How to make brands safer
and stronger
in an AI world
present
Transfer learning
Data mining
Algorithm
Machine learning
Neural network
Strong AI
Backward chaining
Hyperparameter
Variance
Entity annotation
Machine intelligence
Hyper parameter
Autonomous
Deep learning
Entity extraction
Neural network
INTRODUCTION
The era of artificial intelligence is in full swing, and adoption is accelerating rapidly. Nearly 75% of all companies have already integrated AI into their business strategies and 30% have implemented pilot AI programs for a wide range of scaled applications, from accelerating R&D timelines for new products to enhancing customer experiences, according to a recent survey.
And today’s marketers are at the forefront of the new generative AI frontier, deploying applications using neural networks and the natural-language processing capabilities of generative AI platforms. This is led by OpenAI’s ubiquitous ChatGPT software program, buoyed by Microsoft as its biggest investor. It passed the 1-million-user mark in its first five days on the market in November 2022 and now boasts over 100 million users.
01
The global generative AI market, which was estimated at about $10B In 2022, is expected to grow at an annual rate of 35.6% and hit
$109B
by 2030, according to Grand View Research.
Rate of generative AI adoption in the workplace in the United States 2023, by industry
0%
5%
10%
15%
20%
Share of respondents
30%
35%
40%
45%
25%
Marketing and advertising
Technology
Consulting
Teaching
Accounting
Health care
37%
35%
30%
19%
16%
15%
Barriers to
further adoption
But not everyone is hopping aboard the AI train. Concerns about potential data leaks have caused major brands, including Wells Fargo, Verizon, Samsung, Spotify, Amazon and Apple, to at least temporarily restrict or ban employee access to generative AI platforms, including ChatGPT.
02
Digital marketers are among the most frequent users of generative AI technology. An AuthorityHacker survey conducted earlier this year found that 75% of digital marketers are using AI for work and nearly half (49.5%) are using AI tools multiple times per week or more. They range from website builders and SEO practitioners to freelance marketers and in-house agencies.
Notably, the marketing and advertising industry currently leads all other fields in the rate of adoption of generative AI:
Rather than put their proverbial heads in the sand, these companies need to find a way to embrace the massive opportunities of AI while appropriately managing their risk exposure to the technology
Dan Temby, senior VP of technology and analytics, DAC
“
Besides safeguarding proprietary data or protecting trade secrets, DAC has encountered resistance to using generative AI platforms among some marketers due to concerns over the accuracy and quality of the content produced. Those concerns are amplified in industries like financial services, where companies must follow strict regulatory guidelines for claims in marketing and advertising.
Further, all marketers want to make sure the language being produced by tools like ChatGPT and used by employees is consistent with how their brands are represented to the public.
Yet there are strategies for agencies and creative professionals to lean into AI that experts say will make the technology more of a friend than foe. In a recent Ad Age webcast, Chinmayi Bettadapur, head of product management, Salesforce Industries Media Cloud, said, “AI can help companies in three main categories. Predictive AI can help them make better decisions. Automated AI can help them save time and money.
And generative AI can help them increase efficiency and productivity.”
Top factors hindering ai adoption
Lack of knowledge for effective use
45.2%
13.9%
22.5%
6.3%
5.6%
Lack of time for testing and implementation
AI accuracy concerns
High AI cost
Ethical concerns
AI usage complexity
Other
Choice
Number of votes: 925
Keys to success
Adapting to AI-induced transformation will require more than just a mindset shift. Marketers will have to recalibrate their approach to many things, including investments in data and technology, prioritizing skill sets and human resources and basic marketing functions like SEO.
Fortunately, there are several steps that marketers take right now to future proof the organization and ultimately succeed in the world of AI.
03
Top questions you want answered
Here are four of the most frequently asked questions about ChatGPT:
04
Let’s address each in turn.
1. Processing the information.
Generative AI algorithms interpret human language patterns based on statistical probabilities. To the model, the “right” answer is the statistically most likely answer.
Consider a word like “groom.” It can have several meanings: You can groom a pet, you can be the groom at a wedding, or someone can be well groomed. In the language model’s mathematical framework, “groom” is represented by a cloud of numbers in a multidimensional space. This cloud captures the word's various meanings and potential usages based on different contexts.
“Now imagine when the word ‘groom’ appears in a sentence, surronded by other words,” Temby explained. “The model then selects the most suitable version of groom or its related context, and constructs the sentence. It does this for each word in a sentence, one after another. Just like we piece together words to make meaningful sentences, the model follows a similar pattern, but it uses a vast sea of numbers and complex math to get the job done.”
2. Training the model and
getting better answers.
For software engineers and developers, “training” a generative AI model is an exhaustive, multistep process. But when it comes to the one aspect of training that users perhaps care most about—getting better and more accurate responses—the solution is deceptively simple: Learn how to ask better questions. The people who know how to ask the best questions are going to jump to the head of the class.
Creating a safe, controlled environment
In many ways, AI technology is still in its formative stages. The entire space is yearning for utility and efficiency. “There are new AI products and offerings sprouting up every day. They all are basically an application layer that ladder up to one of three or four big AI platforms, and a real need is to provide access to this technology that is secure, controlled and brand aligned,” said Sahlool.
DAC has built a new application programming interface (API) layer that allows marketers to manage all of these platforms together. The goal is to maximize each tool’s individual strengths, fortify the system’s total memory and provide guardrails to protect company information within a controlled, secure and brand-aligned environment. “Think of individual generative AI models as brands of premium tequila, our job is to make delicious
margaritas,” added Temby.
ChatGPT “in the wild” offers little control
DAC'S PROPRIETARY AI PROFILER FEATURES
Brand guidelines
Language rules
Regulatory compliance
Private/secure API data transmission
Audit trail
Access and usage rules
Brand safety compliance + security
Detailed
context
Centralized prompt optimization
User Interface
Neural Network
Prompt is passed directly to LLM
Prompt is passed directly to LLM
Prompt
Completion
Use Default Behavior
Putting it all together
While this model can be applied in a wide variety of sensitive and critical areas to drive business outcomes, Sahlool explains how DAC achieved this result illustrating a specific use case.
DAC is a leading international performance marketing agency that helps brands connect with customers from the enterprise level right down to hyper-local moments. Celebrating our 50th year in business and recognized by Forrester as one of the world’s most significant performance marketing agencies, DAC is passionate about helping brands build, maintain and optimize their online presence, as well as drive traffic and conversions to their local stores. Our integrated services are built around our unique Enterprise-to-Local methodology, underpinned by technology and accelerated by our services (paid media, strategy, content, creative and data analytics). Our campaigns are best-in-class and drive transformational outcomes for our clients, which has led to our success in winning Google Premiere Partner Awards, US Agency Awards and Search Engine Land Awards. For more information about our AI offerings, please contact hello@dacgroup.com.
3. Why do responses so often appear to be wrong?
If the model has the right answer in its training data, it will often give you the right answer because it will be statistically more likely and have higher score. “It’s not willfully giving you false information; it’s making a mathematical decision based on the best information that it has available,” Temby explained.
4. Protecting trade secrets
Protecting trade secrets. Marketers have many questions about sharing sensitive information with platforms like ChatGPT. Let’s say a marketer uploads or copies and pastes a document about a new product, wanting to know what the model would recommend about messaging or brand positioning. The common fear is that some other ChatGPT user asking about the company can find out about the launch ahead of time, effectively stealing a trade secret.
Nasser Sahlool, senior VP of client strategy, DAC
Partly for these reasons, we see this as a time for all marketers to test incrementally into the space, with the safeguards and oversight you would provide in a non-AI environment, but with additional guardrails around brand alignment
“
But that’s not what’s happening.
There are certainly strict privacy and protection policies in place that ensure the model does not have the ability to recall past interactions and inadvertently share your data directly with another user. That being said, some data may be used by OpenAI engineers for the purpose of improving the model, debugging and ensuring a quality user experience.
“
Here’s a visual representation of the DAC approach:
Neural network
Strong AI
Backward chaining
Entity annotation
Hyper parameter
Autonomous
Hyperparameter
Variance
Machine intelligence
User interface
Prompt is passed
directly to LLM
Prompt
Prompt is passed
directly to LLM
Neural network
Use
default behavior
Completion
Centralized governance, context and optimization
User interface
Prompt is passed directly to LLM
Global/DAC rules are appended to behavior
Global rules
Brand rules
Brand safety rules are appended to behavior
Compliance rules are appended to behavior
Compliance
Intercepting LLM generation to apply centralized value
Prompt
Add
behavior
Use Default Behavior
Add
behavior
Add
behavior
Add
behavior
Neural network
Entire prompt is passed to neural network/LLM for completion
External data sets are appended to behavior
Embedding
Context
Client context and/or location specific context is appended to behavior
System behaviors are appended to behavior
System
Completion
Add embedded data
Use Default Behavior
Add
Context
Add
behavior
Add
behavior
Learn more
Guardrail an advanced AI platform and a specialized layer that works in conjunction with industry-leading AI services like OpenAI to offer you a secure and aligned operational environment.
When it comes to data security, we adhere to strict data protection guidelines. Furthermore, we ensure that the APIs we utilize do not store or re-use any prompts that are submitted, offering an additional layer of safety for your data.
DatA security
But Guardrail is not just about security; it’s also about alignment with your brand’s guidelines and values. You can pre-load these into our system, where they are safely stored in our state-of-the-art vector data storage. At runtime, this contextual information is dynamically provided, guaranteeing that all output is in line with your brand’s ethos. Additionally, Guardrail offers advanced validation functionalities, allowing you to
be certain that the generated content not only aligns with your brand but also meets any predefined
quality criteria.
In sum, Guardrail is your comprehensive AI-driven solution for secure and consistent operations, preserving both your data and your brand’s integrity.
Brand safety
AI in marketing
AI for Search
AI for Content & Creative
AI for Reviews
AI for Data
There are a number of text-based AI applications that can take advantage of DAC’s new interface to great effect. Take reviews, for example. A ChatGPT session in a normal environment might provide a serviceable response to a review or customer complaint. Consider this raw, unedited AI response to a consumer review of a major PC brand:
With the new profiler, the response becomes much more thoughtful, as if it really listened to what the customer said, and picked up on various specific comments in a much more conversational and natural sounding reply—without complex prompting or further iteration:
We built a whole review moderation module with all of the rules and guidelines based on our history of working with the brand—the standards, brand tone and voice, and rules on escalations. We injected this new memory layer into the platform, giving much more compelling, engaging and personalized answers.
“
dacgroup.com
For most creative writing functions however, no amount of enhanced functionality can entirely replace human intervention. And that’s perhaps the most important lesson of AI’s ultimate impact on the marketing profession. Temby suggests thinking of an AI generated report as a first draft.
“If I ask a junior analyst to write a report on a marketing campaign, I should assume that it’s not ready for the client until I’ve reviewed it and accepted the edits,” Temby noted. “Why should a report generated by an AI model be any different?”
Ad Age Studio 30 is the creative content arm of Ad Age. Built on the same bedrock of journalistic integrity, Ad Age Studio 30 specializes in multichannel membership content for Ad Age subscribers, as well as custom and sponsored content that resonates with our audience. To partner with Ad Age Studio 30, email James Palma at jpalma@adage.com.
adage.com
Dan Temby, senior VP of technology and analytics, DAC
Legal-approved copy/content
Product knowledge
Proprietary data or information
Long-term ‘memory’
Known high-performing creative
Democratization of knowledge/skills
Instant distribution of improvements
Source: The State of AI in the Online Marketing Industry: 2023 Report, Authority Hacker
How does ChatGPT process the information I provide?
How can I “train” the model to produce better
and more accurate results?
When I share sensitive information with ChatGPT, what is to prevent other users from seeing it as well?
Why do the responses and answers I get back always seems so bad or wrong?
Nasser Sahlool, senior VP of client strategy, DAC
Sometimes change comes faster than our brains have the capacity to process. That speed increases exponentially regarding developments in artificial intelligence. Ad Age Studio 30 and leading international performance marketing agency DAC have partnered to produce this special interactive report to help advertisers and marketers process all they need to know to succeed in the rapidly evolving AI landscape.
Digital marketers are among the most frequent users of generative AI technology. An AuthorityHacker survey conducted earlier this year found that 75% of digital marketers are using AI for work and nearly half (49.5%) are using AI tools multiple times per week or more. They range from website builders and SEO practitioners to freelance marketers and in-house agencies.
Add
behavior
Add
behavior
Add
behavior
Add
behavior
Add
behavior
