ai-first marketing
AI tools like ChatGPT – and another 200 – offer endless possibilities. If you know how to use them.
Our framework involves gaining a holistic understanding of the AI landscape. Instead of attempting to keep up with all the market releases, the focus should be on comprehending the fundamental technologies and models that are being developed.
These foundational models serve as the building blocks for various companies, which either fine-tune these models or incorporate them (or multiple of them) with other systems to create their products.
And the potential for business is also evident. Generative AI tools can swiftly generate an extensive range of effective content, and refine it based on feedback to better suit its purpose. This will impact numerous industries and jobs because adaptation must be quick, intelligent, and well-structured.
With a team of 8 people, and in partnership with the University of Porto, we’ve developed the first European framework that will integrate Generative AI with creative teams, and create a valuable content outcome for our clients, having into account all the data privacy, ethics, and regulation involved.
MARKET CONTEXT
Examining the broader business implications, research suggests that large language models could affect a significant portion of the US workforce, impacting at least 10% of the work of about 80% of workers, and more than 50% of the tasks for 19% of workers. Furthermore, if appropriate software and tools are integrated with these models, they could expedite between 47% to 57% of all tasks while maintaining the same quality standards (study by OpenAI, OpenResearch and University of Pennsylvania.)
A recent study by McKinsey has identified several generative AI use cases across different business functions, which could deliver between $2.6 trillion to $4.4 trillion in economic benefits annually when applied across industries. This accounts for an addition of 15 to 40% to the previously estimated economic value provided by AI. In particular, customer operations, marketing and sales, software engineering, and research and development stand to represent approximately 75 percent of the total annual value from generative AI use cases.
Risks of ignoring Data Privacy & Ethical AI:
1. Breaches & confidentiality
2. AI Bias & Discrimination
3. Lack of trust & transparency
4. Ethical & legal non-compliance
5. Fake outputs and content
THE OPPORTUNITY
This structured model, which will improve the effectiveness and consistency as well as control the risks of using AI, but will also produce more business-oriented solutions.
This model will be focused for the time being on AI working collaboratively with creative teams throughout the process, from the definition of the concept to their creative solutions.
The framework, in a nutshell
The framework,
in a nutshell
the Tech Mapping
> Benchmarking of tools based on business results;
> Information gathering through crowdsourcing, allowing clustering of tools (by case studies, features, etc.);
> Web scraping and advanced AI agents that search the web for new tools and even test them.
The art of prompt
> Unique benchmarking of prompts based on business outcomes;
> Ethical and data privacy assessments;
> Automation of input testing with AI agents;
orchestrate
> Combining AI agents and people to achieve project goals more efficiently;
> Providing AI agents with tools that allow them to connect to APIs, browse the web or request user input (OpenAI plug-ins, Langchain);
BUSINESS-ORIENTED AFTERMATH
> Own assignment templates to map KPIs to tools, prompts and orchestration;
RISK AVOIDANCE
> Specific rules to be provided to models, ensuring quality and ethical compliance and data privacy;
1. the Tech Mapping
> Benchmarking of tools based on business results;
> Information gathering through crowdsourcing, allowing clustering of tools (by case studies, features, etc.);
> Web scraping and advanced AI agents that search the web for new tools and even test them.
The art of prompt
> Unique benchmarking of prompts based on business outcomes;
> Ethical and data privacy assessments;
> Automation of input testing with AI agents;
orchestrate
> Combining AI agents and people to achieve project goals more efficiently;
> Providing AI agents with tools that allow them to connect to APIs, browse the web or request user input (OpenAI plug-ins, Langchain);
BUSINESS-ORIENTED AFTERMATH
> Own assignment templates to map KPIs to tools, prompts and orchestration;
RISK AVOIDANCE
> Specific rules to be provided to models, ensuring quality and ethical compliance and data privacy;
A Secure and Ethical AI Solution
Comprehensive data privacy.
Safeguard sensitive information, prevent unauthorized access, and minimize data breach risks.
Safeguard sensitive information, prevent unauthorized access, and minimize data breach risks.
Bias detection and mitigation.
Fair treatment for all users and avoidance of legal liabilities.
Fair treatment for all users and avoidance of legal liabilities.
Transparency and explainability.
Build trust with customers, foster positive public perception, and retain business.
Build trust with customers, foster positive public perception, and retain business.
Compliance with regulations.
Avoid legal penalties, fines, and maintain market access.
Avoid legal penalties, fines, and maintain market access.
More than just a methodology
This framework is itself a technological innovation – although it can have risks involved like any R&D project, and can sometimes fail.
Our work process consists of an approach of hypothesis, and tested systematically and consistently, while we try to optimize the interactions to improve the success rate. But, like any R&D project, it’s impossible to guarantee the final results to be achieved.
Data structure
Our company strategically applies data to optimize our AI-enabled content creation process. After publishing content, we employ a selection of scoring metrics to gauge its real-world effectiveness, each carefully aligned with the specific objectives of the content.
The metrics are analyzed weekly to derive insights into content performance. We then trace the content back to its elemental components, including the originating prompts and the tools utilized. This process, similar to tracing a symphony back to its originating orchestra, gives us an in-depth understanding of the process that led to the final content.
Though we don’t currently implement fine-tuning, we are actively exploring this avenue to augment our AI models further. Specifically, we are looking at tools such as LoRA for supervised fine-tuning, which offers the advantage of training significantly fewer parameters. While implementation requires expert knowledge and substantial development time, we recognize its potential in enhancing the feasibility of our operations while reducing costs.
use cases
