How Incuda builds user journey models with Snowplow

© incuda GmbH
incuda BI for multi-channel commerce
data-drive your business.

incuda BI for multi-channel merchants
data-drive your business: One better Decision every day © incuda GmbH . 2
we track
behaviour, interest & profitability
of >200 mln. persons
from over 100 countries …
… on a daily basis
Founded 2012, Munich & Düsseldorf. Partners for strategy, business consulting, technology & implementation.
Market focus: E-commerce & multi-channel merchants, international & multi-brand, yearly revenue 20+ mln. €.
Mission: Provide a more mature & powerful BI platform in time of increasing competition.

Agenda
describe an advanced user journey model
see how a journey-based view helps to optimize
- Marketing Performance
- Relationship Marketing
- Product Performance
understand why Snowplow plays a critical role to achieve overall success.
discuss your ideas and questions, ask anytime

Measuring Success: User Journeys
User Journeys are an excelent tool to measure impact of offers and ad items in a
highly parallel marketing environment.
We focus on Journeys which include digital channels, Snowplow is our tool for
collecting information and allows us to build high-quality journeys on a granular
level.
approach description devices channels
Keywords,
banners View/click ad item, Last Click conversion
single single
Campaigns Response across time and channels single multiple
Journeys Contacts over time and conversion for next order multiple multiple
Cohorts mid-/long-term customer lifetime & value multiple multiple

Why is Journey modelling difficult?
Good customers on average use
− 5 marketing channels,
4 contact devices,
− 1.2 customer IDs,
− 1.3 email addresses.
Over 80% of repeat buyers use more
than 1 contact device.
More than 80% of marketing cost &
journeys are on non-buyers.Number of devices per repeat buyer within 12 months.
Impact on:
- Marketing Performance: different devices used during information / conversion phase
- 360° user view on buyers and non-buyers
- Buying & interest behaviour for different product categories, brands, price ranges

What happens if you are working with raw data?
Raw data experience:
- you see many very short buyer journeys
- mobile journeys with clear product interest but no conversion
- buyers visit at other time of day than non-buyers
Our learnings:
− most products are viewed before ordered, but often not on
the same device.
− People use different devices for information phase (mobile,
work, second-screen) and for orders (desktop, home).
− People use different devices during the day
− Family & friends use the same desktop device (cookie ID)
− cross-device detection happens over time,
requires„automatic“ updates of the user journey (ongoing &
historical data)

Process & Application layers provide interaction data
Campaigns & Offers Touchpoints Customer transaction
Order intake
Revenue
Cancel & Returns
Cogs, Pick & Pack
Raw margin, PC-I
Marketing cost,
Commissions, ..
PC-II
Werbemittel-Detailebene Buyer & Nonbuyer contacts Transactions & Customers
ad media
ID
ad media
ID
transact
ID
transact
ID
…
Profit-driven performance marketing
Conversion; Cost per Offer / Ad item
Traffic acquisition
Cost of traffic, Conversion,
Repeat traffic, …
- Sequence of channels
- Conversion rates
- User Lifecycle
CRM outbound
marketing
many few one

Layers are woven into a „Data Network“: more = higher density
Desktop PC (Cookie)
Mobile
Workplace PC
visit
visit
visit
visit
visit
Email
Call Center
Online-Shop
Customer ID
Order ID,
Product data
Catalog, Ad-ID,
Offline Advertising
Contact data
(email, phone)
Order, Case ID
Email click
Email click
Channels Touchpoints Customer transaction
Tracking tools Channel backend data Data Warehouse / ERP
Subscribe
Log-in
Order
Campaign
Email open Subscribers
call

Case Study: Journey length, based on User Consolidation
Case: established international Online Retailer, comparison based on German shop
− Scenario A: modelling by established Journey agency, based on online data
− Scenario B: enhanced modelling based on online & inhouse data (incuda)
Quantitative findings:
− 50% of Buyer Journeys in scenario A with only
1 or 2 contacts (compared to 32% Journeys B)
− Scenario B shows over 2 times more Journeys
with 7+ contacts
Qualitative findings (based on analysis of examples):
Journeys show a more consistent approach to
conversion (product information before orders)
Number of user interactions more in-line with experience of business experts
Validated responses from outbound communictaion confirmed device usage

Case Study (cont.)
Impact:
More complete description of decision funnels, better understanding of channel
interaction
„longer“ journeys provide more opportunities for conversion impulses
More correct allocation of marketing cost to conversion
Why worked approach B better?
- integration of multiple devices in one journey
- integration of more touchpoints (offline, offsite)
- access to inhouse data
- configuration of technical processes to provide required data
- full utilisation of webtracking platform (tracking, parameterisation, cookie types)

Our Learnings
Integrated Journey modelling impacts:
- Marketing Performance: different devices used during
information / conversion phase
- 360° view on buyers & non-buyers, Lifetime estimates and
opportunities for CRM activities
- Buying behaviour for different product categories, brands,
price ranges
Tracking of contact history should include as many touchpoints
as possible. Owned & offline channels (email, print, co-
operations, etc.) improve metrics for ongoing user activity.
Tracking platform must have all the flexibility which you will
need in the future ; blackbox approach might be risky.
We like Snowplow with the Open Source approach, transparent
source code and architectural flexibility.

User – Device detection & consolidation
Users can be merged at any time; automated integration of contact history and
user profile data is required for a performant platform management.
Level 1 Secure matching
- Matches devices, accounts and contact addresses through
events and data that clearly identify a user
- Examples: Orders, Email address, log-in to user account
Level 2 High-probability matching
- Matches devices, accounts, contact data, and other information through events and
data that identify the user of the device with a high probability
- Mapping is reversible in case problems are detected
- Examples: Email click-through, contact information and identifiers
Level 3 Post-technical validation
- Confirm or reject matches, based on functional rules (abstraction from technical level)
- Complete control to analyse & overrride single consolidations, DQ checks
- Examples: Manage household users, friends sessions, etc.

campaign
contact
campaign
contact0
User consolidation: examples for consolidation rules
User
Master User
User Response (e.g. Snowplow online visit)
new user devices known user devices
campaign
contacts
user contact
data
log-in to
user account
subscribe to
service / email
referrer = email
click-through
order (incl.
guest orders)
customer
service request
− initial = 1:1
− Data Load
consolidation
= 1:N
user account
Level 1 consolidation „secure match“Level 2 consolidation „high probability“
− initial = 1:1
− Logical matching rules
− Batch consolidation
on earliest user ID
Post-technical
validation
Post-technical
validation
Functional
validation
Level 3 consolidation „post-technical validation“

Our Learnings
User consolidation needs to be 95% precise, otherwise you
cannot use it for personalisation / targeting and you loose a
big part of the business value.
Transparency on single cases and option to change every
consolidation is critical!
Use tagging & cookie features to track across (sub-)
domains and brands. Use custom fields for additional IDs.
At the same time, make sure that you retain a precise view
on transactions (shipping address, email address per order,
etc.). Otherwise you will run into limitations for
transaction mails.

How we build dynamic Attribution curves
When you have the base data in place, calculation of
journeys is straight forward:
- consolidate contact history & group by conversion events
- update journeys every time a user consolidation gets
updated
- Calculate per contact & single admedia item (detail level)
- Take care of system performance
From the feature list:
− user defined curves; bath-tub, dynamic u-curve; channel
boost factors (DTI, TV, …)
− age of contact relative to journey end
− duration between contacts in journey
− onsite engagement
− early timeout predictor for ongoing business
− compare different models, champion challenger approach

Understanding Journey pattern
The structure of Journeys first gives you an
insight into the usage of you contact channels
and ad items.
The sequence of channels used helps to
understand typical acquisition and conversion
scenarios for new users, active customers, top
buyers, etc.:
- Sequence of marketing channels
- Sequence of device types (mobile,
desktop,…)
- Sequence of daytime hours for information
and conversion
On top, the main impact of an offer or ad item
during the AIDA decision funnel is measured:
some offers have a low conversion, but a high
acquisition or re-activation value.

Our Learnings
Journey structure on user base gives you a first
segmentation of users who behave in a mono-like way
(strong preference for contact channel, device, daytime
hour) and other users with more flexible behavior.
Journey structure shows, if information phase and
conversion happen on the same timeline: which users buy
directly, which have a time-related pattern (e.g. buy in the
evening from home, order on weekend)
AIDA modelling gives you additional metrics (activation,
reactivation) for performance tuning.

Integrated Marketing Calendar
A broader integration of touchpoints
leads to better journey data: include all
marketing & sales activities and relate
them to user contacts.
− Add campaign metadata (clear names)
and cost figures.
− Break ad cost down to single contacts.
− Add a meta layer to connect between
cost reports and click-data tagging.
A standardised description of marketing
activities & cost is the basis for
- flexible and automated calculation of
user journeys
- dynamic attribution modelling.

Marketing performance: manage on ad-item level & profitability
Measure financial success for single ad items.
Define success metrics according to your
business model, e.g. net revenue or profit
contribution instead of KUR.
Automate baseline business and push financial
metrics to tools for Autobidding, campaigns,
email, sourcing, ...
Compare metrics between time periods and
understand changes on different levels (traffic,
advertising, visitor structure, product quality,
return rates) and their impact on marketing
performance.

Customer understanding: User Profiling & Segmentation
Profile user segments based on all data available:
- navigation on the shops
- marketing channel and device usage
- product & category interest
- buying patterns, including revenue, returns,
profitability
- geography, gender, age, statistical data on
households, income, education, etc.
Customer Understanding is based on what users buy,
what they are interested in, and what they are not
interested in…
Detailed click data enables data feeds to support
CRM, Lifecycle alerts and Outbound Segment
marketing.

Our Learnings
Design advertising channels in a consistent way:
- inbound vs. outbound marketing
- direct vs. indirect channels (TV)
- cost models (CPM, CPC, CPO, CPL)
Link marketing cost to single contacts, not orders. Key measure is
the cost for a given contact.
Snowplow tracks campaing information on ad-item level; do not
track on aggregated campaigns!
Use CPM, CPC, KUR? Optimise according to your business model!
e.g. net revenue, profit, or your custom metrics.
Make sure your tracking tool supports flexible mapping of click
data information to marketing cost reports.
Understand your users based on what they do (buy, view) and
what they don‘t do. (requires a complete user view!)

Products meet Sales: the “Product Journey”
Different types of products can have different roles:
from building your brand to attracting visitors and
finally converting into revenue, not all products
behave in the same way.
But at the end of the day, product stock should be
sold out or at leas minimised. And higher demand
should be identified as early as possible.
The „product journey“ identifies interest levels, leads
and prospect buyers for individual articles:
- for onsite, offsite, aftersales channels
- for buyers and non-buyers
- for converted and dropped articles
- based on net revenue & profit contribution

Tracking product interest
Product orders are available from backend systems. The major data source for
Product interest data is the tracking tool.
We use Snowplow tracking to gather event detail data
- Product views
- Basket add or remove events
- Basket convert
- Product impressions on home page, overview pages
- Product detail information from filter, search, etc. events on detail level (large
amount of tracking data!)
- other events, like registrations, service requests, …
The flexibility and scalability of snowplow allows us to get all interaction data we
require and add it to the users contact history.
The contact history allows us, to measure effects over time, e.g. out-of-stock &
purchase probability.

Matching click events to marketing effort and conversion
Product profitability is usually
measured on converted orders.
Brands and categories attracting
traffic but with lower conversion are
undervalued with this “last click” logic
(same last click in online marketing ).
The “product journey” maps product-
or service-related events into the
onsite journey and attributes
marketing cost and journey revenue
to each event.
This gives a more balanced view on
the contribution of products and
brands to conversion (category and
brand management).
Weighting of product views within visits (same color).
All events of a visit (same colour in chart)
are weighted according to the attribution
curve (acquisition-based u-curve).
Event types (view, add, buy) are weighted
according to their stage in the conversion
funnel.

Products meet Sales: the User- and the Product Journey
The Product Journey shows you the attributed value each product on your sales channels
contribute to final sales.
Not only Products with good “last-click” behavior are rated highly, but also
- Products, categories and brands which acquire new customers
- Products, categories and brands which generate repeat visitors, e.g. from Email
- Lost opportunities per product and category, linked to user profiles
Marketing campaigns lead the users to your sales channels, use the product journey to
optimize your onsite conversion.

How Incuda builds user journey models with Snowplow

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to How Incuda builds user journey models with Snowplow (20)

Recently uploaded (20)

How Incuda builds user journey models with Snowplow