A Leader’s Guide To Value Max Enterprise AI

Ameya Kanitkar is the Co-founder and CTO of Larridin, a Bay Area-based startup building an organizational platform powered by AI.

getty​As enterprise AI enters a new phase, the focus on accessing the newest model, the largest context window or the strongest benchmark performance is now giving way to maximizing value.

As AI spending grows with its capacity and the variety of models available, leaders are asking a more practical question: What is the most cost-effective combination of models, infrastructure and human expertise that can reliably produce the business outcome we need?

In truth, enterprises are spending substantially more on AI while expecting to extract more output value from fewer people. Worldwide end-user spending on AI models and platforms is projected to increase 63% in 2026 compared to 2025. At the same time, Gartner predicts 60% of organizations will adopt smaller software engineering teams by 2029.

That growth makes AI spending harder to justify at the board level, pushing enterprise leaders to rethink which models can best perform at each task. Increasingly, the goal is to build a portfolio that matches capability and cost to the value of the work.

This correlates to workforce planning. Companies do not need a senior professional to copy information between spreadsheets. Senior expertise is responsible for solving problems that justify the expense and lower-cost resources are typically assigned work that does not.

AI should be managed the same way as enterprise expertise.

Every leading AI model has different strengths. Some perform better at complex reasoning and planning. Others may be more economical for coding, research, summarization, extraction or repetitive workflow tasks. Companies are increasingly aware of that; however, value-maxxing does not mean simply selecting the least expensive model.

A lower-cost model that repeatedly produces the wrong answer, requires additional prompting or creates hours of human review may ultimately cost more than a premium model that completes the task correctly the first time.

The objective should be to use the least expensive model that is capable of reliably delivering the required outcome. That means companies tend to build portfolios of models rather than one universal AI standard.

Many AI purchasing decisions still focus on token prices. But token price tells leaders very little about the true economics of a workflow.

A more useful measure is cost per successful outcome. That includes model costs, employee time, latency, retries, human review, error rates and the value produced by the completed work. A model that costs half as much per token but requires three attempts is not necessarily less expensive. Neither is a frontier model delivering a five-cent email rewrite that another system could have produced for a fraction of the cost.

Value maxxing requires organizations to understand what they are paying for an outcome, not simply what they are paying for compute.

Open-source models give enterprises another lever to cut costs and reduce dependence on proprietary providers. Many companies still prefer competitive U.S.-developed models over Chinese alternatives for security, governance and procurement reasons.

But benchmarks alone are no longer enough. As public tests saturate, more teams are evaluating models on their own tasks and data.

The right model is not the one with the best benchmark score, but the one that delivers the best economics for the work the company actually does.

The same value-maxxing logic applies to infrastructure.

Given the pace of model development, buying dedicated hardware for most AI workloads can create unnecessary rigidity. A company may make a large capital investment around one model architecture only to find that a substantially better or more efficient model appears months later.

For most enterprises, flexibility has value. Unless a company requires a completely isolated, airtight environment because of security, regulatory or operational requirements, using models through hyperscalers or specialized AI clouds can make more economic sense.

These providers can give companies access to open models without requiring them to purchase and manage their own GPU fleets.

The goal is optionality: the ability to move workloads as model performance and economics change rather than being forced to maximize utilization of hardware the company already owns.

Beyond the technological considerations of value maxxing, the human element matters too. Employees also need to know when different levels of AI capability are appropriate.

Proprietary data from our AI measurement platform found that workplace productivity increased 1.5 times over six months, but those gains were highly uneven. The most AI-proficient users increased output 2.5 times, while the bottom 25% saw no improvement.

The defining skill is increasingly AI resource management: knowing how to frame the task, choose the appropriate model, evaluate the result and escalate to a more expensive system only when necessary.

Giving employees access to AI without the skills or tools to choose the right model for each task can drive up costs without delivering proportional value.

Evaluate models against real company workflows and route different tasks to the systems that deliver the best combination of quality, speed and cost.

Use cloud infrastructure where possible so models can be changed as performance and economics improve.

Track model usage, completion rates, retries, human review, switching costs and business outcomes. Optimize total cost per outcome rather than token prices in isolation.

The competitive advantage will come from how intelligently companies allocate models as new capabilities emerge and regulations continue to evolve. The companies able to continuously match the cost of AI to the value of the task will likely see better ROI for their AI.

That is enterprise AI value maxxing.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?