The 2024 State of AI Marketing

This Week

This week, we’re showcasing the highlights from AI Marketing Institute’s 2024 State of AI Marketing, the top AI tools of the year, how to build your own custom GPTs, tips for getting better outputs from your LLM, a look at AI funding and the latest text-to-image generation model everyone’s raving about.

Read on.

Top 4 Takeaways: 2024 State of AI Marketing

More marketers than ever are embracing AI. Are you?

The 4th annual State of Marketing AI Report by the Marketing AI Institute shows that AI understanding and adoption are accelerating. Ninety-nine percent of the survey’s respondents said they’re personally using AI in some fashion, and the level of AI usage is rising significantly. 

Paul Roetzer, the founder of the AI Marketing Institute, says that everyone, regardless of their title, background, or seniority, must understand how to apply AI to their work before it’s too late.

“Rapid AI advancements in language, vision, prediction, persuasion, reasoning, decisioning, and action are about to augment human capabilities and redefine knowledge work at a rate and scale that the economy has never seen,” he says. “Millions of jobs will be impacted as companies realize the power and potential of AI to drive productivity, efficiency, and profits.”

Here are the top highlights from the study:

AI Usage is Soaring

• 99% of marketers are using AI in some capacity.

• 36% have integrated AI into their daily workflows, a significant increase from last year.

Top Outcomes Desired with AI

• 80% aim to reduce time spent on repetitive tasks.

• 64% seek more actionable insights from marketing data.

Barriers to AI Adoption

• 67% cite a lack of education and training as the biggest hurdle.

• 56% report a lack of understanding and awareness.

AI’s Impact on Jobs

• 47% believe AI will eliminate more jobs than it creates in the next three years.

• 78% expect over a quarter of marketing tasks to be automated by AI within the next three years.

You can find the full report at the AI Marketing Institute’s website.

How to Build Custom GPTs

Custom GPTs are powerful tools for optimizing your work, but they can be time-consuming to build and customize. Lucky for us, Section’s Chase Ballard has created a handy step-by-step guide to help us out. 

“We've developed a lightweight process for creating custom GPTs that can be managed by a single person,” says Ballard. 

Here’s a summary of the four steps:

1. Establish the Scope: Define the specific tasks your GPT will perform and set clear goals to measure success.

2. Craft Prompts: Develop precise prompts to train the GPT effectively. Consider integrating external APIs.

3. Refine Prototypes: Test and iterate the GPT through feedback to improve its performance.

4. Deploy and Maintain: Launch the GPT, either private or public, and update it regularly with improvements and updated instructions.

Head over to the Section blog for the full breakdown. 

5 Tactics for Getting Better AI Outputs

In her article "Why ChatGPT is Bad at Math and What Marketers Need to Know About LLMs," Tanya Olga Brouneus from Anthropic explores the limitations of large language models (LLMs) in mathematical reasoning. 

AI research has long focused on math skills as a sign of intelligence, but despite the expectation that computers excel at math, LLMs struggle in this area due to their reliance on probability and lack of understanding of mathematical structures.

This is an example of what Ethan Mollick calls the "jagged technological frontier" of LLMs—where their abilities are inconsistent and often unpredictable, making it difficult for users to know when they are pushing against the technology's limits.

Brouneus urges us to approach the use of LLMs with a balance of curiosity and caution, emphasizing the importance of clear and specific prompts to yield accurate results. 

As she says, “Stay vigilant and pay attention to the output you're getting, especially for critical tasks where accuracy is important. Don't assume that because an LLM handled one task well, it’ll be just as good at everything else.” 

She offers five techniques to improve LLMs' performance:

1. Think step-by-step: adding "Let's think step by step" to your prompt leads to an improvement

2. Chain of thought: ask the LLM to go through a specific set of intermediate steps to get an answer

3. Tree of thought: encourage the LLM to explore multiple options

4. Few shot prompting: give examples

5. Write computer code: Ask the LLM to write and execute code to find the solution

Brouneus cautions that these techniques are not guaranteed and that you still need to oversee (and double-check) your LLM’s answers. And if your task is high stakes, you might want to use a platform better at math. 

You can read the full article at the Descript Blog.

The Top AI Tools of 2024

The AI Marketing Institute's 2024 State of Marketing AI report highlights the most popular tools in the industry this year. Here they are:

• LLMs: ChatGPT and Claude

• Work Suite: Google Gemini and Microsoft Copilot

• Design: Adobe Firefly and Canva

• Chatbot: Drift

• Video: Descript

• CRM: HubSpot

• Writing: Grammarly, Jasper, Writer

• SEO: MarketMuse

• Text-to-Image: Midjourney

Interestingly, 36% of respondents say they have integrated AI into their daily workflows, a significant increase from 2023, when many were still experimenting with AI. This shift shows how AI is becoming a staple in marketers' toolkits.

Get the report here.

AI Funding Hits All-Time High of $23.2 Billion

AI funding jumped 59% this quarter to the highest quarterly level on record in Q2'24. AI’s exceptional quarter was fueled by massive rounds to a handful of startups, including a $6B Series B to Elon Musk’s xAI.

AI startup funding hit $23.2B, surpassing the previous quarterly record of $23B set in 2021. By comparison, broader venture funding was down 59%.

Discover the other deals fueling the AI boom in CB Insights’ State of AI Q2’24 Report.

Is the Hype Around Text-to-Image Real?

There's a lot of hype around Black Forest AI's newest text-to-image generation model. And yes, it can generate impressive renditions of almost anything, but it still suffers from the same issues as other tools.

Here are some of the quirks I discovered after a few hours of tinkering:

↳ It still can't write text accurately; while it did an (almost) okay job with my Oscar Wilde quote, it failed miserably at labeling fruits.

↳ It's highly biased - spitting out male versions of CEOs, doctors, and lawyers and female versions of housemaids, nurses, and ballet dancers.

↳ In some cases, it does a decent job rending hands; in others, it creates a mangled mess.

↳ When I asked it for Olympic volleyball competitors, it chose the Sports Illustrated style for women and a live match for men.

While it's great to celebrate the advancements of these image-generation models, I think they still need a lot of human guidance. 

You can try out Black Forest AI here.