Reading Variants
Every process has more paths than anyone expects. Which ones matter, and why the count itself is not a finding.
Procedure
Variant analysis is where most real findings come from, and where the most meaningless statistics are produced.
What a variant is
A distinct sequence of activities, from case start to case end.
Two cases with the same steps in the same order are the same variant, whatever their attributes.
Counts are typically large. Hundreds or thousands for a moderately complex process.
Why the count means little
It is a function of activity granularity, not of process discipline.
Refining one activity into three can multiply the count with no change in reality.
Loops explode it: a step repeated two, three or four times produces distinct variants each time.
Report the cost of variation, not the count of variants.
The distribution
A small number of variants usually cover most cases.
A long tail of variants covers one or two cases each.
Plot the cumulative coverage. The shape of that curve is the useful summary: how concentrated the process is.
A very flat curve means there is no standard process, which is itself the finding.
Which variants to look at
The largest by volume, which is the real process whether or not it matches the documentation.
The slowest by median cycle time, which is where the delay is.
The most expensive, volume multiplied by cycle time, which is the correct ranking for prioritisation and is rarely used.
Those containing rework, which indicates something failed upstream.
Those that skip a control, which is a compliance finding rather than an efficiency one.
Not the rarest, unless there is a compliance reason, since a variant covering one case is usually a data defect.
Segmenting them
Where the findings sharpen.
By customer or product type. High-value cases frequently take a longer path than intended.
By region or team, which reveals local practice divergence.
By period, which shows whether variation is increasing.
By channel, which frequently explains most of the variance in one field.
The comparison between two segments is more informative than either alone.
From variant to action
Name what makes the variant different — an extra approval, a return to an earlier step, a skipped check.
Find out why, by asking the people who handle those cases. The log never says why.
Frequently the answer is that the standard path does not work for that case type, in which case the finding is a process design gap rather than a compliance failure.
Quantify: cases affected, additional days, additional touches.
Then decide: absorb the variant into the standard process, prevent it, or leave it and stop reporting it as a deviation.
Ranking by cost, not frequency
The prioritisation most projects get wrong.
Frequency alone over-weights the standard path, which is usually fine.
Median duration alone over-weights rare pathological cases.
Volume times excess duration — the extra days a variant costs against the fastest path — is the correct ranking.
Compute it and sort. The top of that list is frequently a variant nobody had noticed, sitting at moderate volume and moderate delay.
Present the top five with their total cost in days, which is a number an operation acts on.
The cumulative coverage curve
The most useful single summary of variation, and it takes one query.
Sort variants by case volume, descending.
Plot the cumulative proportion of cases covered.
A steep curve means the process is concentrated: a handful of paths cover most work, and standardisation is largely achieved.
A flat curve means there is no standard process, which is the finding.
Compare the curve across segments, which shows whether one product line or region is the source of the dispersion.
Presenting a variant count
The headline finding of most first analyses, and the one most easily dismissed.
Report the count and the distribution together. Two hundred variants where three cover most cases is a different situation from two hundred spread evenly.
Name the top variants in plain language.
Show the long tail's share of volume and of cost, which are frequently very different.
Say which variants are legitimate, because many are correct responses to circumstances the design never anticipated, and treating deviation as error loses the audience immediately.
Take the list to the people who run the process, who will explain most of it in an hour.