Why the Classic ROI Model Fails on Feature Purchases
Look: most product managers still treat a buy-feature like a simple add-on, plugging it into a linear ROI calculator. That’s a rookie mistake. The real math lives in expected value, not just cost-plus-margin. When you ignore the probabilistic payoff, you’re basically gambling with your budget blindfolded.
Crunching the Numbers: Expected Value 101
Here is the deal: expected value (EV) = Σ (probability × payoff). In the context of a feature, probability is the likelihood that the feature will hit a usage threshold, and payoff is the incremental revenue it unlocks. If a feature costs $50k but only has a 20% chance to generate $300k in extra sales, the EV is 0.2 × 300k = $60k. Subtract the cost and you’re looking at a $10k net gain — profitable.
Don’t Forget the Hidden Costs
And here is why: development time, support tickets, and opportunity cost are rarely baked into the headline price. Throw a 15% overhead fudge factor on the $50k, and the net EV slides to $5k. Still positive, but the margin is razor-thin. That’s why a quick sanity check on hidden costs can save you from a costly misfire.
When EV Turns Negative Fast
Imagine a feature that promises a flashy UI upgrade but only 5% of users will actually engage with it. If the payoff per engaged user is $1,000, EV = 0.05 × 1,000 = $50. With a $20k development bill, you’re staring at a $19,950 loss. No amount of hype can mask that math.
Real-World Example: Bonus Buy Slots
Take the case of bonus-buy slots in online gaming. Players can pay to skip the grind and unlock a bonus round instantly. The operator’s EV hinges on the ratio of premium buys to the expected payout of that round. If the payout is calibrated at 85% of the buy-price, the house still walks away with a 15% edge — profitable on volume alone. See the detailed breakdown in this expected value buying feature profitable analysis.
Quick Test Before You Commit
Run a three-step sanity test: 1) Estimate probability distribution from user data. 2) Multiply by projected incremental revenue per user. 3) Subtract total cost, including hidden overhead. If the result is positive, green light the feature. If not, scrap or iterate. No fluff, just numbers.