1 of 57

DECODING DATA WITH AI

2 of 57

Brenda Gianiny

DECODING DATA WITH AI

Axis Research

@BrendaGianiny

George Nassar

Public Opinion Strategies

@GeorgeNassar

Russ Rampersad

Catalist

@russicorn

Ryan Bonifay

ColdSpark

@RBonifay

Veena Chandran

Amazon Web Services

MODERATOR

3 of 57

AN INTRODUCTION TO AI

Veena Chandran, Senior Solutions Architect

Amazon Web Services

4 of 57

RAPID-FIRE AI/ML TERMINOLOGY

5 of 57

How many of these terms do you recognize?

Artificial Intelligence (AI)/ Machine Learning (ML)
Natural Language Processing (NLP)
Generative AI (GenAI)
ML Model and Foundation Model (FM)
Prompt Engineering
Responsible AI

6 of 57

Artificial Intelligence (AI)/ Machine Learning (ML)

Using machines to find patterns in data and mimic human intelligence

Using machines to find patterns in data and mimic human intelligence

7 of 57

Natural Language Processing (NLP)

Branch of AI that focuses on helping computers understand and create human language

8 of 57

ML Model

A machine learning model is a file that can make predictions or answer questions based on patterns it has picked up on in datasets you have shown to it

9 of 57

Generative artificial intelligence (GenAI)

Creates new content and ideas
Powered by Foundation Models

10 of 57

Foundation Model (FM)

A REALLY BIG model, trained on LOTS of data, that is general enough to do lots of different tasks

“a photo of an astronaut riding a horse on mars”

11 of 57

Prompt Engineering

Guiding the AI tool to respond the way you want it to. Use creativity + trial and error to figure out how best to reliably get your answer

12 of 57

Responsible AI

The practice of designing and using AI in a way that prioritizes fairness, transparency, privacy, and ethical impact

13 of 57

PRACTICAL APPLICATIONS FOR AI/ML

14 of 57

What are your goals?

Building the right message

Fundraising

Connecting with voters

15 of 57

Connecting with voters

Help people find exactly what they need on your website
Automating support functions, such as surveys
Personalize communication using information in your CRM

16 of 57

Showing the right item at the right time

PBS is using Amazon Personalize to recommend video content to their viewers based on their viewing history

With Amazon Personalize, just inserting a small set of viewers, videos, and interactions, we saw recommendations that stood up and could possibly scale.

- Mikey Centrella, Direct of Product Management, PBS

17 of 57

Building the right message

Generate starter copy for outgoing emails or social media
Summarize data that campaign principals need automatically

18 of 57

19 of 57

Fundraising

Forecast future donations given historical data
Identify donors most likely to churn, and proactively reach out

20 of 57

DEMO

21 of 57

22 of 57

AI AND DATABASES

Russ Rampersand, Chief Data Officer

Catalist

23 of 57

Operationalizing AI

24 of 57

Operationalizing AI

25 of 57

Terms of Service

26 of 57

Structured vs Unstructured: What kind of data do you have?
Hygiene the Data: Do you want AI to do the hygiene or do you want AI to give you code to execute?
Process the Data: Diffing, Matching and Coalescing
Analysis: What do you need to know about the data set?

Deciding What You Need To Accomplish

27 of 57

Structured vs Unstructured

28 of 57

Data Hygiene

29 of 57

Data Differentials

30 of 57

Data Matching

31 of 57

Facets of a Person

32 of 57

Analysis

33 of 57

Ryan Bonifay, Director of Data & Analytics

ColdSpark

CREATING DATA VISUALIZATIONS WITH AI

34 of 57

NEED FOR DATA VISUALIZATIONS

Uncover trends and patterns
More accessible to a broader audience
Informed decisions made more quickly

35 of 57

HOW TO MAKE DATA VISUALIZATIONS

Data sets that you want to create visualizations for
OpenAI’s ChatGPT or other services
Knowing what information stakeholders need to see

36 of 57

MAKING A DATA VISUALIZATION

Dataset to create visualization
ChatGPT and Graph Maker GPT

37 of 57

MAKING A DATA VISUALIZATION

Attach dataset in prompt box
Graph Maker will define and generate visualization suggestions from the data

38 of 57

MAKING A DATA VISUALIZATION

Enter in the prompt what type of visualization you would like to see

39 of 57

MAKING A DATA VISUALIZATION

Enter edits in the prompt to make the visualization clearer and easier to understand
Save the image of the chart

40 of 57

ADDITIONAL DATA VISUALIZATIONS

Polling data
Digital fundraising performance
Absentee/Early Voting trends
District analysis
Voter turnout data

41 of 57

THINGS TO KEEP IN MIND

Use the analysis done by the GPT to help explore and create additional visualizations
Data overload can lead to clustered visuals
Take time to explore all of the features available with the application you are using

42 of 57

George Nassar, Partner

Public Opinion Strategies

AI AND SURVEY RESEARCH

43 of 57

California U.S. Senate: Nancy Pelosi v. Katie Porter

Utilizing A.I. to…

Simplify Large Amounts of Text
Re-Work Existing Content
Create New Content

44 of 57

Prompt #1: Simplify Large Amounts of Text

Prompt #1

A.I. Response #1

45 of 57

Prompt #2: Re-Work Existing Content

Prompt #2

A.I. Response #2

46 of 57

Prompt #3: Creating New Content

Prompt #3

A.I. Response #3

47 of 57

Prompt #4: Creating New Content

Prompt #4

A.I. Response #4

48 of 57

Analyzing Survey Data

Utilizing A.I. to…

Quickly Summarize Key Findings
Drill Down Deeper into the Data
Find Insights that are not Immediately Apparent

49 of 57

Analyzing Survey Data

50 of 57

Prompt #1: A Top-Level Summary

Prompt #1

A.I. Response #1

51 of 57

Prompt #2: Diving Deeper into the Trends

52 of 57

Prompt #3: Creating New Content

53 of 57

Verbatims

About 20 pages per candidate.
Typically analyzed by reading through, creating word clouds.
Generally, try to focus on what is breaking through with persuadable groups of voters.

54 of 57

Prompt #1: A Top-Level Summary

Prompt #1

A.I. Response #1

55 of 57

Prompt #2: Drilling Down

56 of 57

Prompt #3: Drilling Down

57 of 57

Prompt #4: Drilling Down