Why AI Companies Are Selling at a Loss on Purpose

If a product as expensive to run as an AI model is being handed to you for free or pennies, it is not a gift. It is a hook. Joshua Gold calls it the free bag of crack strategy, and it is the same playbook Oracle ran decades ago. Get you integrated, then raise the price.

In this episode, I talked with Joshua Gold, CEO of thebigword. The Big Word is the company you have never heard of that runs in the background of some of the most critical moments in your life. Every court in the United Kingdom is contracted with them. About half the police stations in the Netherlands use their interpreters. They run Medicare and Medicaid over-the-phone interpreting in the US, and they have had human interpreters on the ground in Afghanistan, Iraq, and Syria under military contracts. Their workforce sits around 15,000 people across more than 50 countries.

Joshua was not supposed to be CEO. He dropped out of college after two weeks, took a sales job at a beer company to learn how to sell, then joined a night shift at thebigword pitching translation services to US banks. He worked his way up, moved to New York right into the teeth of the 2008 financial crisis, pivoted the business to public sector and healthcare because the government was the only one spending, and ended up running a company his father had spent 40 years building. The business has tripled since he took over.

Here is the part that caught me. Thebigword has taken operations cost from 20 cents in the dollar down to 5 cents through workflow automation and AI. That is a 15-point margin swing, not from layoffs, from building AI self-driving workflows into the work itself. And yet they are not worried about machine translation eating the industry, because 99 percent of content translated online is already done by machine. They make their living on the 1 percent that cannot be wrong. Legal witness statements. Medical consent forms. Military briefings. Court testimony. The kind of work where 95 percent accurate is not good enough.

The scale of content is what really drives this home. The total volume of written content online doubles every 15 months. All human-created content, from the invention of writing up to today, doubles again in 15 months, and again after that. So even if the machine-translated share stays at 99 percent, the 1 percent that needs a human growing at that rate is still a compounding business. Their industry has grown around 6 percent a year even while AI has radically changed it since neural machine learning came out in 2013.

Now, to the pricing piece. In the last five years, revenue per head at the global 500 is up 31 percent. If you had told me five years ago that companies were about to squeeze 31 percent more revenue out of every employee, I would have bet you a ginormous recession was coming. Instead we got the lowest unemployment in living memory. The only explanation that fits is AI. And Joshua is careful with the framing. Since World War II, businesses have averaged about 1.5 percent efficiency gain per year. This is not an AI revolution. It is AI evolution, slightly accelerated.

The Achilles' heel Joshua pointed to is where I think a lot of small business owners should pay attention. LLMs now train on an internet where most of the new content is AI-generated. They are tuning themselves on themselves. The music analogy he used stuck with me. All the AI music models were trained on MP3 and Spotify audio, already compressed, already missing 70 percent of the original frequencies. So what comes out is a watered-down average of a watered-down average. Good enough for my car speakers. Not good enough for someone with real headphones. Businesses are going to have to decide which one they serve.

The other story I keep thinking about is the car shop one. My dad runs a car shop, and a friend of his came in after the dealership quoted $15,000 for a new suspension. My dad raced the car, listened, and made an educated guess based on decades of experience. Two small rods, $15 each, slightly bent, causing the whole symptom. He swapped them, the issue disappeared. AI has the knowledge. It can read every forum and service manual. What it cannot do yet is lift the car, try things, and feel when something is off. Joshua calls this the difference between knowledge and understanding, and he thinks experience is about to be the most valuable thing in the workplace again.

At the end of the day, if you run a small business, you do not need to out-AI the AI companies. You need to figure out which slice of your work is the 1 percent that has to be right, and protect it with human judgment while you let AI eat the 99 percent that does not. And you should probably budget for those free AI tools to stop being free.