• by Laura McElhinney , Op-Ed Contributor, March 26, 2025

Like you, when I absently type a prompt into one of the LLMs (“large language models” like ChatGPT, Claude, Gemini, and the latest shiny new thing, DeepSeek), I expect magic and brilliance to be presented to me instantaneously.

Sometimes, my expectations are met. There are even times when the value of the results exceeds my hopes. Sometimes, it takes several prompts to get a usable response.

Then there are those other times, when well… let's just say I get garbage: made-up statistics and answers (aka “hallucinations”) or just gibberish. 

It all comes down to the clarity and quality of the prompts and the data I use to direct the LLM or other artificial intelligence (AI) tools.

Suffice it to say that AI is absolutely fast and it's certainly smart. But signs of its intelligence reflect what is being asked of it and what information is being used to yield actionable outputs.

As AI continues to disrupt the advertising industry's creative and analytics production, we all have a lot more thinking to do about how intelligence happens, artificial and otherwise.

AI's Inconvenient Truth

Everyone uses AI -- whether they have a real reason and grounding in the technology or not. Our colleagues and our competitors use it.

Therefore, we must too. This is the age where the fear of being left behind is sharply intense.

But like anything else, using any tool without training or clear direction will only produce poor results. 

In the rush to implement artificial intelligence, all businesses are discovering an inconvenient truth: Even the most sophisticated AI systems are only as good as the data they're built upon. 

While headlines focus on AI's creative capabilities and potential to automate data gathering and analysis, the less glamorous -- although crucial -- foundation of successful AI operations lies in data preparation, organization, and quality control.

We All Know It, And Yet…

If your foundation is cracked, don't expect AI to build a skyscraper that will stand against the gentlest breeze. 

Think of a photographer working with corrupted image files. “Oh, AI will fix it.” Um, no, it can’t. AI can only enhance what is already there in the image file.

The same goes for any source material. If your image data is degraded, every filter and adjustment only magnifies the flaws. 

With AI, poor data quality doesn't just blur the picture — it fundamentally distorts the content of every campaign, every insight, and every decision that is being automated. 

Yes, our careers and our businesses depend on mastering these systems. But “mastering” means ensuring crystal-clear inputs before we ever press the shutter or input our data analytics.

Given the constant acceleration of AI upgrades, not to mention the veritable newness of AI, no one can credibly call themselves an expert. But we’re all supposed to act like we are. That has infected much of the use of AI. 

You can ask an AI model to give you any outcome you want. That doesn’t mean you’ll actually get the result you’re hoping for. It's like being a parent — if you keep asking your kids the same open-ended question, they will simply tell you what you want to hear.

This dynamic creates a significant challenge for organizations eager to leverage AI's great potential.

The promise of AI to revolutionize everything from customer segmentation to predictive analytics is compelling.

But without proper data governance and preparation, these initiatives risk delivering unreliable or misleading results.

Data Readiness Can’t Be Automated

That’s a matter of human intelligence being behind the machine to guide it.

The path to effective AI implementation requires a fundamental shift in how organizations approach their data strategy.

Rather than viewing AI as a plug-and-play solution, companies need to first address their existing data challenges. 

If you don't solve the issues you have today, AI will not fix it for you. You have to solve the root cause. This means establishing robust processes for data collection, validation, and maintenance before diving into AI implementations.

Organizations must ensure data readiness. That means more than just accessing abundant data. It also entails information that is relevant, current, and free from bias. This requires ongoing curation and validation — a process that cannot be automated away.

Consider the parallel with self-driving vehicles. While the technology is impressive, the challenges of real-world implementation highlight how difficult it is for AI systems to account for every possible scenario.

Self-driving cars may become mainstream in five or 10 years —  we’ll see. But for right now, autonomous vehicles require someone’s hands on the wheel and their attention focused on what is going on around them on the road.

Similar challenges exist in business applications — consumer behaviors change, market conditions shift, and new variables emerge that were not present in historical data.

The solution is not to abandon AI initiatives. Instead, advertisers, agencies, and data science specialists need to approach AI with a clear understanding of what's required for success. 

Start by establishing strong data governance frameworks. Before attempting to apply AI solutions, organizations need clear protocols for data collection, storage, and maintenance. This includes addressing security concerns and ensuring compliance with relevant regulations.

Second, it’s imperative to secure buy-in from every part of the organization. AI initiatives require support across departments, as data quality affects — and is affected by — multiple stakeholders. This means getting executive management on board and fostering collaboration between technical and business teams.

Third, implementing robust training and evaluation regimens. Organizations need human expertise to validate AI outputs against expected results. The data simply isn’t ready to drive itself now, if ever. 

The advertising industry at large tends to view AI as the world's fastest production house. Give it award-winning storyboards, and it creates masterpieces that can be easily targeted at scale. Give it confused briefs and mixed messages, inaccurate or incomplete data, and it mass-produces magnificent failures.

When our data lacks clarity or truth, we're not just making missteps — we're automating them across every channel, every platform, every touchpoint.

At least for the foreseeable future, AI needs human experts, and human intelligence to keep it honest and on track. That is the job now.