80 Prozent der Conversions über Google Shopping Ads werden von Produkten generiert, die jeweils nur ein bis zwei Conversions aufweisen. In der Masse steckt also großes Umsatzpotenzial. Bid Management ist für die Effizienz der wichtigste Performance-Hebel.
8. SHORT HEAD
LONG TAIL
20% der Produkte generieren 80% des Umsatzes
Anzahl der Produkte
U
M
S
A
T
Z
I
N
€
80% der Produkte generieren 20% des Umsatzes
Klassischer Longtail
9. Longtail:
80%
der Shopping Ad Verkäufe werden von Produkten generiert, die nur
1 Conversion
aufweisen.
Shopping Ads Longtail
12. Problematik
Hoher Aufwand
bei manueller Optimierung
Ereignisarmut
Wenige statistische Daten
Über - oder Untersteuerung herkömmlicher Tools
Zeitverzögerung
aufgrund fehlender statistischer Daten
Herausforderung
13. Ziele
Valide Umsatzprognosen
für Produkte mit wenig statistischen Ereignissen
Frühzeitiges Trennen
von Kostentreibern & Umsatzbringern
Setzen des optimalen CPC
für jedes einzelne Produkt im Sortiment
Ziele - Bidmanagementstrategie
19. €0.32 €0.30
Individueller CPC pro Product ID
€0.33
Empfohlene Accountstruktur
Nike Laufschuh Modell 1
Nike Laufschuh Modell 2
Nike Laufschuh Modell 3
36. Steigerung des Bids bei Shopping Campaigns führt nach vollständiger Präsenz bei produktspezifischen Termen
zu einer semantischen Erweiterung ähnlich des Broad-Matches:
Samsung UE55H6270
0,02 Cent Gebot -> kein Erscheinen bei Suche nach “UE55H6270”
0,08 Cent Gebot -> erscheint bei Suche nach “UE55H6270”
0,20 Cent Gebot -> erscheint zusätzlich bei Suche nach “Samsung Fernseher”
0,40 Cent Gebot -> erscheint zusätzlich bei Suche nach “Fernseher kaufen”
0,60 Cent Gebot -> erscheint zusätzlich bei Suche nach “Fernseher”
Gepflegte Google Shopping-Kategorie: Fernseher, Marke: Samsung, Product Type: Fernseher
Besonderheiten Google Shopping Ads
Vorstellung Jan
Vorstellung Chris
….
We hope you understand us well, despite the accent, and that you can pick up on all the details.
The accent originates from Austria – from Linz to be more precise.
I started the Smarter Ecommerce GmbH (Company) there in 2007. We are a software firm specialized in SEA automation. Since the founding, we have been developing our ecosystem with direct Google API linkage.
The core is our own script language, respectively, domain-specific language (DSL) which we call “smeco-script.” We enable large retailers to have complex campaign generation and management automations for their performance with Google TextAds. These are tailor-made solutions adjusted to the respective data feed structures that lead to advantages in performance and maintenance expense for over 200 accounts.
The founding idea of the company was the intensive utilization of the long tail through text ads. This highly individual automation not only enables the comprehensive scripting of text ad keyword combinations in the long and mid-tail areas, but also the agile management of your campaigns on the basis of changing feed parameters such as prices, stock levels, etc.
We thus create the reflection of tactical sales measures, such as the up-to-the-minute communication of prices for selected product range segments, the separate handling of stock levels, the reflection of sales prices with price savings or the pushing of product novelties in the text ads.
Coming from this history, it is only logical that Google Shopping Ads are a very crucial cornerstone of our work. Through our own work in the management of Shopping Ads for customers and the reports of our customers with their own SEA departments, a very large need crystallized in the last year. There is no bid management solution on the market that can really deal well with the special requirements of Google Shopping Ads.
Therefore, in addition to the Ad Engine for the generation and up-to-the-minute managing of extensive text ad campaigns, we have been testing and further developing new approaches for a bid management specialized in Google Shopping Ads in our accounts since the summer of 2014. As the algorithm became increasingly more accurate at the end of 2014 and the tests ran extremely positively, we released an MVP with the most important functions in March 2015 and are currently developing a comprehensive app that will be launched in June 2015. Let’s now take a look together at what the special requirements of Shopping Ads are.
Start der Shopping Ads im August
Davor in Österreich seit ca 1 Jahr als Product Listing Ads verfügbar
Start der Shopping Ads im August
Davor in Österreich seit ca 1 Jahr als Product Listing Ads verfügbar
Start der Shopping Ads im August
Davor in Österreich seit ca 1 Jahr als Product Listing Ads verfügbar
Start der Shopping Ads im August
Davor in Österreich seit ca 1 Jahr als Product Listing Ads verfügbar
Experiences gained from more than 200 AdWords accounts show that the creation of a product-specific campaign structure can be considered as the clear best practice.
Only in this way can every single product be evaluated and managed according to the actual performance.
On the campaign level, a differentiation can naturally be made according to brands or categories; what is important is the setting of an individual product target per product ID among the campaigns to make each product “assailable.”
Only if each product ID has its own product target, every product can be managed according to the actual performance
Which influence this can have on performance was shown by the previous slide. Working on product groups with averaged metrics is very dangerous at times and simply inefficient on a ROAS basis.
This campaign structure thus constitutes a basis for an efficient management of the long tail in the Google Shopping realm.
For the creation of a product-specific campaign structure, we recommend the direct linkage via the Google API – over Bulk Upload in the Google Interface —> it is very slow, and takes 2-3 hours of effort per week, especially with rapidly changing product ranges.
Let’s get down to the bottom of the matter.
So if one wants to make reliable assertions, respectively, estimates about the conversion rates, you have to work with very high confidence interval, in this concrete example with a confidence interval of 95% – the Pearson-Clopper method – in other words, the assertions allow a 95% certainty.
But if you want to make assertions with a high probability there are really large scope the conversion rates show. Especially products with few events do not allow de facto any concrete assertions.
But as we pointed out at the start, big part of your shopping Ad revenue is created by your long tail – that is, products with more than 2 conversions
For instance, products with 50 clicks and 1 conversion? If a relatively certain assertion is to be made here, it can merely be claimed that the conversion rate lies between 0.05% and 10.65%. An online marketing manager cannot do anything with this estimate.
At the same time you have to handle products with 0 sales as well – to predict whether the product will ever perform is a crucial performance lever.
At 10 clicks and 0 conversions, the conversion rate lies with very high probability between 0 and 30.85%. Again, This assumption is not very useful
As you can see, the prediction of the interesting variables like CR and Shopping Cart Values on the basis of classic metrics such as the conversion and the number of clicks is very difficult and, because of the large corridors, not very helpful in part.
And they leave a lot of space for errors.
How could a problem-solving approach look here?
Ebene Product Item
Ebene: Product Type
In practice, we meanwhile use a catalog of more than 100 different attributes, based upon which we calculate correlations with the product performance.
The starting position is always the historical account data with which we determine the existing correlations. This works through a selection of variables, that is, individual correlations are determined for each account on the basis of the actual data. And only those variables that also show the actual correlations to product performance are included in the model.
the algorithm calculates which correlations exist and here are highly exciting correlations for the attributes, which are not very conclusive at first glance.
We see, for example, in the electronics sector that products whose search queries very often contain numbers show a higher CR than product titles without them.
So in the, several regression analyses with different variables are laid on top of each other - dependent on wicht variables correlate to the product performance
In the concrete example, we see a television model for which we exemplarily illustrate different attribute correlations to the product performance. Here we draw on attributes such as the influence of the brand, the influence of the price, the availability, the average number of orders, the size and the age of the product.
So, several regression analyses with different variables are laid on top of each other, so to say, – always with the goal of calculating a correlation with the shopping basket value and the conversion rate. If this is carried out with the proper variables, a prediction of the product performance is also possible without many events – that is, clicks and conversions.
This is analogous to the example of the correlation of the variables of product price and shopping basket value.
As already mentioned, the variables that impact product performance are different in every account. It is also interesting that the variables can also be expanded by the customer at will. A large online shop for sporting goods has, for instance, conveyed the number of sizes per shoe model for the shoe segment. The background is that a shoe with several sizes in stock shows distinctly higher conversion rates.
In sum, the creation of a product-specific campaign structure and the extension of the evaluation basis beyond classic metrics like conversions, clicks, etc., are considered as problem-solving approaches for a valid and quick adjustment of the long tail.