Funnel based reventure attribution of online canpaigns
The article reviews the problems of static attribution models of online stores transactions and suggests a solution in the form of dynamic attribution model based on customer behavior while customer journey on a website.
Збережено в:
| Дата: | 2016 |
|---|---|
| Автор: | |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут телекомунікацій і глобального інформаційного простору НАН України
2016
|
| Назва видання: | Математичне моделювання в економіці |
| Теми: | |
| Онлайн доступ: | http://dspace.nbuv.gov.ua/handle/123456789/131868 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Funnel based reventure attribution of online canpaigns / V.L. Flaks // Математичне моделювання в економіці. — 2016. — № 3-4(7). — С. 158-166. — Бібліогр.: 4 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-131868 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-1318682018-04-05T03:03:12Z Funnel based reventure attribution of online canpaigns Flaks, V.L. Математичні та інформаційні моделі в економіці The article reviews the problems of static attribution models of online stores transactions and suggests a solution in the form of dynamic attribution model based on customer behavior while customer journey on a website. У статті розглянуто проблеми статичних моделей атрибуції доходів від транзакцій в інтернет-магазині і запропоновано рішення у вигляді динамічної моделі, заснованої на поведінці покупця при проходженні воронки замовлення. В статье рассмотрены проблемы статических моделей атрибуции доходов от транзакций в интернет-магазине и предложено решение в виде динамической модели, основанной на поведении покупателя при прохождении воронки заказа. 2016 Article Funnel based reventure attribution of online canpaigns / V.L. Flaks // Математичне моделювання в економіці. — 2016. — № 3-4(7). — С. 158-166. — Бібліогр.: 4 назв. — англ. 2409-8876 http://dspace.nbuv.gov.ua/handle/123456789/131868 004.023 en Математичне моделювання в економіці Інститут телекомунікацій і глобального інформаційного простору НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| topic |
Математичні та інформаційні моделі в економіці Математичні та інформаційні моделі в економіці |
| spellingShingle |
Математичні та інформаційні моделі в економіці Математичні та інформаційні моделі в економіці Flaks, V.L. Funnel based reventure attribution of online canpaigns Математичне моделювання в економіці |
| description |
The article reviews the problems of static attribution models of online stores transactions and suggests a solution in the form of dynamic attribution model based on customer behavior while customer journey on a website. |
| format |
Article |
| author |
Flaks, V.L. |
| author_facet |
Flaks, V.L. |
| author_sort |
Flaks, V.L. |
| title |
Funnel based reventure attribution of online canpaigns |
| title_short |
Funnel based reventure attribution of online canpaigns |
| title_full |
Funnel based reventure attribution of online canpaigns |
| title_fullStr |
Funnel based reventure attribution of online canpaigns |
| title_full_unstemmed |
Funnel based reventure attribution of online canpaigns |
| title_sort |
funnel based reventure attribution of online canpaigns |
| publisher |
Інститут телекомунікацій і глобального інформаційного простору НАН України |
| publishDate |
2016 |
| topic_facet |
Математичні та інформаційні моделі в економіці |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/131868 |
| citation_txt |
Funnel based reventure attribution of online canpaigns / V.L. Flaks // Математичне моделювання в економіці. — 2016. — № 3-4(7). — С. 158-166. — Бібліогр.: 4 назв. — англ. |
| series |
Математичне моделювання в економіці |
| work_keys_str_mv |
AT flaksvl funnelbasedreventureattributionofonlinecanpaigns |
| first_indexed |
2025-07-09T16:19:51Z |
| last_indexed |
2025-07-09T16:19:51Z |
| _version_ |
1837186935600185344 |
| fulltext |
~ 158 ~
Математичне моделювання в економіці, №3-4, 2016
UDC 004.023
V.L. FLAKS
FUNNEL BASED REVENUE ATTRIBUTION
OF THE ONLINE CAMPAIGNS
Abstract. The article reviews the problems of static attribution models of
online stores transactions and suggests a solution in the form of dynamic
attribution model based on customer behavior while customer journey on a
website.
Keywords: Ecommerce, online store, conversion, cost of customer
acquisition, customer, merchant, revenue, attribution.
Introduction
Ecommerce field is one of the most dynamic segments of Ukrainian and worldwide
economics. In many ways this was made possible due to more efficient processes
including those for customers acquisition. This efficiency is based on the
availability of a large amounts of personalised customer data, especially before
transaction. For instance, data on a specific advertising campaign, that a customer
contacted before transaction or data on the products that interested a customer who
hadn’t placed an order.
In classic brick-and-mortar retail there is a saying that there exist three key
factors of success: “location, location and location”. It is a location of a physical
store that determines the quantity and the quality of potential customers, who can
be attracted by the trade offer of a store.
In internet there is no possibility to place a server on crossroads or another
crowded place. Thus the task of customer attraction in ecommerce is solved by
online marketing. As a result the major part of a gross margin is invested in
marketing of online store. Like in classic brick-and-mortar retail, a big part of gross
margin is invested in stores rent and maintenance. These expenses ensure that a
potential customer gets familiar with a product.
That’s why management of the marketing budget is a key challenge for online
retail.
Glossary for article:
Attribution — a process of defining the impact of interaction, that took place
during a customer journey.
Conversion — user action that took place on a website, usually a transaction.
Retailer — businesses that sell goods directly to individuals.
Conversion value — the value from a user action that was conducted on a
website or in mobile application. For instance, the revenue from an order.
User session — a chain of actions of an exact user, that are combined by a
traffic source and a period of time.
Traffic source — a source that a user had interacted with before a transition
took place. For instance, “google / cpc”.
Ó V.L. Flaks, 2016
~ 159 ~
Математичне моделювання в економіці, №3-4, 2016
1. Business task description
Managing of marketing budget is based on a segments estimation with grouping by
properties of advertisement channels, products and users. To put such a grouping
into practice, the value estimation should be implemented based on unsampled
data, granulated to each transition.
There exist a few peculiarities of the customer behavior that complicate the
business task:
1. Over half of transactions are made after 2+ paid transitions. Thus we need a
model of revenue attribution from each transaction for several transitions.
2. A considerable part of paid visits does not result in transactions. That’s why
it’s impossible to clearly estimate the impact of these expenses on the revenue from
the interaction chains resulted in transactions.
3. In many cases a customer uses several devices. It means that technically it’s
impossible to define the relations between user actions on different devices. Even if
a certain chain resulted in a transaction. At the same time a possibility to connect
interaction chains of authorised users does not solve this task for 100% visits.
4. There is a natural time lag between the moment of costs expenditure for a
visit and the moment of getting revenue from a user, who conducted this visit. The
time lag is caused by a period needed for making decision as for the purchase.
This period can make from a couple of days to several weeks and depends on the
product price. At the same time online store has to make decisions as for budget
allocation as soon as possible, since the delay determines inefficiently expenses
and revenue loss.
2. Limitations of existing models
The most popular models to estimate the efficiency of advertisement channel are
static attribution models. In these models the value of visit is determined by its
position in an interaction chain before transaction.
In terms of a Last Click (Pic. 1) Attribution Model 100% of conversion value is
attributed to the last channel in a user interaction chain.
Picture 1 – Last Click Model
~ 160 ~
Математичне моделювання в економіці, №3-4, 2016
Since this model is the most common, let’s consider its limitations in more
detail.
1. This model does not consider the impact of all the other interactions except
for the last one.
2. ROI is calculated as follows:
, (1)
where the revenue and the expenses are taken for the same period. The problem is a
part of revenue in the beginning of a reporting period is generated due to the
expenses of a previous period. And at the same time the revenue at the end of a
reporting period will be generated in the following reporting period.
3. This approach considers the revenue on the 1st transaction, but ignores the
revenue on the subsequent transactions of this user and the expenses for his/her
retention.
In terms of a First Click (Pic. 2) Attribution Model 100% of conversion value is
attributed to the 1st channel in a user interaction chain.
Picture 2 – First Click Model
In terms of a Linear (Pic. 3) Attribution Model all the channels in a user
interaction chain are attributed the same value.
Picture 3 – Linear Model
~ 161 ~
Математичне моделювання в економіці, №3-4, 2016
3. Attribution based on limitation period
The model is built on time decay (Pic. 4). The closer to the moment of conversion
there is a point of interaction, the more valuable it is considered. In terms of this
model the period of half life makes 7 days by default. It means that an interaction
that took place 7 days before the conversion is twice less valuable than the one that
took place the same day with conversion, and the one that took place 2 weeks
before conversion is 4 times less valuable compared to the interaction of the same
day. Exponential decay is analysed for the whole period of retrospective analysis,
usually for 30 days.
Picture 4 – Attribution based on limitation period
The key limitation of all the static models is that they disregard real impact of
all interactions to passing the customer journey.
Let’s consider an example. For instance there exist a customer funnel including
4 steps:
1. The first visit to a website
2. Visiting a product pagе
3. Adding to cart
4. Transaction
And there are 4 traffic sources:
1. A — a channel through which a user came to a website for the first time
2. R — retargeting
3. E — email marketing
4. D — direct
With the same sequence of traffic sources and visits the different scenarios are
possible (Pic. 5 – 7):
~ 162 ~
Математичне моделювання в економіці, №3-4, 2016
Picture 5 – Customers journey through the sales funnel
Picture 6 – Product page view, add to cart and transaction take place during the last
visit
Picture 7 – Product page view, add to cart take place during the first visit
It is obvious that the impact of a traffic source in each scenario differs, though
the sequence of interactions remains the same.
~ 163 ~
Математичне моделювання в економіці, №3-4, 2016
4. Requirements for the model
Thus we can formulate the requirements for the model of traffic sources estimation:
1. It should consider the impact of each transit in an interaction chain, not only
the last one.
2. The connection between expenses and revenue should be considered for the
users cohorts, not for static periods of time.
3. The value of a visit should not be fixed. It should be determined by the
visit’s impact to customer passing the sales funnel.
4. The expenses spent for visits that did not result in transactions should also be
considered.
5. Solution
The input needed for calculation is the following:
1. User sessions
2. User actions and their priority within interaction chain
3. Values of conversions.
The output of the calculation is the following: conversion value attributed to
each session.
Let’s base the solution on the following thesis:
1. The business objective is to lead a customer through the sales funnel. That’s
why we should estimate not only transaction, but also micro conversions resulted
in transaction.
2. The lower probability to pass the funnel step, the more difficult it is for a
customer to make it. Thus the higher value of a session that resulted in such a
transition.
Let’s use the following method of calculation:
1. Defining user actions in a sales funnel and prioritize them (Pic. 8)
Picture 8 – Steps of the sales funnel
2. Calculating probability of each user action as a proportion of sessions that
resulted in a transition, among those in which the previous transition took place.
Let’s have a look at an example where users are distributed between the funnel
steps in the following way (Tab. 1):
~ 164 ~
Математичне моделювання в економіці, №3-4, 2016
Table 1 – Users share for sales funnel steps
Action Users
Visit 100.0%
Product page 49.0%
Add to cart 7.8%
Checkout 3.4%
Purchase 2.1%
In this case the probabilities of transits to the next step will be as follows
(Tab. 2, Pic. 9):
Table 2 – Probabilities for sales funnel steps
Action Users Probability
Visit 100.0% 100%
Product page 49.0% 49%
Add to cart 7.8% 16%
Checkout 3.4% 44%
Purchase 2.1% 62%
Picture 9 – Passing probabilities of the sales funnel steps
3. Assigning priorities to user actions when attributing values. The prioritization
is based on business objectives (Pic. 10):
Picture 10 – Steps priorities of the sales funnel
4. Defining the value for each user action in accordance with its probability and
assigned priority.
Attributing 100% of value starts from the step with the 1st priority. For the step
with the highest priority the value is calculated as 1 minus probability of transition
to this step. For all the other steps the value is calculated in descending order of
priorities as 1 minus probability of transition to this step multiplied by 1 minus
value attributed for the steps with higher priority (Pic. 11).
http://context.reverso.net/translation/english-russian/in+descending+order+of
http://context.reverso.net/translation/english-russian/in+descending+order+of
http://context.reverso.net/translation/english-russian/in+descending+order+of
http://context.reverso.net/translation/english-russian/in+descending+order+of
http://context.reverso.net/translation/english-russian/in+descending+order+of
http://context.reverso.net/translation/english-russian/in+descending+order+of
http://context.reverso.net/translation/english-russian/in+descending+order+of
~ 165 ~
Математичне моделювання в економіці, №3-4, 2016
Picture 11 – Value distribtion by the sales funnel steps
The result of sales funnel steps’ estimation (Tab. 3):
Table 3 – Value distribtion by the sales funnel steps
Step Users Probability Priority Value
Visit 100.0% 100% 5 2%
Product page 49.0% 49% 2 32%
Add to cart 7.8% 16% 3 25%
Checkout 3.4% 44% 4 3%
Purchase 2.1% 62% 1 38%
5. The value of session is defined as the sum of values for every user action that
took place within this session for the 1st time (Pic. 12).
Picture 12 – Sessions value evaluation based on the steps value
The value of user action is directly proportional to the attributed value and
directly proportional to the difference between 1 and probability of this action.
The attributed action is calculated as the difference between the conversion
value and all the values attributed to actions with higher priority.
If the sampling is statistically valid, the probability of transition from one step
to another has to be segmented between:
– new and returned users;
~ 166 ~
Математичне моделювання в економіці, №3-4, 2016
– depending on the type of a landing page;
– type of traffic source.
Let’s have a look at a formula of the session value in a general way:
(2)
where
,
(3)
= , (4)
conversion value attributed to the exact session;
the minimum step order that has been achieved in this session;
the maximum step order that has been achieved in this session;
probability of the value on step-a;
probability of conversion cancellation;
value of a step-a;
probability of a transit to the exact step of a sales funnel;
function the returns priprity of the step by its order .
Сonclusion
The suggested approach allows to attribute the conversion value in accordance with
the true impact of a traffic source into the profit from a micro conversion. Unlike
static models it allows to adaptively distribute the conversion value. To implement
the suggested solution there can be classified as a limitation a necessity to collect
enough statistic data.
REFERENCES
1. G. Cristal, Ad Serving Technology: Understand the marketing revelation that
commercialized the Internet. – Paperback, 2014
2. M. Jeffery et al., Data-Driven Marketing: The 15 Metrics Everyone in Marketing
Should Know Know. – Hardcover, 2010.
3. Mary E. Tyler, Jerri Ledford, Google Analytics. – Wiley, 2006 г.
4. Forrester eCommerce Forecast:
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
Стаття надійшла до редакції 30.07.16.
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
http://www.forrester.com/US+eCommerce+Forecast+2013+To+2018/fulltext/-/E-RES115513
FUNNEL BASED REVENUE ATTRIBUTION
OF THE ONLINE CAMPAIGNS
Introduction
5. Solution
The input needed for calculation is the following:
Сonclusion
|