German Fashion eCommerce Case Study: How I Used Custom Bidding in Display & Video 360 to Generate More Revenue From the Same Media Investment

 

Revenue +32% | Orders +21% | Average Order Value +9% | Return on Advertising Spend +31%

For a multi-brand fashion eCommerce retailer selling primarily in Germany, paid media was already operating at significant scale.

The business was investing approximately €2.5 million per month across paid media, with around €1.3 million running through Display & Video 360.

The existing Display & Video 360 campaigns were already generating substantial sales. Automated bidding was in place and Custom Bidding was not part of the existing setup.

So this was not a case of fixing campaigns that were fundamentally underperforming.

The opportunity came from looking deeper at what the retailer was asking the bidding system to optimize toward.

Customers were generating meaningful signals throughout the purchase journey. They were engaging with products, adding products to baskets, progressing through checkout and eventually purchasing at significantly different transaction values.

The existing conversion-focused approach was not making full use of this combination of customer progression signals and transaction value when determining the expected value of individual impression opportunities.

That led to the hypothesis behind the project:

Could the retailer's own conversion and revenue signals help Display & Video 360 make better bidding decisions and generate more sales revenue from the media investment already available?

That became my Custom Bidding use case.

Case Study Snapshot

Area	Business Context
Business	Multi-brand fashion eCommerce retailer
Primary Market	Germany
Additional European Markets	Austria, Netherlands and France
Products	Women's and men's fashion, footwear, bags, accessories and premium/lifestyle brands
Product Price Range	Approx. €30 to €500+
Average Order Value Before the Test	Approx. €130
Core Customer Base	Fashion shoppers, primarily 18 to 44
Audience Environment	Prospecting, product-interest audiences, existing customers and lower-funnel audiences
Business Model	Transactional eCommerce
Overall Paid Media Investment	Approx. €2.5M per month
Display & Video 360 Investment	Approx. €1.3M per month
Custom Bidding Before the Initiative	Not in use
Initial Custom Bidding Test Scope	Approx. €200K monthly media investment
Scaled Custom Bidding Scope	Approx. €750K monthly media investment
Initial Evaluation Period	Approx. 6 to 8 weeks after training and stabilization
Primary Business Objective	Generate more sales revenue without a corresponding increase in media investment
Primary Business Metrics	Sales revenue, orders, Average Order Value and Return on Advertising Spend
Supporting Diagnostics	Add to basket, checkout progression, conversion rate, device performance and customer mix
Buying Platform	Display & Video 360
Measurement	Campaign Manager 360 and Floodlight

Where the Opportunity Came From

The existing campaigns were converting.

But conversion volume alone did not fully represent the retailer's economics.

Consider three completed transactions:

Order A → €65

Order B → €180

Order C → €410

All three represented one purchase.

From a conversion-count perspective:

1 = 1 = 1

From the business perspective:

€65 ≠ €180 ≠ €410

There was also useful information appearing before the final transaction.

A customer viewing products was behaving differently from someone adding products to a basket.

Someone reaching checkout was demonstrating stronger commercial intent again.

The customer journey broadly looked like:

Product Engagement

→ Add to Basket

→ Checkout

→ Purchase

→ Sales Revenue

The opportunity was to use meaningful information from that progression while keeping completed transactions and actual revenue firmly anchored as the commercial outcome.

Why I Chose Custom Bidding

A standard conversion-focused bidding strategy can optimize effectively toward the conversion objective it receives.

Value-based bidding can go further and optimize toward conversion value.

My requirement went beyond simply telling the platform:

Maximize transaction value.

I wanted the bidding logic to learn from multiple meaningful conversion outcomes across the purchase journey, while still using completed transactions and actual sales revenue as the strongest expression of business value.

That was where Custom Bidding became relevant.

The strategy could incorporate the retailer's own Floodlight conversion activities and sales revenue into a custom definition of value.

The difference was important.

I was not simply changing the optimization target from:

Conversions

to:

Revenue

I was creating a richer bidding objective around:

Meaningful Customer Progression

•  

Completed Transactions

•  

Actual Transaction Value

The Hypothesis

The working hypothesis was:

If Display & Video 360 could learn from meaningful conversion progression and transaction value, it could identify impression opportunities with greater expected commercial value and allocate the existing media investment more effectively.

Conceptually:

Historical Impression Data

•  

Floodlight Conversion Outcomes

•  

Transaction Revenue

↓

Custom Bidding Model

↓

Predicted Impression Value

↓

Auction-Level Bid Decision

↓

Orders

↓

Sales Revenue

That gave me a clear hypothesis to test rather than simply introducing a new bidding feature into the account.

My Role

I owned the strategy from the business problem through to evaluation and scaling.

My involvement covered:

Identifying the limitation in the existing optimization approach

↓

Developing the Custom Bidding hypothesis

↓

Selecting the conversion and commercial signals that should inform the strategy

↓

Reviewing the Floodlight and transaction-revenue measurement requirements

↓

Defining the Custom Bidding logic

↓

Selecting the initial campaign scope

↓

Structuring the controlled comparison

↓

Monitoring algorithm training and impression scoring

↓

Evaluating the commercial results

↓

Scaling the strategy across additional eligible Display & Video 360 activity

The important part for me was keeping the project anchored to the commercial objective throughout.

Getting the Measurement Foundation Right

Before building the algorithm, I reviewed the Floodlight setup supporting the relevant eCommerce activity.

The core signals were:

Customer Action	Role in the Strategy
Product Engagement	Supporting indication of meaningful product interest
Add to Basket	Stronger indication of purchase intent
Checkout	High-intent commercial progression
Purchase	Primary completed commercial outcome
Sales Revenue	Actual monetary value of the transaction

For completed purchases, the Floodlight Sales activity captured transaction-level information including revenue.

This mattered because the retailer sold products across a wide price range.

A customer might purchase:

€45 accessories

or

€140 footwear

or

€280 premium fashion

or build a:

€500+ multi-product basket

A bidding objective based only on transaction count loses that variation.

How I Defined Value

I avoided creating arbitrary monetary values for actions that did not directly generate revenue.

I did not decide that:

Product View = €2

Add to Basket = €10

Checkout = €30

Those numbers would create an artificial economics that did not exist in the retailer's actual business.

Instead, I separated the signals according to their role.

Supporting Conversion Signals

Product engagement

Add to basket

Checkout

These helped provide information about meaningful customer progression.

Primary Conversion Signal

Purchase

This represented the completed commercial outcome.

Commercial Value Signal

Floodlight Sales Revenue

This represented what the transaction was actually worth.

The Custom Bidding objective could then learn from those outcomes without pretending that every behavioural event had a directly attributable euro value.

Why I Did Not Use Every Available Signal

The retailer had far more behavioural data available.

Homepage visits.

Category views.

Product views.

Wishlist interactions.

Session depth.

Time on site.

Basket activity.

Checkout activity.

Purchases.

Revenue.

Adding every measurable event would have made the algorithm more complicated without necessarily making it more commercially useful.

I concentrated on signals with a credible relationship to purchase progression.

That kept the strategy aligned around:

Intent

→ Transaction

→ Revenue

rather than rewarding activity simply because it was measurable.

How Custom Bidding Changed the Auction Decision

Custom Bidding did not mean manually assigning different bids to different customer actions.

I was not creating rules such as:

Basket user → bid €8

Checkout user → bid €15

Previous purchaser → bid €20

Instead, I defined the outcomes that represented value to the retailer.

Display & Video 360 used historical campaign data and machine learning to understand how previous impression opportunities related to those outcomes.

The operating logic was:

Historical Eligible Impressions

•  

Floodlight Conversion Outcomes

•  

Transaction Revenue

↓

Model Training

↓

Relationships Between Impression Signals and Commercial Outcomes Learned

↓

New Eligible Impression Opportunity

↓

Predicted Impression Value

↓

Bid Decision

That moved the retailer's own business information closer to the individual auction decision.

What That Meant in Practice

Imagine two eligible impression opportunities became available.

Historical data associated characteristics surrounding Opportunity A predominantly with relatively shallow customer behaviour and limited downstream commercial value.

Patterns surrounding Opportunity B had historically been associated more strongly with purchase progression, completed transactions and greater sales revenue.

The model could therefore produce different predicted values for those opportunities.

Lower Predicted Commercial Value

→ lower bidding priority

Higher Predicted Commercial Value

→ higher bidding priority

Display & Video 360 could then use those predicted values when determining the bid.

I did not need to manually decide what an individual customer or impression was worth.

The model learned from historical outcomes and applied those relationships across future eligible impression opportunities.

Why Revenue Changed the Quality of the Optimization

Consider two customers who both purchased.

Customer A

1 order

€80 transaction value

Customer B

1 order

€350 transaction value

A conversion-count objective records:

1 purchase vs 1 purchase

The retailer receives:

€80 vs €350

That distinction becomes especially important at scale.

The business was not simply trying to maximize the number of transactions.

It wanted to maximize the commercial output of the available media investment.

Starting With Approximately €200K, Not €750K

The wider paid media operation was approximately:

€2.5M per month

Approximately:

€1.3M per month

was running through Display & Video 360.

I did not immediately move most of that investment onto an unproven Custom Bidding strategy.

The initial controlled scope represented approximately:

€200K in monthly media investment

I selected eligible campaigns where:

Floodlight measurement was reliable

transaction revenue was available

historical conversion volume was sufficient

campaigns represented meaningful commercial scale

the existing bidding approach provided a useful comparison

The rest of the Display & Video 360 operation continued under the existing bidding approaches.

Creating a Meaningful Comparison

The objective of the initial test was specific:

Could Custom Bidding produce greater commercial output from a broadly comparable level of media investment?

I therefore kept the major surrounding variables as stable as practically possible.

That included:

Audience strategy

Inventory approach

Creative strategy

Frequency controls

Conversion configuration

Budget levels

This was still a live commercial environment, so perfect laboratory isolation was impossible.

But I wanted to avoid a situation where audiences, creatives, inventory and bidding all changed simultaneously and any revenue movement became impossible to interpret.

Accounting for Fashion Seasonality

Fashion eCommerce introduces another important variable:

Demand changes.

Promotional periods, seasonal collections, discounting, Black Friday, Christmas and clearance activity can materially affect:

Conversion rate

Orders

Average Order Value

Revenue

and therefore:

Return on Advertising Spend

I did not want a promotional spike to be mistaken for a bidding improvement.

The comparison therefore focused on periods with reasonably comparable trading conditions and avoided using exceptional promotional events as the primary basis for judging the Custom Bidding strategy.

Where commercial conditions differed, I treated that difference as part of the performance analysis rather than automatically attributing the movement to bidding.

Training and Validating the Algorithm

Before expanding the strategy, I checked how the algorithm was scoring eligible impressions.

I wanted to understand:

whether sufficient impressions could be scored

whether the algorithm was generating meaningful positive scores

whether the score distribution behaved sensibly

whether the underlying conversion signals were being received correctly

The retailer's scale was useful here.

The campaigns already generated substantial impression and conversion volume, giving the model enough historical information to learn from meaningful commercial outcomes.

The progression was:

Business Hypothesis

→ Measurement Validation

→ Custom Bidding Configuration

→ Impression Scoring Validation

→ Training

→

Controlled Live Activation

The Initial Evaluation Period

I did not judge the strategy from the first few days.

The model needed time to train, activate and stabilize before the commercial comparison became meaningful.

The initial evaluation therefore covered approximately:

6 to 8 weeks following training and stabilization

During this period I monitored both commercial outcomes and the diagnostic signals underneath them.

What I Watched

Business Metrics

Sales Revenue

Orders

Average Order Value

Return on Advertising Spend

These determined whether the strategy was commercially successful.

Diagnostic Signals

Product engagement

Add-to-basket activity

Checkout progression

Conversion rate

Spend and delivery

Device performance

Customer mix

These helped explain why the business metrics were moving.

For example, if add-to-basket activity increased substantially while purchases remained flat, that would suggest the additional intermediate activity was not translating into commercial value.

If orders increased but Average Order Value deteriorated significantly, I would investigate whether the strategy was generating more lower-value transactions.

New vs Returning Customers

I also monitored the customer mix.

This mattered because an apparent improvement in Return on Advertising Spend could come from concentrating more heavily on customers who were already highly likely to purchase.

That might produce attractive short-term reporting while contributing less incremental customer acquisition.

I therefore looked at the relationship between:

New Customers

and

Returning Customers

alongside the primary commercial metrics.

The objective was to understand whether the strategy was changing where the revenue came from, rather than looking only at the headline revenue number.

Device Performance

I also reviewed performance across device environments.

Fashion journeys frequently move between:

Mobile

Desktop

and other touchpoints before purchase.

A strong aggregate result can hide very different behaviour underneath it.

Breaking down performance by device helped identify whether the improvement was broadly distributed or being driven disproportionately by one environment.

Again, this was diagnostic rather than the ultimate success measure.

The business outcome remained sales revenue.

From Initial Test to Scaled Deployment

Once the initial Custom Bidding scope had accumulated enough data and demonstrated stronger commercial performance, I expanded it across additional eligible Display & Video 360 activity.

The progression was approximately:

Existing Automated Bidding

↓

~€200K Initial Custom Bidding Scope

↓

Training + Stabilization

↓

6 to 8 Week Commercial Evaluation

↓

Performance Validation

↓

Expansion Across Additional Eligible Campaigns

↓

~€450K Monthly Custom Bidding Scope

↓

Further Commercial Validation

↓

~€750K Monthly Custom Bidding Scope

The €750K therefore represented the scaled state of the strategy, rather than the starting point.

Custom Bidding earned additional budget responsibility through performance.

Why I Stopped at Approximately €750K

The retailer had approximately €1.3M per month running through Display & Video 360.

Custom Bidding ultimately covered approximately €750K.

I did not force the remaining investment into the same bidding approach.

Different campaigns served different objectives.

Different inventory environments had different requirements.

Not every campaign had the same conversion characteristics or data conditions.

And Custom Bidding was not applicable to every type of Display & Video 360 inventory.

The final setup therefore looked broadly like:

~€1.3M Monthly Display & Video 360 Investment

↓

~€750K Custom Bidding

•  

~€550K Other Appropriate Bidding Approaches / Campaign Requirements

The bidding strategy followed the business and campaign requirement.

The Commercial Outcome

After training, stabilization and scaling across the eligible campaign scope, the commercial performance improved materially against the comparable previous approach.

Revenue +32%

The Custom Bidding scope generated substantially more sales revenue from broadly comparable media investment.

Orders +21%

More completed transactions were generated.

Average Order Value +9%

The average value of completed transactions increased.

Return on Advertising Spend +31%

The retailer generated materially more sales revenue for every advertising euro invested.

The Results Together

Business Metric	Result
Media Investment	Broadly stable
Orders	+21%
Average Order Value	+9%
Sales Revenue	+32%
Return on Advertising Spend	+31%

The relationship between the commercial metrics also made sense.

Orders +21%

combined with:

Average Order Value +9%

produces approximately:

Revenue +32%

Mathematically:

1.21 × 1.09 = 1.3189

or approximately:

+31.9% revenue growth

With media investment remaining broadly stable, the revenue improvement translated into a similar increase in Return on Advertising Spend.

Before and After

Before

Existing Automated Bidding

↓

Established Conversion Objective

↓

Campaign Optimization

↓

Orders

↓

Revenue

After

Meaningful Floodlight Conversion Signals

•  

Transaction Revenue

↓

Custom Bidding

↓

Machine-Learned Impression Value

↓

Auction-Level Bid Decisions

↓

More Orders

•  

Higher Average Order Value

↓

+32% Revenue

↓

+31% Return on Advertising Spend

That was the commercial reason for scaling the strategy.

What Made the Difference

The outcome came from connecting several pieces that already existed inside the retailer's ecosystem:

Reliable Floodlight measurement

•  

Meaningful purchase-progression signals

•  

Accurate transaction revenue

•  

Sufficient historical campaign data

•  

A commercially relevant Custom Bidding objective

•  

Controlled testing

•  

Training and stabilization

•  

Commercial evaluation

↓

Better-informed auction-level decisions

The platform provided the machine learning and bidding capability.

My job was to determine what business problem we were solving, what signals deserved to influence the algorithm, how the strategy should be tested and whether the resulting performance justified scaling it.

The Business Impact

The retailer started with an established Display & Video 360 operation that was already generating substantial sales.

Custom Bidding was not being used.

I identified an opportunity to make the retailer's own conversion and revenue information more influential in auction-level decision-making.

The strategy started with approximately:

€200K per month

It was trained, evaluated and compared against the existing approach.

Once the commercial improvement became clear, I expanded it to approximately:

€450K per month

After further validation, it reached approximately:

€750K per month

The scaled commercial outcome was:

Broadly Stable Media Investment

→ Orders +21%

→ Average Order Value +9%

→ Sales Revenue +32%

→ Return on Advertising Spend +31%

For me, that was the real value of Custom Bidding in this case.

It allowed the retailer's own definition of commercial value to influence individual media-buying decisions at scale.

And once that happened across hundreds of thousands of euros in monthly investment, improvements at the auction level translated into a materially stronger business outcome.