ask when VOI exceeds attention cost
Your sales team wants to run a $15,000 customer survey before you commit $200,000 to a new product feature. Your gut says just build the thing. Their gut says you need data first. Neither gut feeling is a number - and you need a number, because $15,000 buys you a quarter of an engineer.
Value of Information is the gap between your Expected Value with better data and your Expected Value without it. If that gap exceeds the cost of getting the data - in dollars, time, or attention - go get it. If not, decide now and move.
Value of Information (VOI) answers one question: how much should I pay to reduce uncertainty before I commit?
You already know how to build a decision tree and collapse it with Expected Value. VOI adds one layer: what if you could get better probabilities for the uncertain outcomes in your tree before you pick a branch?
Formally:
VOI = E[X with better information] − E[X with current information]
If VOI > cost of getting the information, go get it. If VOI < cost, decide now with what you have.
The "cost" is not just dollars. It includes the opportunity cost of delayed Execution, the attention of whoever has to run the study, and any advantage you lose by waiting (a competitor ships first, a vendor deal expires, a candidate takes another offer).
Every P&L decision is made under uncertainty. The temptation is binary: either you agonize and over-research, or you "move fast" and skip diligence entirely. Both destroy value.
Over-researching burns Budget on studies, delays Revenue, and often yields data you never actually use in the decision. This is the classic failure mode of large organizations - twelve-week analysis cycles for decisions that were obvious at week two.
Under-researching is the startup failure mode. You commit Capital Investment based on vibes, then discover the market was telling you something you could have learned for $5,000 and two weeks.
VOI gives you a decision rule for when to stop thinking and start doing. That is worth more than most frameworks, because the meta-decision - should I gather more data or act now? - is the one Operators get wrong most often.
On the P&L, VOI shows up as:
Start with a decision tree you have already built. You are at the root, looking at your branches.
Step 1: Compute your Expected Value with current information.
This is just the standard decision tree collapse. You assign your best-guess probabilities to each uncertain outcome, compute Expected Value at each branch, and pick the branch that wins under your decision rule. Call this number EV_now.
Step 2: Ask what "perfect information" would change.
Imagine an oracle tells you exactly which outcome will happen before you choose. Now you can always pick the best branch. Compute the Expected Value of that scenario - weighted by the probability of each state the oracle might reveal. Call this EV_perfect.
VOI_perfect = EV_perfect − EV_now
This is the ceiling - the maximum you should ever pay for information about this decision. No survey, pilot, or consultant can beat perfect information.
Step 3: Adjust for imperfect information.
Real data is noisy. A customer survey might predict actual Demand with 70% accuracy. A pilot launch in one city tells you something about national Demand, but not everything. So the real VOI sits between zero and VOI_perfect:
0 ≤ VOI_actual ≤ VOI_perfect
You estimate VOI_actual by updating your probabilities using the information source's historical accuracy - what fraction of the time it correctly predicts the actual outcome - then re-running your decision tree with those updated probabilities. For each possible signal the source could send, compute the posterior probability of each outcome, find the best action given that posterior, and weight by the probability of receiving that signal.
Critically: if the updated probabilities never change which branch you pick, VOI is zero regardless of what the source costs. This is the most common and most useful result. Check for it first before grinding through the full calculation.
Step 4: Compare VOI to total cost.
Total cost = dollar spend + opportunity cost of delay + attention cost.
If VOI_actual > total cost, get the information.
If VOI_actual < total cost, decide now.
If they are close, your decision is robust either way - just pick and move.
Use VOI when:
Skip VOI when:
You run a SaaS product at $2M ARR. You are deciding whether to build an enterprise tier (12 weeks, $180,000 Implementation Cost). Your decision tree has two uncertain outcomes: enterprises adopt (you estimate 55% probability, yielding $400,000 incremental Revenue in year one) or they do not (45% probability, yielding $20,000 from a handful of small upgrades). Your sales team proposes a $12,000 survey of 200 prospects that historically predicts adoption with 75% accuracy.
EV_now (no survey): (0.55 × $400,000) + (0.45 × $20,000) − $180,000 = $220,000 + $9,000 − $180,000 = $49,000. Your best current option is to build.
EV_perfect: If the oracle says 'adopt' (55% chance), you build and net $400,000 − $180,000 = $220,000. If the oracle says 'no adopt' (45% chance), you skip the build and net $0. EV_perfect = (0.55 × $220,000) + (0.45 × $0) = $121,000.
VOI_perfect = $121,000 − $49,000 = $72,000. This is the ceiling - perfect knowledge is worth up to $72K.
VOI_actual (75% accurate survey): P(survey says 'adopt') = (0.75 × 0.55) + (0.25 × 0.45) = 0.4125 + 0.1125 = 0.525. When the survey says 'adopt,' the updated probability of adoption is 0.4125 / 0.525 ≈ 0.786. EV of building given that signal: (0.786 × $220,000) + (0.214 × −$160,000) = $172,857 − $34,286 = $138,571. You build. P(survey says 'no adopt') = 0.475. Updated probability of adoption drops to 0.1375 / 0.475 ≈ 0.289. EV of building given that signal: (0.289 × $220,000) + (0.711 × −$160,000) = $63,684 − $113,684 = −$50,000. You skip - building has negative Expected Value at those odds. EV_with_survey = (0.525 × $138,571) + (0.475 × $0) = $72,750.
VOI_actual = $72,750 − $49,000 = $23,750. The survey costs $12,000 plus roughly $3,000 in opportunity cost (two weeks of delayed launch). Total cost = $15,000. Since $23,750 > $15,000, run the survey.
Insight: The survey is worth running because it changes your action in the 'no adopt' scenario - you save $180K in wasted build cost roughly a third of the time. Notice that VOI comes entirely from the cases where the signal flips your decision. If the signal never flips it, VOI is zero.
Same SaaS company. You are deciding whether to add a Slack integration (2 weeks, $15,000 Implementation Cost). You estimate 80% chance it increases Expansion Revenue by $50,000 and 20% chance it adds only $5,000. An $8,000 usage study could tell you which scenario is more likely (70% accuracy).
EV_now: (0.80 × $50,000) + (0.20 × $5,000) − $15,000 = $40,000 + $1,000 − $15,000 = $26,000. Build it.
EV_perfect: Oracle says 'high' (80%): build, net $35,000. Oracle says 'low' (20%): don't build, because $5,000 − $15,000 = −$10,000. Skip, net $0. EV_perfect = (0.80 × $35,000) + (0.20 × $0) = $28,000.
VOI_perfect = $28,000 − $26,000 = $2,000. The ceiling on any information is $2,000. The study costs $8,000. Do not run the study. Even perfect knowledge is not worth the price.
Insight: When your best option wins by a wide margin across scenarios, information has low value. The probabilities would have to shift dramatically to change your decision - and the study cannot shift them that much. This is the most common VOI result: you discover you should just decide.
VOI is the difference between your Expected Value with better data and without it - if that gap is less than the cost of getting the data, stop researching and commit.
The maximum VOI (perfect information) sets a hard ceiling - never spend more than that on any study, pilot, or consultant for this decision.
VOI is zero whenever the information would not change your action. If you would build the feature regardless of the survey result, the survey is pure waste - even if the uncertainty makes you uncomfortable.
Ignoring the cost of delay. Operators compute dollar cost of a study but forget the opportunity cost of weeks lost. A $5,000 survey that delays a $200K Revenue launch by a month has a real cost far above $5,000 - you need to include the time value of that delayed Cash Flow.
Treating all uncertainty as worth resolving. Some uncertain outcomes in your decision tree do not affect which branch you pick. If the Expected Value gap between 'build' and 'don't build' is $200K under every plausible scenario, no information changes the decision. Spending Budget to narrow that uncertainty is a failure mode disguised as diligence.
You are considering hiring a specialized data engineer at $160,000 in annual Total Compensation (salary, benefits, taxes, and equipment). You estimate a 60% chance the hire accelerates a data pipeline that generates $300,000 in Revenue and a 40% chance the pipeline is not viable regardless of staffing (Revenue = $0). A two-week paid contract trial with the candidate costs $18,000. The trial predicts pipeline viability with 80% accuracy. Should you run the trial?
Hint: First compute EV_now for the hire/no-hire decision. Then compute EV_perfect. Check if the ceiling exceeds $18,000 before computing VOI_actual.
EV_now: Hire: (0.60 × $300,000) + (0.40 × $0) − $160,000 = $180,000 − $160,000 = $20,000. Don't hire: $0. Best option: hire, EV = $20,000. EV_perfect: Oracle says viable (60%): hire, net $140,000. Oracle says not viable (40%): don't hire, net $0. EV_perfect = (0.60 × $140,000) = $84,000. VOI_perfect = $84,000 − $20,000 = $64,000. Ceiling is well above $18,000, so the trial could be worth it. VOI_actual (80% accurate): P(trial says 'viable') = (0.60 × 0.80) + (0.40 × 0.20) = 0.48 + 0.08 = 0.56. When the trial says 'viable,' P(viable|trial=viable) = 0.48/0.56 ≈ 0.857. EV of hiring = (0.857 × $140,000) + (0.143 × −$160,000) = $120,000 − $22,857 = $97,143. Hire. P(trial says 'not viable') = 0.44. P(viable|trial=not viable) = 0.12/0.44 ≈ 0.273. EV of hiring = (0.273 × $140,000) + (0.727 × −$160,000) = $38,182 − $116,364 = −$78,182. Don't hire. EV_with_trial = (0.56 × $97,143) + (0.44 × $0) = $54,400. VOI_actual = $54,400 − $20,000 = $34,400. Since $34,400 > $18,000, run the trial. The trial's value comes from avoiding the $160K hire when the pipeline is not viable - the 'not viable' signal flips your decision from 'hire' to 'don't hire.'
Your marketing team wants to spend $25,000 on A/B testing before committing $100,000 in Marketing Spend to a new channel. You estimate a 50/50 chance the channel returns 3x ($300K Revenue) or 0.5x ($50K Revenue). The A/B test predicts the true outcome with 65% accuracy. Is the test worth it?
Hint: Compute EV for both 'invest' and 'don't invest' first. Then check: after updating your probabilities with the worst signal the test could send, would you still invest? If so, what does that tell you about VOI?
EV_now: Invest: (0.50 × $300,000) + (0.50 × $50,000) − $100,000 = $175,000 − $100,000 = $75,000. Don't invest: $0. Best: invest, EV = $75,000. EV_perfect: Oracle says 3x (50%): invest, net $200,000. Oracle says 0.5x (50%): skip, because $50,000 − $100,000 = −$50,000. Net $0. EV_perfect = (0.50 × $200,000) + (0.50 × $0) = $100,000. VOI_perfect = $100,000 − $75,000 = $25,000. The ceiling exactly equals the test cost - so only a near-perfect test could justify the spend. Now check whether the 65% test ever changes your action. P(test says '0.5x') = (0.35 × 0.50) + (0.65 × 0.50) = 0.50. When the test sends the worst signal ('0.5x'): P(3x|test=0.5x) = (0.35 × 0.50) / 0.50 = 0.35. EV of investing given that signal = (0.35 × $200,000) + (0.65 × −$50,000) = $70,000 − $32,500 = $37,500 > $0. You still invest. When the test sends the best signal ('3x'): P(3x|test=3x) = 0.65. EV of investing = (0.65 × $200,000) + (0.35 × −$50,000) = $130,000 − $17,500 = $112,500 > $0. You invest. You invest regardless of what the test says. VOI_actual = $0. Do not run the test. The 65% accuracy is too weak to overcome the asymmetry between upside ($200K) and downside ($50K). Even the worst signal leaves investing with positive Expected Value. The test would need accuracy above 80% before the '0.5x' signal could flip your invest decision. Below that threshold, the information is not just overpriced - it is literally worthless.
Value of Information is the natural next step after you have internalized Expected Value and the decision tree. Expected Value taught you to collapse uncertain outcomes into a single number. The decision tree taught you to map sequential choices and uncertain outcomes into a structure you can compute over. VOI adds the meta-question: before you collapse that tree and commit, is there a data source that would change which branch you pick - and is it worth the cost? This connects forward to Sensitivity Analysis (which parameters actually move the needle), Risk-Adjusted Value (when the Variance of outcomes matters, not just the mean), and Capital Budgeting (where VOI determines whether to run pilots before committing large Capital Investment). It also reinforces opportunity cost - every dollar and week spent gathering information is a dollar and week not spent on Execution.
Disclaimer: This content is for educational and informational purposes only and does not constitute financial, investment, tax, or legal advice. It is not a recommendation to buy, sell, or hold any security or financial product. You should consult a qualified financial advisor, tax professional, or attorney before making financial decisions. Past performance is not indicative of future results. The author is not a registered investment advisor, broker-dealer, or financial planner.