Briefly about the main advantages of Google Analytics + BiqQuery

• If you have loyalty cards, you can link offline and online purchases
• No sampling issues as in Google Analytics
• You can see LTV over several years by city, or product category, or a manager, if you like
• You can reduce the costs for channels that bring repeat customers
• You can store data for machine learning

However, it’s difficult to work with raw data, as each report needs to be built manually.

Conclusion

There are no secrets, we believe this case to be a success for the following reasons.

• The client has already been through hell with cheap contractors and realized that SEO and context advertising are a serious craft and should be treated accordingly
• The client had tried to manage campaigns on their own, which enabled them to “speak a common language” and better integrate contextual advertising into the business.
• The client has a working business model, we just had to structure and fine-tune everything.