1. ARTIFICIAL INTELLIGENCE IN
MARKET RESEARCH
Deep dive // November 24th 2015
Tom De Ruyck
Managing Partner
Steven Debaere
Data Scientist
Discussing 2 case studies
2. ARTIFICIAL INTELLIGENCE
BIG DATA
MACHINE LEARNING
AUTOMATION
PREDICTIVE ANALYTICS….
PATTERN RECOGNITION
CONFERENCE TALK
DEEP LEARNING
| @tomderuyck @steven_debaere | @InSites
4. UMBRELLA TERM
Artificial Intelligence is the broader concept of
machines being able to carry out tasks in a way
that we would consider « smart »
(Bernard Mahr, Dec 2016, Forbes)
| @tomderuyck @steven_debaere | @InSites
5. WE KEEP HYPING IT
massive promotion implies massive adoption?
| @tomderuyck @steven_debaere | @InSites
6. HYPE OR REALITY?
67%
INTERESTING TREND,
TOO EARLY TO TELL OR
23%
GAME CHANGER
(GRIT 2016)
MUCH ADO ABOUT NOTHING
| @tomderuyck @steven_debaere | @InSites
15. TEST
IMPACT
DETECTION ABILITY: 78% (QT) & 71% (QL) ACCURACY
WHAT
A
T
I
FROM REACTIVE TO PROACTIVE MANAGEMENT
SEE WHAT THE MODERATOR CAN’T SEE
MODERATION TIME
PROACTIVE COMMUNITY MANAGEMENT
PREVENTION CAPABILITY: IMPACT CRM APPROACH
| @tomderuyck @steven_debaere | @InSites
16. LEARN
Increase adoption by trade-off
between predictive accuracy,
believability & actionability
Volume and quality
necessary preconditions Prediction model as
pattern reveal tool
Database
Adoption
Insights
PROACTIVE COMMUNITY MANAGEMENT
| @tomderuyck @steven_debaere | @InSites
19. BUILD
A SMART ASSISTANT
INSIGHT ACTIVATION BOT
FIND ME SOME INSIGHTS
ABOUT PACKAGING
Case 3
LET’S MEET THE
CONSUMER
Case 1
SHOW ME THE LATEST
NEW INSIGHTS
Case 2
| @tomderuyck @steven_debaere | @InSites
20. TEST
A
T
I
IMPACT
INSIGHT ACTIVATION: ADOPTION & SATISFACTION
INSIGHTS ANYTIME, ANYWHERE, IN YOUR POCKET
CONSUMER-CENTRIC DECISION MAKING
MARKETING MANAGER TIME
WHAT
INSIGHT ACTIVATION BOT
| @tomderuyck @steven_debaere | @InSites
21. LEARN
In addition to core-
behavior, challenging
to enable small talk
Logic
Map out human behavior
for defining use cases
Relevance
Associating a
personality and
humanizing the chatbot
for increased adoption
Adoption
INSIGHT ACTIVATION BOT
| @tomderuyck @steven_debaere | @InSites
23. ARTIFICIAL INTELLIGENCE IN
MARKET RESEARCH
Deep dive // November 24th 2015
www.insites-consulting.com
Tom De Ruyck
Managing Partner
Steven Debaere
Data Scientist
Discussing 2 case studies
Thank you for
your attention!
Notes de l'éditeur
Current conference talk
Buzzword alert!
What is all this buzz?
Is this a hype?
A hype is ….
Whenever we talk about smart tasks being carried out by the computer, we can use the umbrella term Artificial Intelligence
Because we all have been hyping Artificial Intelligence, this means we have been massively adopting it
Right?
A recent industry report shows that we are not yet in the wide adoption phase:
The 67% of marketers that say AI is an interesting trend, too early to tell or much ado about nothing, allow us to assume they didn’t even start adoption or experimentation.
Let’s hope the 23% put there money where their mouth is..
But let’s pause for a second;
All the conference presentations about the huge possibilities of AI for MR say us that pioneers/trend watchers see something in it?
Numbers show us that the majority of the industry is still awaiting or does not know what to expect
The only way forward is to stop talking about the possibilities, but move to the next phase;
Let’s bring in help Hollywood to insipire us. Shia Labeouf, please, can you give us advice of what our industry has to do with AI?
We need to build, measure & learn
We have to start experimentation to what we can do with it, how it can affect us or how we need to adapt to embrace the benefits?
We, at InSites Consulting already started this experimentation process;
In two AI adoption projects, we have been through the build, measure & learn phase
Today, we are here to share
We adopt AI in two future-proof environments
Insight generation
Insight activation
Why future-proof?
Insight generation: Communities as % or MR budget
Insight activation: Impact gap of insights
Casestudies:
online research communities
Minority report, to predict future participant behavior, in order to prevent destructive behavior from impacting the community
Isight Activation Studio
Galvin, the personal assistant for consumer insight activation
Challenge of Member disengagement for CCB’s
Low quantity if the member does not participate enough in the topics that are organized in the community
low quality if what the member says doesn’t say anything at all.
As a result, if those types of behavior occur frequently, the community may not be able to deliver useful insights anymore.
Therefore, as we want to sustain ongoing ccb’s on the long-term, it’s crucial to battle member disengagement.
The moderator can do a pretty good job, but it still requires extra energy that could be better used for analyses & insight generation motivation
We adopt AI in a community context:
Predict member disengagement using prediction models
Moderatore uses these models to become automatically alerted of expected unconstructive member behavior, so preventive actions can be taken
How does it works?
We construct prediction models using using machine learning on the data-rich environment
We identify patterns in historical data to explain future events.
The more intuitive explanation is that it tried to find habits. Human behavior can be actually quite predictable, this is the same for a community context.
Now, we have our own « minority report » in a community context
where murder is a threat for the world Tom Cruise lives in, member disengagement is the same for moderators & structural ongoing CCB’s
Delete? Enough time?
Now how, can we make this more practical in a community context.
By combining the two dimensions (quantity & quality) & their respective activation levels (low & high), we can come up to a four quadrant framework
This allows you to identify easily the future performance of each member.
Now, we can classify each member into one of the quadrants & see the future of each member
With this project, we aim to maximize capabilities to identify & prevent member disengagement, while maximizing moderator convenience
The impact:
Automational: automatic detection of destructive behavior leads to time reduction in community moderation
Transformational: through the prediction models, we can go from reactive (when it’s already to late) to proactive community management (prevention)
Informational: through the prediction models, subtleties & destructive community patterns arise that lead to destructive community behavior
Process learnings:
Data preparation is time-consuming, but key for prediction quality
Objective is to enjoy benefits of AI in a community context & to maximize adoption;
Therefore, we gave up predictive accuracy through explore other models that are more justifiable & comprehensible for the moderator
Prediction model as pattern reveal tool
Models show that narcissistic moderators & negative community environments increases chances of member disengagement
Challenge of user assistance for Insight Activation Environments
Who is my consumer? Which insights define my consumer?
Insight inspiration for new projects?
The Marketing department overloaded with consumer insights; in meeting & urgent need for insight?
Guiding users through your insight database
Galvin is connected to the insights database, having the right information at hand to help marketers.
It uses Luis AI to understand language
It uses cortona to become an intelligent personal assistent
Both powered by AI technologies
With this project, we aim to maximize user assistance in the insight
The impact:
Automational: because it’s a 100% digital chatbot, the marketing manager’s time can be spend more wisely
Transformational: the chatbot for market research impersonates your personal market research assistent
Informational: access to insights anytime, anywhere, right availble in your pocket