Going From Data To Ideas With AI
Re-imagining the strategic and creative planning process with LLMs
To a human living in 2023, the speed of information dissemination is unprecedented. Powered by technology, network effects, and online communities, the perception that the world is moving faster reflects our new reality.
But, in a rapidly changing world, the traditional planning process begins to break down. For example, the output of a 6-month marketing campaign planning cycle is often stale by the time it reaches the shelves.
At Addition, we've been exploring ways to embed functions of brand strategy and creative planning within AI systems via Large Language Models (LLMs). In doing so, brands and agencies can accelerate the time it takes to go from research to ideas by 10-100x while maintaining strategic rigor.
This type of lofty statement requires a live demo to become tangible. So, let’s start with a question that we can use to go from data to insights to ideas.
How should brands respond to the Maui fires?
To answer this question and get to a tangible output, we’ll have AI conduct research, develop a brief, and develop an idea for a brand activation that supports victims of the Maui fires.
It’s a topic that is both time-sensitive and strategically nuanced, not to mention vital to the many people whose lives have been impacted.
Time is of the essence, so we’re going to move quickly.
Collecting Research
To get started, we'll conduct some online research. We will do so by deploying an AI-powered web-scraping tool that consumes online content and uses the information it finds to go deeper on the topic. Running this web scraper on "how to help with the Maui fires," I can quickly collect thousands of results from the web, news, and social media. These documents are vectorized using large language embeddings and loaded into a vector database for information retrieval.
Exploring Insights
Once the scraper has finished, I'm ready to begin working with the data set. We'll load it into one of the Strategy Tools we've developed at Addition that will enable me to run the data through custom LLM pipelines to extract information and come up with insights.
I provide the tool with a prompt using a /slash-command, and from there, the system retrieves the most relevant information and passes it through a set of LLM functions that:
Extracts the most relevant information
Identifies an insightful perspective within the information
Presents the insight in a pithy 1-2 sentence format along with its sources
I give it the following prompt:
"/Generate insight about how brands should get involved in the Maui fire recovery."
Now the AI goes to work on the source material and identifies some key considerations:
- The Maui fires have caused immense damage and displaced many residents. There is a huge need for donations and support.
- Many organizations and funds have been set up to coordinate donations and distribute aid. However, distribution has been challenging due to disorganization and influx of donations.
- Residents are requesting specific items like non-perishable foods, water, hygiene items. There is concern about coordinating what is needed most.
- Businesses are looking for ways to help the recovery efforts through donations.
It then goes on to give me an insight:
The impulse to donate what we have readily available doesn't always fill the unique needs of a devastated community. True support means brands collaborating with relief organizations to provide items specifically requested by those rebuilding their lives.
It’s not earth-shattering. But it’s practical, and since the cost of ‘re-briefing’ AI is pretty much zero, I’m going to run with it and see where we land.
Generating Ideas
Next, we will use this insight to generate an idea using a similar LLM pipeline. In this case, instead of generating an insight, the AI will use the data sources to write a creative brief and generate an idea based on it.
For this demo, we’ll pick a brand that has a vested interest in Maui’s recovery: Expedia.
To augment our initial data set, we’ll pull run a web crawl for “Expedia Maui Fires”. This additional data will help ensure that the AI’s thinking reflects the brand’s current response to the fires.
Once I've loaded in the additional data, I give it a prompt:
“/Generate an idea for Expedia to team up with relief organizations to support the Maui fire recovery”
I hit enter, and the AI goes to work, documenting its thought process as it goes from reviewing the research, to generating a brief, and ultimately coming up with an idea.
The AI starts by identifying some key findings from the source materials that it will use to craft its idea.
- Deadly wildfires caused widespread devastation in West Maui, displacing residents and damaging tourism infrastructure
- Officials are discouraging non-essential travel to West Maui to allow focus on relief efforts
- Many hotels and resorts in West Maui are closed, refunding bookings
- Airbnb and some travel companies are providing flexibility on cancellations/refunds
- Travel to rest of Maui not affected, officials encourage supporting these areas
- Donations can be made to relief funds to help recovery efforts
- Airbnb offering housing units to displaced residents
- Travel companies communicating inconsistently on policies
The AI then goes on to identify a cultural tension, insight and tactical opportunity.
Cultural Tension: The tension between Maui's dependence on tourism revenue and the need to pause tourism to focus on relief efforts.
Insight: Expedia has the capability and resources to provide innovative solutions that bridge this tension.
Tactical Opportunity: Leverage Expedia's technology, customer reach and relationships with relief organizations to redirect canceled trips into relief aid.
It then uses this brief to generate an idea:
Expedia will launch "Trips for Recovery", allowing travelers to instantly convert any canceled Maui booking into a donation to relief organizations. When a user cancels their Maui trip on Expedia.com, they will be presented with the option to donate the value of their booking. Expedia will match donations up to $500,000. The seamless user experience turns canceled trips into real aid, while the matching motivates increased donations. It leverages Expedia's strengths - its platform, customer relationships and capabilities - to drive impact. Promoted via email and social media, it rallies Expedia's partners and massive customer base worldwide to support recovery efforts in Maui quickly and efficiently. Measurement will track total donated trips and matching funds raised for relief organizations. The campaign emotionally resonates by giving travelers an easy way to turn disappointment into hope and healing for Maui residents.
Implications
Going from zero information to insights, a strategy, and a concrete idea to move forward with took me less than half an hour, not counting the 45 minutes I spent waiting for the web scraper to finish running.
This begs the question: can we figure out a better way to accomplish strategic marketing planning with AI?
As demonstrated in this demo, the answer is undoubtedly yes. But will we? Will the entrenched ways of working that have been ingrained in us through years on the job act as a barrier to reinventing the strategic process? Or will financial pressures and the tangible benefits of innovation win out, pushing us to reconsider how we do things?
As we learn more, I’ll report my findings here.
Bonus: here’s a video of the process if you’re in need of something to share in social.
Addition is an AI company for the brands of tomorrow.
Visit our website to learn about the work we do with brands and agencies
Read about us in the Wall Street Journal
Follow us on Twitter