The Executive Decision Series (Post 5): Markov Models for LTV

Executive Takeaway: Markov models matter because they don’t just show where customers or deals are today—they reveal how groups are likely to move tomorrow—toward growth, retention, or churn. They reveal not just the what (likelihood of churn, revenue forecast) but the pathways—how customers and deals actually flow over time. That creates runway for action, giving leaders prescriptive insight into where to intervene, when to act, and which levers will change outcomes before it’s too late.

What Executives Need to Know About Markov Models

In business, outcomes don’t happen in one step. Deals move through stages, customers move between segments, and value unfolds across time. A Markov Model is a way of capturing those movements.

At its core, a Markov Model is a map of states (or stages) and probabilities:

States are conditions customers or deals can be in (e.g., High, Medium, Low, or Churned).
Probabilities describe how likely it is to move from one state to another next period (e.g. weekly, monthly, quarterly).

That’s it. Simple. The power comes from what happens when you follow those probabilities forward: you can calculate, with precision, the most likely outcomes, the value of different paths, and how interventions (like improving retention or boosting engagement) change the picture.

Why Markov Models Matter for Executives

Executives don’t need equations; they need early warning. Markov models turn today’s state mix into tomorrow’s likely distribution, giving you the runway to intervene before value is lost.

Most dashboards stop at descriptive statistics: counts, averages, churn rates. Useful, but backward-looking. By the time someone shows up in “Churned,” it’s already too late to act.

A Markov Model goes further. It’s a mass probability model, which means it doesn’t predict what any one customer will do, but rather how a group of customers in the same state today will distribute across states in the future. In other words, it shows the flows — who is likely to slip toward Churn, and who still has a path back to High.

For example, if you place 1,000 customers in At-Risk today, the model can tell you how many, on average, will still be At-Risk, how many will recover to High, and how many will churn over the next 1, 3, or 5 months. Those probabilities play out over time, revealing the most likely journeys customers will follow.

That probabilistic forecast gives executives time to act. Instead of reacting after value is lost, you see risks and opportunities while there’s still leverage — whether that means investing in retention, pushing upgrades, or shifting resources to where they’ll have the biggest lift.

Bottom line: Markov shifts you from describing the present to shaping the future—providing prescriptive insight on where to intervene, when to act, and which levers will change what happens next.

Executives don’t need the equations. What matters is the idea: value is not static, it moves. And Markov Models let us see — and act on — those movements.

A Simple Example: Customer Lifetime Value (LTV)

Everyone knows LTV — the total value a customer generates over time. But many LTV formulas are too simplistic: one average revenue number × one churn rate. They assume customers never change.

This simple “average churn” formula ignores how behavior changes over time. In real life, churn isn’t a flat constant: it’s often front-loaded, with a higher risk in the early months (trial fatigue, onboarding friction) that tapers later as remaining customers become more committed. Revenue can shift, too—some customers ramp up spend as they adopt more features, while others cool off or seasonally fluctuate. A single “average churn” and “average revenue” gloss over these dynamics, masking where—and when—you actually win or lose value.

An RFM lens (Recency, Frequency, Monetary Value), for example is but one method that lets us model that movement and value it properly.

An RFM Lens on Customer Behavior

Before we model value, we need a way to summarize how customers behave over time. One widely used method—especially in digital and ecommerce—is RFM, which captures the underlying dynamics of a customer relationship:

Recency — how recently a customer engaged or purchased. (Fresh activity signals momentum; long gaps signal risk.)
Frequency — how often they engage or purchase in a period. (Regular, repeated actions indicate habit and stickiness.)
Monetary Value — how much they spend when they do engage. (Higher ticket or basket size lifts impact even at the same frequency.)

I won’t dive into the mechanics of RFM today. What matters is that RFM tracks changes in engagement over time—who’s heating up, cooling off, or at risk. With those signals (i.e. an RFM score and rank) in hand, we can project likely trajectories as probabilistic outcomes over different time horizons, giving the business runway to act (e.g., rescue At-Risk customers, nudge Mediums to High) before outcomes harden.

States (RFM-style)

We combine the three RFM dimensions into a score and rank that we can track over time. This makes it easy to group customers into five states that reflect relationship quality and risk:

S1 — High (Active/High-value): recently active, high spend/frequency
S2 — Medium: engaged, but not top tier
S3 — Low: sporadic, low monetary value
S4 — At-Risk / Lapsing: drifting away; likely to churn soon
S5 — Churned: absorbed (no return in this simple version)

How the Markov Model Works

To build a model like this, we first have to analyze historical customer data to establish how people actually progress along the journey. This gives us the baseline probabilities—how often customers stay in place, upgrade, downgrade, or churn. And although it’s not visually exciting, a simple table is the cleanest way to capture this information before any math is involved. Each row answers the question: “given where a customer is today, where are they likely to be next month?”—whether that’s staying in the same state, moving back up to High

Each month, a customer can stay in their current state, upgrade back to High (S1), downgrade toward Churn (S5), or (for Churn) remain Churned. The probabilities are summarized in a transition matrix (rows = current state, columns = next state).

From \ To	S1	S2	S3	S4	S5
S1	0.75	0.25	0.00	0.00	0.00
S2	0.20	0.00	0.80	0.00	0.00
S3	0.10	0.00	0.00	0.90	0.00
S4	0.05	0.00	0.00	0.00	0.95
S5	0.00	0.00	0.00	0.00	1.00

For example, the S4 row:

5% → S1 (they come back and buy),
95% → S5 (they churn and stay there).

The transition matrix above is the blueprint for how customers flow over time through your journey. Each row says, “given where a customer is today, where are they likely to be next month?”—whether that’s staying put, upgrading back to High, slipping toward At-Risk, or exiting to Churn.

Drilldown: click to see how you would go about generating RFM Scores from Historical data

1. Track Each Customer’s RFM Score Over Time

Every month (or week/quarter depending on the business), you calculate each customer’s RFM score and rank.
Example: In January, Customer A is “Medium” (S2). In February, they make several purchases and move up to “High” (S1). In March, they go quiet and slip to “Low” (S3).

By doing this across all customers, you create a time series of states for each individual.

2. Count Transitions Between States2. Count Transitions Between States

Next, you look at how customers moved from one month to the next.
Example: In February, 200 customers were in Medium (S2). By March:
- 40 upgraded to High (S1),
- 120 stayed Medium (S2),
- 30 slipped to Low (S3),
- 10 dropped into At-Risk (S4).

You tally these moves for every state → state pair.

3. Convert Counts Into Probabilities

For each starting state, you divide by the total number of customers that month to get transition probabilities.
In the example above for Medium (S2):
- P(S2→S1) = 40 ÷ 200 = 0.20
- P(S2→S2) = 120 ÷ 200 = 0.60
- P(S2→S3) = 30 ÷ 200 = 0.15
- P(S2→S4) = 10 ÷ 200 = 0.05

Now you’ve built the transition matrix. Each row represents the “from” state, and each column the “to” state. Rows sum to 1.

4. Use the Transition Matrix to Project Forward

When you attach dollar rewards to each state, the same matrix turns into a forecast of expected value (LTV).

With this matrix, you can now say: “Given a customer is Medium today, here’s the probability they’ll be High, Medium, Low, At-Risk, or Churned in 1 month, 3 months, 5 months.”

Looking Beyond Next Month: The 5-Period Forecast

But what matters for LTV is not just the next step—it’s how those moves play out over time. When we extend the model forward, the one-month probabilities compound into a multi-month forecast. This lets us see how today’s customer base is likely to shift across High, Medium, Low, and Churn as months go by.

For example, here’s what the picture looks like five months ahead. Notice how the weight steadily drifts toward Churn (S5), unless customers are pulled back into High (S1) along the way. That’s the story of lifetime value in action: retention, upgrades, and churn, all happening together in the same flow.

From \ To	S1	S2	S3	S4	S5
S1	0.377	0.111	0.103	0.110	0.299
S2	0.152	0.045	0.044	0.041	0.718
S3	0.067	0.020	0.019	0.022	0.872
S4	0.022	0.006	0.006	0.007	0.959
S5	0.000	0.000	0.000	0.000	1.000

Furthermore, when you attach a dollar value to arriving in each state, the same matrix translates those flows into expected revenue (LTV)—and makes the key levers obvious (e.g., lifting Medium→High or rescuing At-Risk→High) to move value, not just metrics.

Attaching Dollars (Rewards)

To turn states into LTV, we attach a dollar value to arriving in each state this month:

State	Reward on Arrival (This Month)
S1 (High)	100
S2 (Medium)	-20
S3 (Low)	-20
S4 (At-Risk)	-20
S5 (Churned)	0

S1 (High): +$100 net (think: $120 revenue – $20 cost)
S2–S4 (Medium/Low/At-Risk): –$20 (cost to serve without a purchase)
S5 (Churned): $0

We discount future months by 20% annually (≈1.7% per month) to reflect time value and risk.

Think of the transition matrix as the weights for what happens next month. Each row (your current state) multiplies the reward table by the likelihood of landing in each next state, and when you sum those weighted rewards you get the expected value for that month. Do the same step again for future months (applying a discount each month), and you’ve turned state movement into a forward-looking, dollar-valued forecast of LTV.

5-Period LTV Results

After applying the transition probabilities for five months, the reward values for each segment are:

Start State	5-Period Value (Discounted, $)
S1 (High)	227.19
S2 (Medium)	56.98
S3 (Low)	23.01
S4 (At-Risk)	14.74
S5 (Churned)	0.00

How to interpret:

If a customer is in S1 (High) today, their expected value over the next 5 months is $227.19.
If they’re in S2 (Medium) today, it’s $56.98.
If they’re in S3 (Low) today, it’s $23.01.
If they’re in S4 (At-Risk) today, it’s $14.74.
If they’re in S5 (Churned) today, it’s $0.

Even though only Highs (S1) generate positive income directly, the other tiers still carry positive expected value because they have a chance of moving upward over the five-period horizon. In fact, the only way Medium, Low, and At-Risk customers generate future value is if some of them eventually hit High at least once.

For example, some Mediums (S2) will upgrade to High before churning, and that pathway generates meaningful revenue. Lows (S3) and At-Risk (S4) are worth less, but their value is not zero because a fraction of them are rescued or reactivated into higher states. In other words, their score reflects not today’s income, but the weighted average of all the possible journeys they might still take — and as long as some of those journeys lead back to High, the expected value stays above zero.

What the Results Mean

The table above is the bottom line: if a customer starts in one of these states today, this is the expected revenue they will generate over the next five months, after accounting for movement between states, the likelihood of churn, and a 20% annual discount (≈1.7% per month) for time-value and risk.

High (S1) customers generate the most value. They’re sticky, keep paying, and have the longest runway.
Medium (S2) customers are still valuable, with upside if some move back to High.
Low (S3) customers have limited value unless they upgrade.
At-Risk (S4) customers are worth little unless they’re rescued back to High—and most won’t be.
Churned (S5) customers contribute virtually nothing going forward.

Scenarios: Using the 5-Period Forecast to Create Runway

Scenario 1 — Rescuing At-Risk (S4)

Start with 1,000 At-Risk customers today.

Baseline expected value (per S4): $14.74 → $14,740 per 1,000.
If a retention program doubles S4→S1 this month (from 5% → 10%, shifting probability from Churn), the 5-period value per S4 rises to $29.56.
Incremental lift: +$14.82 per customer → ~$14,822 per 1,000.

Scenario	Value per S4 ($)	Value per 1,000 ($)	Incremental Gain ($)
Baseline	14.74	14,740	–
With Retention Lift	29.56	29,561	+14,822

Why this works: more S4s jump back into High (S1) sooner, producing positive net rewards instead of drifting into Churn (S5). That’s runway in action.

Scenario 2 — Nurturing Mediums (S2)

Now take 1,000 Mediums.

Baseline expected value (per S2): $56.98 → $56,980 per 1,000.
A targeted program that delivers a +10% relative lift in S2→S1 (from 20% → 22%, with S2→S3 reduced to keep the row sum at 1) raises the 5-period value per S2 to $63.53.
Incremental lift: +$6.55 per customer → ~$6,554 per 1,000.

Scenario	Value per S2 ($)	Value per 1,000 ($)	Incremental Gain ($)
Baseline	56.98	56,980	–
With Nurture Lift	63.53	63,533	+6,554

Why this works: nudging more Mediums into High compounds into more time in the value-creating state (S1), even over just five periods.

Remember: These aren’t predictions about individual customers; they’re cohort-level forecasts that show where a whole segment is headed — and how much value is on the line if you intervene now. Markov gives you the where, when, and how much to act with confidence.

Where Else Markov Models Are Used in Business

Markov Models show up in more places than you might expect. Whenever outcomes unfold across states over time, they’re a natural fit. A few examples:

Multi-Touch Attribution (Marketing): modeling how customers move through touchpoints on the path to conversion, and which channels drive incremental impact.
Pipeline Forecasting (Sales): deals progress across opportunity stages; probabilities of conversion are derived from historical flows.
Customer Retention & Churn Prediction: capturing how users slide between “engaged,” “dormant,” and “churned” states, with interventions to rescue at-risk segments.
Product Adoption Journeys: tracking how customers move from free to basic to premium tiers, and the revenue implications of those transitions.
Credit Risk & Finance: modeling borrower states (current, delinquent, default) to estimate losses and manage portfolio health.
Operations & Support: predicting how service tickets or cases flow through queues and resolutions, to improve staffing and turnaround times.

The pattern is always the same: you have states, transitions, and rewards. Markov Models turn those building blocks into actionable foresight.

Executive Takeaway

Value isn’t static — it moves. For example, customers shift between High, Medium, Low, and At-Risk before some inevitably churn. Deals move forward, stall, or fall out of the pipeline. Markov Models give us a simple but powerful way to capture that movement: states, transitions, and rewards.

Unlike the traditional one-line LTV formula (average revenue ÷ churn), this approach reflects real customer dynamics: churn risk is often front-loaded, upgrades and downgrades matter, and interventions can change the trajectory.

By analyzing historical data, we can map the probabilities of moving between states and assign dollar values to each outcome. The result is a forward-looking forecast of both value and risk — one that tells us not just where customers are today, but where they’re likely to be tomorrow.

For Executives, the message is clear:

See the hidden flows behind your KPIs.
Quantify future value instead of relying on static averages.
Act on leverage points — like rescuing At-Risk customers or nudging Mediums to High — to unlock outsized gains.

If your business depends on customers, deals, or processes that move through stages, Markov Models can help provide the clarity to anticipate outcomes and the runway to take action.