How to update a decision when new evidence arrives
The essentials in 30 seconds
- To update a decision when new evidence arrives, multiply the prior probability by how likely the evidence is under success, then rescale.
- In the illustrative case, a prior of 0.30 rises to 0.53 after a second meeting and 0.87 after a term sheet.
- Three weeks of silence then lower it to 0.75.
- A commit rule of 0.70 turns each updated probability into an action.
- The EQOS Asia Sentinel service recomputes the model monthly (EQOS Asia method page, as of October 2026).
Decisions rarely stay still. A partner replies, a competitor moves, a pilot returns data, and the original probability of success no longer fits. Bayesian updating gives a disciplined way to revise that probability as evidence arrives. This guide shows the arithmetic on an illustrative partnership and connects the habit to the wider set of decision making frameworks.
On this page
Compare the EQOS Asia offers and request a 30-minute pre-diagnostic.
What is Bayesian updating and why does a decision need it?
Bayesian updating revises the probability of a hypothesis when new evidence arrives. It starts from a prior, the probability held before the evidence. It asks how likely the evidence would be if the hypothesis were true and how likely if it were false. The revised probability, called the posterior, combines the three. Executives use the same logic informally whenever they say that a fact raises or lowers their confidence.
The formula for a yes or no hypothesis is short: posterior equals prior times the likelihood under success, divided by the total probability of the evidence. A decision maker needs only four numbers and a calculator. The result tells how strongly one new fact should move a belief, which a gut feeling cannot do reliably.
How do you set the prior probability for a decision?
The prior comes from experience with similar cases. If three in ten comparable partnership talks end in a signed deal, 0.30 is a fair starting point for a new one. Starting from the base rate protects against the pull of a vivid story about the particular partner. The prior also tells the team how much evidence a decision will need before it moves.
State the prior before collecting the evidence and write it down. A prior adjusted after the fact is just a conclusion. If the team disagrees, run the calculation with two priors, such as 0.20 and 0.40, and see how far apart the posteriors land after each piece of evidence.
The arithmetic behind the update has a name. Bayes’ theorem for business decisions sets out the formula, shows why natural frequencies make it easy to apply, and explains how a screening flag with high accuracy can still leave the hypothesis far from certain when the base rate is low.
How do you update a probability after a first piece of evidence?
Take a partnership talk, all numbers illustrative. The prior chance of a signed deal is 0.30. The partner agrees to a second meeting. Such an agreement occurs in 80 percent of deals that close and in 30 percent of deals that fail. The two likelihoods differ by a factor of almost three, which makes the meeting a useful signal.
Multiply the prior by the first likelihood: 0.30 × 0.80 = 0.24. Multiply the complement by the second: 0.70 × 0.30 = 0.21. The total is 0.45, so the posterior is 0.24 divided by 0.45, or 0.53. One meeting moved the probability from 0.30 to 0.53.
A second full case helps the habit take hold. A worked example of Bayesian updating follows a decision through several pieces of evidence, with a prior, likelihoods, and a posterior recorded at each step, so that a team can copy the layout for its own choice.
How do a second and a third piece of evidence move the probability?
The posterior becomes the next prior. A written term sheet arrives, which occurs in 60 percent of deals that close and 10 percent of those that fail. The calculation is 0.533 × 0.60 = 0.320 and 0.467 × 0.10 = 0.047, so the posterior is 0.320 divided by 0.367, or 0.87.
Evidence can also lower the probability. Three weeks pass without a reply, which occurs in 30 percent of deals that close and 70 percent of those that fail. Then 0.873 × 0.30 = 0.262 and 0.127 × 0.70 = 0.089, so the posterior is 0.262 divided by 0.351, or 0.75. Each step takes a minute and leaves a record that a colleague can audit.
How do you turn an updated probability into a decision?
A probability becomes a decision through a threshold set in advance. Suppose the team commits a costly hire only when the chance of a signed deal reaches 0.70. After the second meeting the probability of 0.53 sits below the line, so the team waits. The line reflects the cost of being wrong in each direction and the appetite for risk.
After the term sheet, 0.87 clears the line and the hire goes ahead. After the silence, 0.75 still clears it, so the commitment stands and the team schedules a follow-up call. Agreeing the threshold before the evidence arrives keeps the reading of each update honest.
Which mistakes weaken a Bayesian update?
Four habits weaken the method. Counting the same fact twice inflates the posterior. Using likelihoods that nobody has checked turns arithmetic into decoration. Treating a vivid anecdote as strong evidence overweights it. Skipping the base rate returns a number that feels precise and starts from nothing.
A quick safeguard is to write the likelihood pair next to each piece of evidence, with its source. When the two likelihoods are close, such as 0.50 and 0.45, the evidence is weak and the probability barely moves. A review of past updates shows which pieces of evidence earned their weight. Over a year, the log builds a track record for the team.
How does the EQOS Asia decision engine keep a decision current?
A decision gains value when its score follows the facts. The EQOS Asia method supports this with the Sentinel service: the model is reusable monthly, and a recompute each month refreshes the score and raises real-time alerts when a threshold is approached (EQOS Asia method page, as of October 2026).
The engine recomputes deterministically from updated inputs, which differs from the Bayesian revision above but serves the same purpose. A reader can try a first score in the simulator, and the 30-minute pre-diagnostic, scheduled in Singapore time (SGT), sets the inputs worth tracking. Teams in Singapore, ASEAN, and Hong Kong can review the score on a fixed monthly rhythm.
See also in this guide
Frequently asked questions
What is a prior in Bayesian updating?
A prior is the probability held before the new evidence arrives. It comes from the base rate of similar cases, such as the share of comparable talks that end in a signed deal. Writing it down first keeps the update honest and lets a team compare reasoning later.
How do I get the likelihoods for an update?
Estimate how often the evidence appears when the hypothesis is true and how often when it is false, using past cases or expert judgment. State the source and a range. The ratio between the two numbers shows how strong the evidence is.
Can evidence lower a probability?
Yes. Evidence that appears more often when the hypothesis is false lowers the probability. In the example, three weeks of silence move the chance of a signed deal from 0.87 to 0.75. The commitment still stands because the result stays above the 0.70 rule.
Does the EQOS Asia method update with new data?
The Sentinel service recomputes the model monthly and sends real-time alerts when a threshold is approached (EQOS Asia method page, as of October 2026). The recompute uses the deterministic core, so the same updated inputs always return the same score, which makes each change traceable.
Test your knowledge
1. What is a prior?
- The probability held before new evidence
- The cost of the decision
- The final recommendation
Show the answer
The probability held before new evidence.
2. What is the posterior after the second meeting in the example?
- 0.30
- 0.87
- 0.53
Show the answer
0.53.
3. What lowers the probability from 0.87 to 0.75 in the example?
- A term sheet
- Three weeks of silence
- A second meeting
Show the answer
Three weeks of silence.
4. What is the commit threshold in the example?
- 0.50
- 0.90
- 0.70
Show the answer
0.70.
5. How often does the EQOS Asia Sentinel service recompute the model?
- Monthly
- Weekly
- Yearly
Show the answer
Monthly.
About the author
Stéphane Jambu
President, EQOS Forecast Asia
Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile
Sources: EQOS Asia method page (eqos-asia.com/method.html), simulator page (eqos-asia.com/simulator.html), and contact page (eqos-asia.com/contact.html), consulted on 21 October 2026. Page updated on 21 October 2026.