What Markov analysis is
Markov analysis, in workforce planning, is a method for estimating how an organisation’s employees will be distributed across its levels in the future, built on transition rates measured from the past: how many of the people at a level stayed there, how many moved up, and how many left.
It answers a question that a count of today’s employees cannot: if things carry on as they are, where will the people holding level three posts in three years’ time come from? Its answer is a figure derived from measured behaviour, not a manager’s estimate.
The transition matrix in Markov analysis
The method rests on a single table. Each level has a row that distributes the people who were at that level at the start of the year across where they were at its end. There are four destinations: staying at the same level, moving up to the level above, moving down or across, and leaving the organisation. Each row necessarily adds up to 100%, because everyone who was at the level went to one of those four destinations and there is nowhere else to go. A row that does not reach 100 is a row from which one of the paths has been lost, not a row describing a different reality.
In the illustrative matrix below nobody moves down or across, so that destination has no column. The figures were chosen to show the calculation. Each organisation measures its rates from its records, and in the sources we reviewed we found no published transition rates for the Saudi market suitable for transfer from one organisation to another.
| From level | Stayed | Moved up | Left the organisation |
|---|---|---|---|
| Level three (leadership) | 90% | No level above | 10% |
| Level two (supervisory) | 75% | 10% | 15% |
| Level one (operational) | 72% | 8% | 20% |
The stay column is not the retention rate of a level. People promoted out of a level are still employed, so a retention rate counts them as retained, while the matrix counts them as lost to the level they left.
A one year Markov analysis, worked in full
Suppose today’s distribution is 10 people at level three, 40 at level two and 150 at level one. Applying the rows gives the following after one year:
- Level three: those who stay, 10 × 90% = 9, plus those promoted into it from level two, 40 × 10% = 4. The total is 13.
- Level two: those who stay, 40 × 75% = 30, plus those promoted into it from level one, 150 × 8% = 12. The total is 42.
- Level one: those who stay, 150 × 72% = 108, with no inflow from inside the organisation. The total is 108.
The headcount falls from 200 to 163, and the shortfall sits in the base: level one ends the year 42 people short, while the two upper levels grow by 5 between them. That is what the method shows and a turnover table does not. The organisation needs no hiring at the two upper levels, because promotion feeds them, and at the base it needs more than its leaving rate suggests, because the base loses people through departures and through promotion together.
Hence the first practical lesson of Markov analysis: a promotion is a loss to the level its holder leaves, exactly as a resignation is, even though it is a gain to the organisation. At level one in the example, departures account for 30 of the 42 people lost and promotions for the other 12, so a hiring estimate built on departures alone would arrive at 30 rather than 42.
Markov analysis over three years
The effect of the method becomes clear when the calculation is run again on its output. Running the same rows for three consecutive years, with no external hiring, gives the following, listed from level three down to level one:
- After one year: 13, 42 and 108, a total of 163.
- After two years: about 16, 40 and 78, a total of about 134.
- After three years: about 18, 36 and 56, a total of about 111.
The figures for the second and third years are rounded to whole people, and the calculation carries the fractions forward. What matters is not the falling total but two opposite movements: the leadership level grows from 10 to about 18, while the operational level shrinks from 150 to about 56. The ratio of base to top was 15 to 1 and becomes about 3 to 1. The organisation’s shape changes to that extent while it applies the same policy it has always applied.
None of this is a forecast of what will happen, because the run assumes three years without a single hire. It is a demonstration of what the structure does by itself if left alone: the top is fed by promotion and needs nobody, and the base is fed by recruitment alone. Once that shape is seen, the hiring question is no longer “how many do we need?” but “at which level do we need them?”.
The assumptions behind Markov analysis
The method rests on three assumptions, each of which can break, and knowing them is the difference between using it and using it where it does not fit:
- Stable rates. The calculation assumes that last year’s rate holds for a coming year. Anything that changes behaviour undermines that: a change to the pay scale, a reorganisation, or a new competitor entering the market. After any change of that kind the rates are measured again rather than carried over.
- A uniform level. A row treats everyone at a level in the same way, but they are not the same: an employee with two months of service and one with nine years have different probabilities. The wider the level, the weaker this assumption, so splitting a large level into two bands gives a more precise result than keeping it as a single row.
- Groups, not individuals. A rate of 8% says that twelve of the hundred and fifty will move up; it does not say who. Using it to decide the future of a named employee applies a group figure to an individual question.
A further caution concerns how the output is interpreted, not how it is calculated. The method describes what has been and extends it forward, so by its nature it preserves any flaw in past behaviour. If the rate of promotion from one particular group in an organisation was lower than from others, its matrix will reproduce that pattern in the future, because the figure transferred the reality without judging it. Taking the output as what ought to happen misreads it. Markov analysis is also purely descriptive arithmetic, and it carries no ruling on ending an employment relationship or on hiring.
Rows and columns in a Markov analysis
The table can be examined in two directions, and each answers a different question. A row asks: of the people who were at this level, where did they go? That is the question of loss. A column asks: of the people now at this level, where did they come from? That is the question of supply.
The column can be the direction that gets overlooked, and in planning it can be the more useful of the two. The column for level three in the example holds two sources only: those who stayed there and those promoted from level two. It has no entry from outside at all, because the matrix was not built with one. A table of this kind does not describe an organisation that hires into its leadership level from the market, and such an organisation needs to add recruitment as a separate input rather than leave it out of the calculation. An external input can be added for each level, and there is more than one way to formulate it; we do not favour any of them. What matters is that recruitment is a separate input, not a row in the matrix. The split of internal movement between promotions and lateral moves is described by a separate measure, the career path ratio.
A related constraint lies in how the model is built: leaving is a state from which there is no return. Someone who leaves does not come back in the calculation, although in practice former employees can come back. An organisation to which some of its leavers return counts their return as an external input, not as a path inside the matrix; otherwise the condition that each row adds up to 100 no longer holds.
What Markov analysis needs before it is run
The inputs of Markov analysis are few but strict: defined and stable levels, with each employee’s place in them known; the date each employee entered and left each level, not only the organisation; and several years of records rather than one, because a single year may contain an exceptional event that the calculation then turns into a rule. Measuring over more than one year is better than measuring over one, but the point at which the figure settles depends on the size of the organisation and on how much its activity fluctuates.
The first condition is where the difficulty can lie. An organisation that changed its levels part way through the measured period has no continuous series, and a figure calculated on a broken series describes nothing. That is why arranging jobs into levels, which job classification is one way of doing, is a precondition of the analysis and not a result of it.
Size matters for a second reason. Markov analysis is calculated on rates, and a rate calculated on a small number can move a great deal when a single person leaves: at a level of five people, one departure moves the leaving rate by 20 percentage points. The point at which an organisation becomes too small for the method to suit it is not settled by the definition.
What Markov analysis is not
- A headcount plan. The plan sets the approved number for each job, which is a decision. Markov analysis estimates the expected number if behaviour does not change, which is an estimate. The value lies in comparing the two: the gap between what is approved and what is expected is what gets planned for.
- Capacity planning. That discipline looks at the capacity needed to carry out a known body of work, whereas Markov analysis looks at what the workforce will become by itself.
- Succession planning. It works on particular roles and named people, whereas Markov analysis works on flows between levels. The two complement each other: the analysis says that level three will lose one person in the coming year, and the succession plan says who replaces them.
- Root cause analysis of resignations. That analysis asks why people leave, whereas Markov analysis takes the leaving rate as given and does not explain it. The leaving rate is an input to Markov analysis, while explaining it is the work of root cause analysis.
Before Markov analysis informs a decision
What Markov analysis offers is less the figure it produces than the question it forces: where do the people at each level come from? An organisation that finds its level two fed entirely by promotion knows that any slowdown in hiring at the base will reach that level later, not at once. One that finds its rate of promotion from level one to level two is 8% when it needs 15% knows that its problem lies in developing people, not in recruiting them.
Neither conclusion appears in any single number; both appear in the relationship between the rows. So the table matters more than the result, and carrying the result alone into a presentation keeps the part of the analysis that says least.
This is an explanation of the concept, not legal advice.
A standalone Saudi HR system
One employee file holding the contract, the documents and their expiry dates, the attendance record, leave, salary and end-of-service entitlements. End-of-service, overtime and leave-balance calculations are built into the system.
A standalone system on its own subscription. The connection to Qoyod Accounting is now available.