Success Framework for Marketing Generative AI: Cautiously exuberant

Highlights/Executive Summary

1. Generative AI is exploding onto the Data Analytics Scene with many potential use cases. Firms and Consumers have shown extreme exuberance about the potential use cases. Not all use cases are ready for prime time, and firms must select them carefully.
2. Like any form of analytics, firms need to ask the question, “What business problem am I trying to solve” and What is the potential impact and value add?
3. Firms need a Generative AI Plan and Success Framework for Marketing that grounds their transformation journey in the highest impact, safe, and well-governed use cases.  The Success Framework must include a process for evaluating Marketing Generative AI potential use cases by assessing the value, readiness Level, and safety.

Introduction 

The following article is an expert opinion piece about Generative AI, primarily in marketing, with some examples from Chat GPT4.0.

Generative AI has many potential applications, from Image and content creation to writing marketing briefs, to analyzing and summarizing campaign results. In addition, Generative AI can fill in missing data (aka Synthetic Data) or variables from known inputs for customer profiling and targeted marketing. These use cases have generated much excitement across a variety of domains, and yet at the same time, some experts warn of the need to go slower in developing Generative AI. Whatever the warnings, executives need to understand the technology and feasibility within their own firms of any particular use case and weigh the mix of AI activities aligned to strategic goals. For example, marketers and executives need to ask How generative AI for marketing fits in with the overall AI strategy?  How do firms evaluate their generative AI use cases based on value and success?

A Brief Definition of Generative AI: 

This is what Wikipedia defines as Generative AI.  We won’t belabor the public definition, but it is worth noting the point about the misuse of generative AI.

Generative artificial intelligence or generative AI is a type of artificial intelligence (AI) system capable of generating text, images, or other media in response to prompts.[1][2] Generative AI models learn the patterns and structure of their input training data, and then generate new data that has similar characteristics.[3][4]

Generative AI has potential applications across a wide range of industries, including art, writing, software development, healthcare, finance, gaming, marketing, and fashion.[8][9] Investment in generative AI surged during the early 2020s, with large companies such as Microsoft, Google, and Baidu as well as numerous smaller firms developing generative AI models.[1][10][11] However, there are also concerns about the potential misuse of generative AI, such as in creating fake news or deepfakes, which can be used to deceive or manipulate people.[12]

 Section 1: Jumping in with Excitement, an example of AI Hallucinations. 

As in many business endeavors, there are important tradeoffs to consider when leveraging Generative AI versus not starting on the journey and the associated technology barriers. Starting a generative AI journey is key, and doing nothing is not an option as there are many cost and operational benefits to moving forward with this transformation. However, as we can see from some examples in other domains, understanding how Gen AI works is key. We highlight an example of a use case gone sideways in the legal field next.  This example highlights the need to understand how the data sources feed generative AI work and the outputs of any use of Generative AI and large learning models.

An interesting episode of ABC’s “The View” can help frame the paranoia and misunderstanding of Generative AI. Whoopi Goldberg expresses concerns about using ChatGPT for Court Cases and Wedding Vows. While Ms. Goldberg has valid points, it is clear that most of the panelists, being commentators/ journalists and not data scientists, did not fully understand how Generative AI and ChatGPT actually work. Have a look at the video:

Link to the View Episode.
https://youtu.be/f169CFC7Y7M

To be clear, any data scientist can tell you that large learning models work on a methodology, so it is not true that we can’t understand what they are doing and how they work. We can! We can understand their functionality, their success metrics, and more. All models fit the data and have output diagnostics to tell you how good they are. GNI, RFI, RMSEA, etc.   Without going into the mechanics, the main thing we can’t control is what they learn, their output, and, in some cases, the data they are pulling from, especially from the open internet.  Therefore, good data management principles are more important than ever, especially when combining public sources with our own data. Knowing your data sources and validating them should be the mantra. So, to Whoopi Goldberg’s lament, we say that if ChatGPT or Large Learning Models (LLMs) were asked to write a case or a summary of legal research from accurate, validated legal data sources, it could. Understanding how Chat GPT was built is important, as understanding the data sources and model inputs is vital to understanding how they work. In our example above, obviously, the lawyer didn’t understand how the model worked, and unfortunately, neither do many other professionals, well, at least not yet. The example given on The View was from a lawyer who didn’t know how LLMs work or how data science works and used unconstrained, unvetted, unverified sources from the open internet versus confirmed legal data sources.

2)    Starting your journey:  Determining which Marketing Use Cases

Like any form of analytics, firms must ask, “What business problem am I trying to solve, and what is the potential impact and value add”?  Once that question is answered, a value can be assigned to the use case and a rating regarding ease of implementation.  These assessments can help you prioritize your use cases and to create a Generative AI plan.  Also, you will want to understand how the mix of the different use cases helps you to learn and build capabilities over time.

Marketing pros need to weigh the value generated with their ability to execute on any potential use case with adherence to privacy and regulatory concerns. 

Obvious overall good governance of AI, in general, is paramount to ensuring Generative AI stays on track to deliver the benefits while staying compliant with privacy principles and existing laws.

Marketing Use Cases

Firstly, we will not discuss all of the Marketing Data Science Algorithms that power up the use cases that follow, but we will briefly mention the major algorithm types to help marketers be better consumers of data science. We suggest you meet with your data science team for a deep dive.  In marketing circles, it is important for executives and their teams to understand how some key learning algorithms work, including Decision Trees, Random Forests, and other deep learning processes. Natural Language processing is one of the most important types of Machine Learning in marketing as it allows marketers to understand customer sentiments received in various channels: social media, chatbots, metaverse, and more.   In our opinion, these AI and ML tools are best used to augment human intelligence and creativity. 

Section 2: Marketing, Sales, and Customer Use Cases and their Value Drivers. 

The Use Case, a brief description and the KPI’s Impacted:

1. Marketing content

Helps augment agency and copywriting efforts and can enhance brand language and copy tone.  Email Marketing: email templates, calls to action, and subject lines.

Automation of campaign briefs.

• Reduce Creative Costs
• Increase Brand Perception
• Increase email open rates

2. Targeting and Personalization

Tailor individual messages and offers to individuals or segments.
Tailored content based on customer needs.

• Increase offer relevancy and acceptance.
• Revenue Uplift

3. Generate new images based on brand position and customer feedback

Gen-AI can generate unique and personalized images by leveraging public and firm data sources.

• Reduce create costs.
• Increase brand/product differentiation.

4. Customer Experience

Answer customer questions, write FAQs, and draft responses to customer communiques.

Sentiment Analytics

Analyze customer notes and social media feedback and reviews and create summaries.

• Reduce data management costs.
• Increase speed to insight.

5. Summarizing Analytics and Research Reports

Generate summaries and actionable insights from various sources.

• Reduce Operating Cost
• Speed to Insight

1. Images and Logo

Generate new brand-related images and new logo designs.  Allows for consistency of brand elements while generating new images/logos.

• Minimize Agency Costs

7. Marketing and Marketplace research.

Questionnaire design summarizing results.

Operational Efficiency

8.RFP Development and Response.

Leverage past RFPs and Requirements documents to generate new RFPs/RFP Responses.

• Speed to market
• Consistency of responses

9. Data Management

Fill in missing data through extrapolation and known sources.

• Operational efficiency
• Effectiveness of targeting,
• The success rate increased of targeting models and program results.

Section 3:  Success Framework based on good AI governance. 

Overall, a Success Framework for Generative AI needs to be implemented.  When beginning the journey of governing Gen AI and thinking about success, the firm must understand the value of the Generative AI pilot or program. What does the use case solve, what does it cost to deploy, and how will it help improve business results? Does the firm have the ability to execute and deploy the use cases?  An overall AI maturity assessment needs to be conducted, and the firm needs to weigh the level of generative AI maturity in the context of overall AI maturity. The CMO or Digital officer must work closely with the data and analytics unit to ensure the data needed for the use case is available, verified, and understood.   The potential output of any potential use case needs to be analyzed for safe use and adhere to the firm’s AI Governance Framework.  If the firm does not have an AI Governance Framework, then that needs to be implemented before Generative AI Pilots are launched.

6-Step Marketing Success Framework

1. Create and Engage with the right team across disciplines and within the marketing team.  Digital Marketing, Marketing Analytics. Brand Channel Management, Technology, and Data Partners.
2. Identify and vet your chosen use cases. Ensure the value added and align on and set the prioritization. Will the use cases drive cost savings, good customer outcomes, or generate strictly revenue uplift?
   1. Decide on Marketing Success Criteria:  Customer Lifetime Value, break-even ROI, Conversion, Return on Customer, NPS, and More.
   2. Assess the ability to execute and deploy the use case.
   3. Agree on the definition of done.  Determine when the use case or test is complete and what level of success is needed for rollout.
   4. Identify seed funding or investments upfront.  Marketing promotional costs, media spend, creative or data costs, etc.
3. Understand your input data source and validate them.  Be aware of public versus private data sources used as input.
4. Understand how your Large Learning Model(s) will leverage/consumer data sources.  Check with your data scientists and compliance experts on data usage.  Will open internet sources be used or only proprietary sources?  Are there any potential consumer privacy implications?
5. Ensure oversight through governance:  data ethics, privacy regs, and data security.  Validate the models (Efficacy, accuracy, validity).   Ensure logic testing of the output against the original business question(s).
6. Piloting/Test and Learn(execution):  Understand how tests are deployed and the forecasted results (Set KPIs/marketing hold-back samples/baselines, etc.). Understand IT timelines and infrastructure constraints.

I highly recommend conducting a Generative AI Assessment to drive your AI strategy.  This strategy is necessary to help your organization establish value and risk in Generative AI.  I hope you enjoyed the article and look forward to your comments and questions.  If you want to discuss the Marketing Generative AI success framework further, I can be reached at tony@anthonybranda.com or a.branda@tcs.com.

Anthony Branda is the lead Partner for Marketing Data Analytics in the Advisory and Consulting Practice at TCS and the Founder of the Analytics Hall of Fame. Formerly a CAO, CDO, and CDAO for Citigroup, Commbank, RBS, Ahli-United Bank, and Embrace Home Loans.

He founded a Master’s in Marketing Analytics at Pace University with Tom Davenport and is published in the Journal of Marketing Analytics and Applied Marketing Analytics.