AI-Driven Text Coding
…in the Real World
www.digitaltaxonomy.co.uk
1
We make smarter software for MR
• Automated coding of openended text answers
• Unique “blended” model
combines multiple AI
techniques
• Better productivity, speed and
quality
•
ETL for mere mortals!
•
Automate repetitive data processing
tasks with a simple drag and drop
workflow
•
•
Fully customisable and extensible
Extensive data transformation tools
2
Wouldn’t it be great…
“big screen amazing sound system”
“Large screen and good sound systems”
“Louder noise, bigger screen, more immersive”
“The occasion, sound, and screen size”
“The sound is much better and the wide screen is great”
“Larger screen and atmosphere”
Verbatim comments
Insights
3
CodeIT allows flexibility
across requirements and
methods
Project Cost
Low
High
Consistent
Low
Requirements
Code Frame Specificity
High
Directional
Specific
The Real-World Is Complicated
Traditional
Manual
Coding
Tracking
Segmentation
Profiles
Repeatable
NPS
Csat
Ad/Product Test
Ad Hoc
Attitudinal
In-Survey
Fast
Text Analysis
Process Time
Slow
AI
Methods
Manual Coding
4
The Codeit “Stand-Alone” Approach
5
•
Fast Startup
•
Multi-User
•
Text Analysis & AI
•
Flexible Tasks
•
API Integration
Databases
Data Collection
Reporting
Vendors
The Codeit “Blended AI” Approach
7
Using Historical Data
Training Data
“big screen amazing sound system”
“Large screen and good sound systems”
“Louder noise, bigger screen, more immersive”
“The occasion, sound, and screen size”
“The sound is much better and the wide screen is great”
“Larger screen and atmosphere”
Verbatim comments
1.
2.
3.
4.
5.
Atmosphere
Bigger screen
Better Sound
Immersive
Night out
17%
100%
67%
17%
0%
Coded output
8
Using Historical Data
Training Data
“big screen amazing sound system”
“Large screen and good sound systems”
“Louder noise, bigger screen, more immersive”
“The occasion, sound, and screen size”
“The sound is much better and the wide screen is great”
“Larger screen and atmosphere”
AI Autocoded
Up to 60%
AI Assisted Coding
Up to 30%
Uncoded
Up to 40%
Verbatim comments
9
sample studies
drugstoretracker–4questions
# of responses
actual hours
estimated manual hours
improvement
wave 1
11918
24
125
523%
wave 2
11996
30
126
421%
wave 3
12000
34
126
372%
wave 4
12028
35
127
362%
47942
123
505
410%
total/average
financialtracker–1question
# of responses
actual hours
estimated manual hours
improvement
wave 1
3108
24
52
216%
wave 2
3543
19
59
311%
6651
43
111
258%
total/average
10
operational efficiencies
Human
Coding
Automate
Coding Rate
50-75 responses/hour
40-80%
Coding
Speed
300% Improvement
Timing
50-80% Reduction
Costs
30-60% Reduction
11
What next for AI in the Survey World?
• “Life begins at a billion examples”
• Current training data and AI model is siloed
• Quantity of data in each silo is relatively small
12
What next for AI in the Survey World?
• Codeit architecture can “pool” data to enhance AI training
• Pooling data within a company, across companies or sector-wide
would turbocharge AI for Market Research
13
Conclusions
• It is a mistake to look for a single “silver bullet”
• The best results are achieved by blending techniques to suit the task at hand
• It is a mistake to think of AI as simply replacing humans.
For now, AI is optimized if it is used to assist humans rather than trying to
do everything for them.
• This “new wave” of AI is only 5 years old … Plenty of room for further
innovation in models and techniques
14
Coding Methods
• Sequential
• Block
• Guided AI/Machine Learning
• Text/Sentiment Analysis Segments
• Brand Coding
Sequential Coding
Block Coding
Guided Sequential Coding – Automated AI
Guided Block Coding – Automated AI
Segment Coding – Text Analysis
Code Frame Auto-Generate
Segment Coding – Text Analysis
Automated
• Topics
• Sentiment
Brand Coding – Code Frame Auto-Generate
•
Secure In-Network
•
Manage Sample
•
Transform/Combine
•
Text Analysis
•
Complexity
•
Non Tech Users
Request more information