Portfolio Probability Management — Reading the Entire Pipeline with Bayes

"Sales that only look at the trees miss the forest. Only when you view the entire forest through probability does strategy emerge."

Introduction: From Individual Deals to the Entire Pipeline

In Series Part 1, we looked at how the win probability P(Win) of an individual deal is calculated. However, a real-world sales organization doesn't manage just 1 deal. On the desk of the Head of Sales, tens or hundreds of deals are progressing simultaneously.

"This quarter's revenue target is 50 billion won. Can we achieve it with the deals currently in the pipeline?"

The traditional answer to this question is a simple summation of individual sales reps' self-assessments:

Deal	Rep's Estimated Probability	Deal Value	Expected Contribution
Company A ERP Implementation	80%	12B	9.6B
Company B Cloud Migration	60%	8B	4.8B
Company C Security Solution	90%	5B	4.5B
Total		25B	18.9B

18.9 billion? It falls far short of the 50 billion target. But how much can we trust this number? Who decided the "80%" for Company A? Is that 80% an overestimate or an underestimate?

Bayesian portfolio management provides a mathematically honest answer to all these questions.

Part 1: Bayesian Calculation of Expected Revenue

1.1 The Basic Formula: Probability-Weighted Revenue

If EXAWin's P(Win) is calculated for each deal $i$ , the Expected Revenue of the entire pipeline is:

E[\text{Revenue}] = \sum_{i=1}^{N} P_i(\text{Win}) \times V_i

Here, $V_i$ is the expected contract value of deal $i$ .

1.2 3-Tier Forecasting Reflecting Uncertainty

However, P(Win) alone is not enough. Utilizing the Credible Interval learned in Series Part 1, a 3-tier forecast reflecting the uncertainty of the entire pipeline is possible.

Using the 95% credible interval $[P_{i,lower}, P_{i,upper}]$ for each deal:

\text{Conservative Forecast} = \sum_{i=1}^{N} P_{i,lower} \times V_i

\text{Expected Forecast} = \sum_{i=1}^{N} P_i(\text{Win}) \times V_i

\text{Optimistic Forecast} = \sum_{i=1}^{N} P_{i,upper} \times V_i

1.3 Scenario: A 5-Deal Pipeline

Deal	Deal Value	P(Win)	α + β	CI Lower	CI Upper	Maturity
Company A ERP	12B	78.5%	41.0	66.1%	91.0%	🌳 Mature
Company B Cloud	8B	62.3%	12.5	35.8%	88.8%	🌿 Growing
Company C Security	5B	85.2%	55.0	75.9%	94.5%	🌳 Mature
Company D AI	20B	45.1%	8.0	15.6%	74.6%	🌱 Early
Company E Maintenance	3B	91.3%	120.0	86.2%	96.4%	🌳 Mature

3-Tier Revenue Forecast:

\text{Conservative} = 120 \times 0.661 + 80 \times 0.358 + 50 \times 0.759 + 200 \times 0.156 + 30 \times 0.862

= 79.3 + 28.6 + 38.0 + 31.2 + 25.9 = \textbf{20.30 Billion}

\text{Expected} = 120 \times 0.785 + 80 \times 0.623 + 50 \times 0.852 + 200 \times 0.451 + 30 \times 0.913

= 94.2 + 49.8 + 42.6 + 90.2 + 27.4 = \textbf{30.42 Billion}

\text{Optimistic} = 120 \times 0.910 + 80 \times 0.888 + 50 \times 0.945 + 200 \times 0.746 + 30 \times 0.964

= 109.2 + 71.0 + 47.3 + 149.2 + 28.9 = \textbf{40.56 Billion}

Management Report: "Based on the current pipeline, the expected revenue for this quarter is 30.4 billion won. Conservatively, it is around 20.3 billion, and optimistically, 40.6 billion. To achieve the 50 billion target, we need to secure approximately 10~20 billion in new deals."

This is the Bayesian approach, reporting the range of uncertainty alongside a single number ("30.4 billion"). Management can make risk-aware decisions between the conservative scenario (20.3B) and the optimistic scenario (40.6B).

Part 2: Deal Priority Matrix — Where to Pour Our Energy

2.1 P(Win) × Deal Value Matrix

Sales resources are finite. 5 sales reps cannot manage 50 deals simultaneously. The key is which deals to focus on.

We classify them using a 2D matrix:

	High Deal Value (≥ 10B)	Low Deal Value (< 10B)
High P(Win) (≥ 70%)	🔴 Top Priority: Immersive Investment	🟡 Efficiency: Fast Closing
Low P(Win) (< 70%)	🟠 Strategy: Intensive Nurturing or Abandon	🟢 Observe: Minimal Management

Applying this to the 5 deals above:

Company A ERP (12B, 78.5%) → 🔴 Top Priority
Company D AI (20B, 45.1%) → 🟠 Strategy (Value is highest but probability is low → Needs intensive nurturing)
Company C Security (5B, 85.2%) → 🟡 Efficiency (Wrap up contract quickly)
Company B Cloud (8B, 62.3%) → 🟡/🟠 Borderline (Needs additional signals)
Company E Maintenance (3B, 91.3%) → 🟡 Efficiency (Almost confirmed, minimal management)

2.2 Adjusting Priorities Considering Evidence Maturity

Even if P(Win) is high, if the evidence maturity is 🌱 Early, the probability is unstable. For the Company D AI deal (20B, 45.1%, 🌱 Early), the CI is [15.6%, 74.6%], an extremely wide range.

\text{Expected Value} = P(\text{Win}) \times V = 0.451 \times 200 = 9.02B

However, in the conservative scenario:

\text{Conservative} = P_{lower} \times V = 0.156 \times 200 = 3.12B

The gap between the expected value of 9.02B and the conservative 3.12B is 5.9B. The larger this gap, the greater the risk.

EXAWin displays this as Expected Value Volatility (EVV):

\text{EVV} = (P_{upper} - P_{lower}) \times V

Deal	EVV	Interpretation
Company D AI	(0.746 - 0.156) × 200 = 11.80B	Highly unstable, urgent need for more evidence
Company B Cloud	(0.888 - 0.358) × 80 = 4.24B	Unstable, needs evidence collection
Company A ERP	(0.910 - 0.661) × 120 = 2.99B	Stable, focus on closing
Company C Security	(0.945 - 0.759) × 50 = 0.93B	Very stable, wrap-up stage
Company E Maintenance	(0.964 - 0.862) × 30 = 0.31B	Confirmed, just manage

Strategic Insight: The Company D AI deal has an expected value of 9B, but an EVV of 11.8B. This is a strong warning: "This single deal could determine the quarter's performance, but we still lack evidence." The best sales personnel must be assigned to Company D to quickly raise the evidence maturity from 🌱→🌿→🌳.

Part 3: Optimal Resource Allocation — The Math of Limited Time

3.1 Marginal Expected Value

We can calculate the marginal expected value that 1 hour of a sales rep's time has on each deal. For a deal with a P(Win) around 50%, 1 hour of additional activity might raise the probability by 3~5%p. However, for a deal already at 90%, investing another hour might not even raise it by 1%p.

This is consistent with the Law of Diminishing Returns:

\text{MEV}_i = \frac{\partial P_i}{\partial t} \times V_i

The MEV is highest for deals where P(Win) is in the middle range (40~60%). Investing time in already certain deals or already abandoned deals is inefficient.

3.2 Practical Allocation Example

If out of a 40-hour week, the available time for sales activities = 30 hours:

Deal	P(Win) Range	Allocated Time	Strategy
Company D AI	45.1% (Mid)	12 hours	Intensive Nurturing — Technical PoC, meeting with decision-makers
Company B Cloud	62.3% (Mid)	8 hours	Signal Collection — Additional demo, quote adjustment
Company A ERP	78.5% (High)	5 hours	Closing — Final terms negotiation
Company C Security	85.2% (High)	3 hours	Wrap-up — Contract review
Company E Maintenance	91.3% (Almost confirmed)	2 hours	Management — Check signature schedule

Core Principle: "Don't waste time on certain deals. Invest in high-value deals among the uncertain ones." This is the essence of probability-based resource allocation.

Part 4: Pipeline Health Metrics

4.1 Weighted Pipeline Coverage

Traditional pipeline coverage is simply the ratio of "Total Pipeline / Target":

\text{Simple Coverage} = \frac{\sum V_i}{\text{Target}} = \frac{48B}{50B} = 0.96x

However, this number ignores P(Win). Bayesian weighted coverage is:

\text{Weighted Coverage} = \frac{\sum P_i \times V_i}{\text{Target}} = \frac{30.42B}{50B} = 0.61x

And conservative coverage:

\text{Conservative Coverage} = \frac{\sum P_{i,lower} \times V_i}{\text{Target}} = \frac{20.30B}{50B} = 0.41x

Management Interpretation: A simple coverage of 0.96x looks "almost achievable," but a weighted coverage of 0.61x means "a significant gap exists," and a conservative coverage of 0.41x implies "serious risk." Reality speaks through probabilities, not just numbers.

4.2 Goal Achievement Probability

More precisely, we can directly calculate the "probability of achieving the 50 billion target." Assuming each deal is independent, the distribution of total revenue becomes the sum of individual Bernoulli random variables.

Using a Monte Carlo simulation with 10,000 iterations:

In each simulation, deal $i$ is won (revenue $V_i$ ) with probability $P_i$ , and lost (revenue 0) with probability $(1 - P_i)$ .
The ratio of times the total revenue is ≥ 50 billion out of 10,000 = Goal Achievement Probability

The probability of achieving 50 billion with the current pipeline = approx. 12.3%.

"How can we raise the achievement probability to over 50%?" → Raising the P(Win) of the Company D AI deal from 45% → 70% = 34.7% → + Adding 2 new 10B scale deals (P(Win) 50%) = 51.2%

Part 5: Portfolio Rebalancing — When to Abandon and When to Push

5.1 Cut-off Threshold

You can't push every deal to the end. A deal where P(Win) continuously drops and falls below 30% even when the evidence maturity is 🌳 Mature is effectively lost.

EXAWin's recommendation conditions for abandonment:

P(Win) < 25% AND Evidence Maturity ≥ 🌿 Growing
S⁻ > S⁺ in 3 consecutive meetings (Negative signals overwhelm positive ones)
Momentum P(Win) is 15%p or more below the base P(Win) (Recent trend worsening)

5.2 Nurture Priority

Conversely, there are deals where P(Win) is still low but the growth potential is large:

P(Win) = 30~50% AND Evidence Maturity 🌱 Early (Still in the early stages of evidence collection)
Top 20% in deal value
At least 1 SA 5.0 signal in a recent meeting

These deals have the potential for a sharp rise in P(Win) with just 1~2 key meetings. This is strategically leveraging a core characteristic of Bayesian analysis — a single early piece of evidence can cause a massive shift in probability.

Part 6: Time-Series Portfolio Tracking — Diagnosing the Forest's Health Weekly

6.1 Weekly Pipeline Dashboard

Every Monday, the Head of Sales checks the following metrics on a single screen:

Metric	This Week	Last Week	Change	Signal
Total Deals	48	45	+3	🟢 New Inflow
Weighted Pipeline	30.42B	28.7B	+1.72B	🟢 Growth
Conservative Pipeline	20.30B	19.8B	+0.5B	🟡 Slight Increase
Goal Achievement Probability	12.3%	10.8%	+1.5%p	🟡 Improving
Average P(Win)	58.2%	56.7%	+1.5%p	🟢 Healthy
Deals with P(Win) < 30%	8	7	+1	🔴 Abandon Candidates Rising
Silence Penalty Warnings	5	3	+2	🔴 Urgent Follow-up

6.2 Trend Analysis: Is the Pipeline Getting Healthier?

Weighted pipeline trend over 4 weeks:

Week 1: 24.5B
Week 2: 26.7B
Week 3: 28.7B
Week 4: 30.4B

Linear trend: approx. +2B per week. If this trend is maintained, it will be around 46B after 8 weeks. It approaches the 50B target but falls short.

Strategic Judgment: "The current trend is good, but achieving the target requires additional stimulus. An increase in P(Win) for the Company D AI deal or an influx of large new deals is the key variable."

Conclusion: Eyes to See the Forest

The P(Win) of an individual deal is the health status of a single tree. But what the Head of Sales needs is the health of the entire forest.

Bayesian portfolio management enables the following:

Honest Revenue Forecasting: Presenting not just the expected value, but the conservative/optimistic ranges.
Optimal Resource Allocation: Focusing energy on deals with the highest marginal expected value.
Early Risk Detection: Identifying unstable deals through EVV (Expected Value Volatility).
Strategic Decision Making: Providing objective criteria for abandoning, nurturing, and closing.
Building Management Trust: A culture of conversing with probabilities, not gut feelings.

In Series Part 3, we will cover Competitive Analysis using Conditional Probability — "If a competitor enters this deal, how much does our P(Win) change?"

Author: EXA Bayesian Research Lab
Published in: EXAWin Technical Series — Vol. 2
Keywords: #BayesianPortfolio #PipelineManagement #SalesForecast #ResourceOptimization #EXAWin

BA04-3. [Series Part 2] Portfolio Probability Management — Reading the Entire Pipeline with Bayes