Stop spending AI tokens on trash

Stop spending AI tokens on trash

One of the biggest misconceptions I encounter about AI is that success comes down to writing better prompts.

Prompting certainly matters, but it isn’t what determines whether an AI initiative delivers business value. In my experience working with customers, the deciding factor is much simpler: the quality of the data feeding the AI.

You can craft the perfect prompt, but if the underlying information is outdated, incomplete, irrelevant, or poorly governed, you’ll still get poor results. That’s why I often tell customers they’re spending their AI tokens on trash. Every query consumes resources, and every inaccurate response wastes time, money, and trust.

The conversation shouldn’t start with prompts. It should start with data.

AI needs the right data, not more data

Many organizations assume AI needs access to more information. I disagree. What AI really needs is access to the right information.

An HR assistant doesn’t need engineering documentation. A finance chatbot shouldn’t search marketing presentations. Every AI application should be connected to the data that supports its specific purpose.

That starts by asking a simple question: What problem are we trying to solve?

Once the business objective is clear, identifying the systems and data required becomes much easier. Instead of exposing AI to everything, organizations can focus on providing relevant, up-to-date, and trustworthy information that yields meaningful outcomes.

That’s how you maximize the value of every token.

Responsible AI starts with responsible data

Responsible AI is often discussed in terms of governance, transparency, and ethics. Those conversations are important, but they overlook a critical prerequisite: responsible data.

Data doesn’t remain accurate forever.

Engineering documentation changes. Internal policies evolve. Product fixes are released. Business processes improve. Information that was correct six months ago may already be obsolete.

That’s why data integrity can’t be treated as a one-time cleanup effort before launching an AI project. It must become an ongoing discipline. Organizations need to continuously evaluate whether their information is accurate, up to date, secure, and appropriate for the AI systems that consume it.

When the quality of your data declines, the quality of your AI declines with it.

Data ownership isn’t just an IT responsibility

One of the biggest organizational challenges I see is confusion over who owns data.

Too often, businesses assume this responsibility belongs entirely to IT because IT manages the infrastructure. But storage teams aren’t responsible for determining whether an HR policy is current or whether an engineering document is still valid.

I like to compare infrastructure teams to the postal service. They know how to deliver the envelope safely and securely, but they aren’t responsible for what’s written inside.

The people who create the data understand its meaning. The business understands how it’s used. Infrastructure teams ensure it’s available, protected, and accessible.

Successful AI requires all three groups working together.

I prefer to think of this as data custodianship rather than ownership. Everyone has a role in maintaining the quality, security, and usefulness of the organization’s most valuable asset: its data.

Business leaders must be part of the AI conversation

AI projects are often treated as technology initiatives when they’re really business initiatives.

The most successful projects I’ve seen involve business leaders from the very beginning. They define the desired outcome, establish success metrics, and identify where AI can create measurable value.

That alignment changes the conversation.

Instead of asking IT to “implement AI,” organizations begin solving specific business problems, whether that’s reducing manual work, improving customer experiences, or helping employees make faster decisions.

It also creates buy-in across the organization because everyone understands what’s in it for them.

Employees don’t want AI to replace meaningful work. They want it to eliminate repetitive tasks so they can spend more time solving problems, serving customers, and creating value.

Don’t forget about scale

A successful pilot doesn’t guarantee a successful deployment.

I’ve seen organizations build AI solutions that work beautifully for a handful of users, only to discover that supporting thousands of employees dramatically changes the economics.

More users generate more prompts. More prompts consume more tokens. Costs increase quickly if scalability wasn’t part of the original plan.

That’s why organizations should think beyond proving that AI works. They need to prove that it works efficiently at production scale without creating unsustainable costs or disappointing users.

Planning for scale early prevents unpleasant surprises later.

Start with your data

If you’re launching an AI initiative, my advice is straightforward.

Start by defining the business outcome you’re trying to achieve. Then identify the data needed to support that goal and ensure it’s clean, current, secure, and governed. Bring together the business leaders, data creators, and infrastructure teams so everyone understands their role in maintaining data quality and measuring success.

Only then should you focus on prompts.

The organizations seeing the greatest return from AI aren’t necessarily using the newest models or writing the cleverest prompts. They’re the ones investing in trusted data, shared accountability, and clear business objectives.

Because at the end of the day, AI can only be as valuable as the information you give it. When your data is clean, relevant, and well governed, your tokens become an investment instead of an expense, and that’s when AI starts delivering the business outcomes everyone is looking for.