What it means
The key idea is called the Markov property: the future depends only on the present state. If a customer buys Brand A this month, the chance they buy Brand A or Brand B next month is the same whether they have been loyal for a decade or just arrived, so history beyond the present is ignored.
To use it, an analyst lists the possible states, such as paying on time, 30 days late, 60 days late and written off. Then they build a transition table showing the probability of moving from each state to every other state in one period, usually estimated from past records.
Multiplying today's distribution of customers or accounts by that table gives next period's distribution. Repeating the calculation shows how things evolve, and many systems settle into a steady state in which the proportions stop changing even though individual customers keep moving.
Finance teams use the method for credit risk, where it estimates how many loans may default; for market share forecasting; for customer lifetime value; and for pricing and risk models that track credit ratings over time. Operations teams use it for equipment reliability and inventory conditions.
The main limitation is the assumption that probabilities stay the same from one period to the next and that history does not matter. In reality, a recession, a new competitor or a pricing change can alter the transition probabilities, so the results are best treated as scenario guides to be updated with fresh data.
In practice
Real-world examples.
Example
A telecommunications company tracks customers across three states: on a premium plan, on a basic plan or cancelled. Using last year's movements it forecasts that after twelve months, 18% of premium customers will have cancelled and adjusts its retention budget.
Example
A bank groups its loan book into current, 30 days overdue, 90 days overdue and defaulted. Markov analysis on monthly movements helps the risk team estimate how many dollars of the $50,000,000 portfolio may be lost over a year, which sets its loss provision.
Example
A factory models a machine that is either running or under repair, with a 5% chance of breaking down each day and a 60% chance of being fixed each day. The maintenance manager uses the steady-state result to decide how many spare machines to keep.
Formula
Calculation
Next-period distribution = Current distribution x Transition probability matrix
Two brands share a market of 10,000 customers, starting with 5,000 each. Each month, 80% of Brand A customers stay and 20% switch to Brand B, while 70% of Brand B customers stay and 30% switch to Brand A.
Brand A next month = (5,000 x 0.80) + (5,000 x 0.30) = 4,000 + 1,500 = 5,500. Brand B next month = (5,000 x 0.20) + (5,000 x 0.70) = 1,000 + 3,500 = 4,500. Applying the table again gives Brand A = (5,500 x 0.80) + (4,500 x 0.30) = 4,400 + 1,350 = 5,750 and Brand B = 4,250. The steady state is reached when Brand A holds 0.30 / (0.20 + 0.30) = 60% of customers, which is 6,000 customers.Case study
Seen in the real world.
Cobalt Streaming is an illustrative, fictional subscription service that wanted to forecast subscribers across three states: free trial, paying and lapsed. Its analysts built a monthly transition table from two years of data and found that 40% of trial users became paying and 25% of paying users lapsed each quarter.
Running the table forward, the model showed that, without changes, paying subscribers would level off at about 45,000. The marketing team then tested a win-back offer, estimated to move 10% of lapsed users back to paying, and re-ran the model to see the new steady state.
The forecast predicted around 52,000 paying subscribers, enough to justify the cost of the offer. In this illustrative story the value of the method was that it turned a debate about marketing into a testable number.
Watch out
Common mistakes.
- Assuming the transition probabilities never change, when competition, pricing and the economy can shift them quickly.
- Forgetting that the method ignores history, so a long-standing customer is treated the same as a new one.
- Reading the steady state as a prediction of what will happen next month, when it describes where the system heads over many periods.
Questions
People also ask.
Where do the probabilities come from?
They are usually estimated from past records, counting how often accounts or customers moved from one state to another in earlier periods.
Does Markov analysis give exact forecasts?
No, it gives probability-based expectations, which are only as good as the data and the stability of the underlying behaviour.
What tools do analysts use?
Simple cases fit in a spreadsheet using matrix multiplication, while larger models use statistical software.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related