Are marketers using AI totally wrong? | Harley Allaby, Samurai Code

There is a stat going around right now that should make every founder and marketing leader uncomfortable.

According to a study from MIT, recently highlighted by Fortune, 95% of AI implementations that companies have attempted on their own have failed. Not underperformed. Not "needs improvement." Failed.

The remaining 5% that actually worked? They all had one thing in common: AI experts were involved in the implementation.

That is the starting point of a conversation I had with Harley Allaby on a recent episode of Marketing by Design. Harley is the co-founder of Samurai Code, an AI implementation company based in Columbus, Ohio. Before launching Samurai Code, he helped scale Exact Medicare from a small team doing about a million dollars in revenue to over 800 people and $100 million in revenue in three years.

The engine behind that growth? AI agents his team built from scratch.

So when Harley talks about why AI fails inside most businesses, he is not speculating. He has seen it from both sides. He built the thing that worked, and now he walks into companies every week where the thing did not work.

This article breaks down the core lessons from that conversation.

The Sword vs. the Samurai

Harley has a metaphor he uses with clients that cuts right to the problem.

Most businesses that try to adopt AI end up buying a sword they do not know how to use, instead of hiring a samurai.

What he means is this: a founder reads an article, watches a YouTube video, or sees a competitor post about some new AI tool. They get excited. They buy the tool. They try to plug it into their business. And it falls flat.

This is not because the tool is bad. It is because the tool was never designed for their specific business. It was a generic product, or worse, a traditional SaaS product that had AI squeezed into it as a feature rather than built around it as a foundation.

Harley put it plainly in our conversation: "What we're seeing happen is organizations are like, okay, we know we need to use AI, so let's just go and do it. And then they go and get this tool that is supposed to be the AI tool that's gonna change their lives, change their business, and it falls on its face."

The tool is the sword. Without the expertise to wield it, you are just swinging at air.

The Real Problem Is Not the Technology

Here is where it gets interesting. The failure is almost never a technology problem. It is a strategy problem.

Big companies, the Fortune 500 types, can afford to burn time, resources, and money figuring out an AI solution through trial and error. They have the risk capital to set money on fire, learn from the wreckage, and try again. They know that once they crack it, the return will be enormous.

Small and mid-sized businesses do not have that luxury. They cannot absorb a six-figure experiment that produces nothing. They need it to work the first time, or at least the second.

That gap is where most of the failure lives. It is not that smaller companies are less capable. It is that they are applying a big-company experimentation model to a small-company budget. They are trying to figure out AI by themselves because they assume that is what everyone else is doing.

And in most cases, everyone else is doing the same thing. Badly.

Where Harley Starts: Constraint or Quick Win

When Samurai Code begins working with a business, Harley described two paths they evaluate.

The first path is the high-leverage option. This means identifying the single biggest constraint in the business and pointing AI directly at it. For example, if a company has more leads than their sales team can handle, the business will not grow until that bottleneck is solved. That is a high-leverage spot to deploy an AI agent because the ROI is immediate and measurable.

The second path is the low-hanging fruit. Sometimes the biggest constraint is also the hardest problem to solve. It might take months. In those cases, Harley looks for a simpler win first. Something that saves money or time quickly, gives the team a taste of what is possible, and builds confidence before tackling the bigger project.

Both approaches are valid. The point is that neither one starts with "let's buy a tool and see what happens." Both start with understanding the business first.

This is the part most DIY implementations skip entirely.

Your Data Is Probably a Mess

Even when a business identifies the right problem to solve, there is another obstacle that Harley runs into constantly: the data is not ready.

Most small businesses do not have clean, organized, accessible data. They are running on a patchwork of spreadsheets, legacy software, disconnected platforms, and tribal knowledge that lives in someone's head.

Before an AI agent can do anything meaningful, it needs data it can reference and act on. If the data is scattered across five different systems that do not talk to each other, you are not deploying AI. You are doing data cleanup with extra steps.

Harley was direct about this: "The things that tend to create the most resistance for us up front are having your data in a working order. The average small business does not have proper databases in place."

He also flagged something that gets overlooked. It is not enough for the data to be accessible. The systems need to communicate. Cloud-based software with actual APIs and endpoints is the baseline. If your tools do not integrate natively, you may need middleware just to get them talking before the AI layer can sit on top.

This is not the exciting part. Nobody posts about data hygiene on LinkedIn. But it is the part that determines whether your AI investment produces results or produces frustration.

Small Language Models Over Large Ones

One of the more tactical insights from the episode was Harley's explanation of why smaller, focused AI agents consistently outperform one large, monolithic agent.

He shared an example from a car dealership project. The instinct was to train one massive agent on everything: every car model, every financing option, every service record, every piece of inventory. The result was a bloated context window that struggled to surface the right information at the right time. The outputs were inconsistent.

The fix was to break it apart. Instead of one giant agent, they built several smaller ones. One agent was an expert on car models. Another handled financing. Another managed service. Those agents could communicate with each other when needed, but each one stayed focused on its domain.

The result was dramatically better performance.

This principle, using small language models instead of large ones, is an emerging trend in the AI space. Microsoft recently demonstrated that a collection of small, specialized language models outperformed doctors in identifying and diagnosing medical conditions. The takeaway is counterintuitive but consistent: more focus beats more data.

For business owners, the lesson is practical. If someone is pitching you a single AI tool that claims to do everything, be skeptical. The best implementations are modular, specific, and tailored to how your business actually operates.

The Competitive Reality Nobody Wants to Talk About

I asked Harley a question that I think a lot of people are quietly wondering: are my competitors already doing this?

His answer was nuanced but honest.

If your competitors are publicly traded Fortune 500 companies, yes. They have been investing in AI for years and they have teams dedicated to it. You are behind, and AI is giving them the ability to come for your lunch more aggressively than ever before.

If your competitors are other mid-sized or local businesses, most of them are in the same spot you are. They know they should be doing something. It is on their list. They have not started yet.

That means there is still a window. If you act now, you can establish a real speed advantage over your local competition. But that window is closing.

Harley framed the opportunity around a stat that stuck with me: "We're seeing companies today grow without adding headcount. That is unheard of. You're seeing startups hit revenue per employee numbers of $500,000."

Five hundred thousand dollars in revenue per employee. That number was not possible before AI entered the picture. For lean operators, this is not a theoretical future. It is happening now, inside companies that figured out where to point the technology.

What You Can Do Before You Spend a Dollar

If you are a founder or marketing leader reading this and thinking about where to start, here is what I took away from the conversation with Harley.

Get your data in order. Before you hire anyone or buy any tool, audit where your data lives. Is it centralized? Can your systems talk to each other? If the answer is no, that is your first project.

Identify your constraint. What is the one thing preventing your business from growing right now? Is it lead volume? Lead qualification? Onboarding speed? Customer service load? That constraint is likely your highest-leverage AI opportunity.

Stop trying to make your business fit a tool. The whole point of AI is that it can be tailored to your business. If you are changing your workflows to accommodate a product, you have it backwards.

Start small if you need to. You do not have to solve the biggest problem first. Pick a low-risk use case, build confidence, and learn how your team interacts with AI before you tackle the mission-critical stuff.

Talk to someone who has done it. The 5% success rate is not random. Those companies had experts guiding the implementation. You do not need to figure this out alone, and honestly, the data says you probably should not try.

The Bottom Line

AI is not magic. It is not a silver bullet. And buying a tool off the shelf is not an AI strategy.

The businesses getting real results are the ones treating AI the way they would treat any serious operational investment: with strategy, specificity, and expert guidance. They are not chasing trends. They are solving specific problems inside their business and building systems that fit how they actually work.

That is what Harley and the team at Samurai Code are doing for small and mid-sized businesses. And it is the mindset every founder needs to adopt if they want to be in that 5%.

You can listen to the full conversation with Harley on Marketing by Design wherever you get your podcasts.

‍

Marketing by Design is powered by the American Marketing Association Columbus Chapter and produced by MMG Design. Each episode breaks down how high-performing marketers and founders built their success and told their stories, by design.

*Get the Visual Swipe File: https://www.mmg.studio/visualswipefile

‍