We have decades of proprietary operational data that we want to use to train a custom machine learning model, but the development costs are high. How do we use absolute and relative valuation frameworks during strategic planning to evaluate if this investment will actually increase our exit multiple?

Question

Tyler Smith · Accepted Answer

Training a custom model on proprietary data sounds sophisticated, but you must evaluate it as a cold business investment. You need to determine if the development costs will yield a positive return on investment using both absolute and relative valuation frameworks.

First, look at relative valuation. Buyers pay premium multiples for companies that possess defensible intellectual property. If your custom model allows you to deliver results that competitors cannot replicate because they lack your decades of historical data, you have built a strong moat. This proprietary technology shift can elevate your company from a standard services multiple to a tech-enabled platform multiple, significantly increasing your enterprise value on the market.

Second, evaluate the project through absolute valuation, which focuses on discounted future cash flows. Will this custom model dramatically lower your operating costs, or will it allow you to charge premium prices for superior predictions? If the model merely automates a process that could be handled just as well by a cheaper, off-the-shelf API, the absolute value is low. The cash flow improvement will not justify the high development and maintenance costs.

Use Keith Cunningham's Thinking Time to analyze this trade-off. Ask yourself: Is the primary value of this model its ability to generate higher near-term cash flow, or is it a strategic asset designed to attract a specific strategic buyer?

If you cannot tie the custom model directly to a measurable increase in cash flow or a highly defensible IP asset that a buyer will pay a premium multiple for, do not build it. Stick to cheaper, third-party integrations.