Key Takeaways

  • Mercor CPO Oswald Nitski insists that the perceived ROI problem for enterprise AI isn't a problem at all. Instead, it's a necessary period of exploration and experimentation where companies have more patience for returns.
  • The future of AI value lies in highly specialized models tailored for individual companies. These models will require unique, enterprise-specific evaluation and training data to perform in their exact setting.
  • AI spend, currently around 3% of developer salaries at some firms, is set to increase significantly across the industry. This is because much of this spend (like on coding agents) offers compounding gains, not just direct costs of goods sold.
  • Despite concerns about spend, companies like Mercor are experiencing intense demand, indicating a healthy market where the value of AI solutions often outpaces the ability to scale delivery.

Enterprise AI ROI: It's Not a Problem, It's a Phase

Founders often hear whispers about enterprise AI's poor return on investment. But Oswald Nitski, CPO at Mercor, pushes back hard on this. He sees it differently: “I don't think there's an ROI problem right now. I think we're in a period of exploration and experimentation where there's more tolerance, more patience to get that ROI calculation right now.”

This isn't just semantics; it's a crucial reframe. Instead of viewing AI investments as underperforming, Nitski suggests we're simply in the early innings. Companies are still figuring out the best use cases, collecting initial data, and learning what works. This phase demands patience, not panic. The early adopters, like a startup exploring a new market, know the path isn't linear. They expect to experiment and iterate before hitting a clear, quantifiable ROI, a tolerance that's surprisingly high today.

The Future Is Hyper-Specialized: Your Data, Your Model

If current AI tools feel generic, that's because many are. But Nitski sees a fast shift. The next wave of value comes from models built specifically for one company, one context. He states, “I buy it [specialized models for every company]. I think it's also self-serving towards Mercor in that we think that every specialized model will need enterprise specific eval training data to show the model how to perform in its setting.”

This means the true competitive advantage won't just be accessing a powerful base model. It will be feeding that model your unique, proprietary, enterprise-specific data to create something tailored. Think of it less like buying a ready-made suit and more like hiring a bespoke tailor. This is where companies like Mercor, which scales human data annotation, play a vital role. They help businesses gather and refine the specific data needed to make these hyper-specialized models perform brilliantly, turning latent demand into measurable gains.

Optimizing AI Spend: Look Beyond Direct Costs

When you count AI spend, are you looking at the right things? Nitski challenges the conventional wisdom that all AI costs should be measured as direct cost of goods sold (COGS). He points out that investments like Salesforce's in Anthropic, or a company's spend on coding agents, deliver different kinds of value. “It totally depends on the use case,” Nitski says. “When you're looking at coding agent spend for your software engineers, that's not always like that's not cogs for your work... That doesn't like if that's really high, that could still be giving you compounding gains.”

This perspective is key. AI expenditure, while significant (Mercor's own AI spend is high relative to salaries), often contributes to productivity gains that compound over time, rather than just directly reducing immediate costs. Nitski believes this trend will only grow. “I hope that we can move towards a future of better accounting of the outcomes being driven by token spend,” he reflects, predicting that “macro the percentage will increase over time to to more than 3% [of developer salaries].” Despite these rising costs, the market is hot. Mercor itself ends “every week with so much more money in the bank like the business is is very healthy and we can't we can't spend money fast enough to service all of the demand that we have.”

What to Do With This

Stop viewing early AI spend as an ROI failure. Instead, identify a specific problem within your business that could benefit from a highly specialized AI model. Dedicate resources to gather your unique enterprise data for training and evaluation, then pilot the project, measuring both direct cost savings and compounding productivity gains.