ADI and CV2 Calculation

Created by Shyam Sayana, Modified on Sat, 14 Nov at 7:38 AM by Shyam Sayana

TABLE OF CONTENTS


What are ADI and CV²?

Demand profiling places every item into one of four profiles: Smooth, Intermittent, Erratic or Lumpy. The profile describes how the item's demand behaves over time, and it helps decide how the item should be forecast and planned. A Smooth item is the easiest to forecast. A Lumpy item is the hardest.

The profile is decided by two measures:

  • ADI (Average Demand Interval) shows how often demand occurs.
  • CV² (Squared Coefficient of Variation) shows how much the quantity varies when demand occurs.

Each measure is compared with a cut-off (1.32 for ADI and 0.49 for CV² by default), and the combination gives the profile. This follows the standard Syntetos-Boylan-Croston (SBC) classification.

This article is for advanced users who want to see exactly how the values in the Demand Profile table are produced. It covers the rules, the formulas, and worked examples for all four profiles.


Terminology

TermMeaning
PeriodOne time bucket of the data: a month, or a week for weekly data. The date range chosen in the segmentation is split into periods.
Demand periodA period in which the item had demand greater than zero.
n (Periods used)The number of periods analysed for an item, from its first demand to the end of its analysis period.
kThe number of demand periods inside the analysis period.
Mean (average)The sum of the values divided by how many values there are. Written μ.
DeviationA value minus the mean. A value of 120 against a mean of 100 has a deviation of +20.
Squared deviationA deviation multiplied by itself. +20 and −20 both give 400, so values above and below the mean count equally and bigger deviations count more.
VarianceThe mean of the squared deviations. It shows how widely the values spread around the mean.
Standard deviation (σ)The square root of the variance, in the same units as the demand. The system uses the population standard deviation, which divides by the number of values rather than by one less.
CV (Coefficient of Variation)Standard deviation ÷ mean. It shows variability relative to the size of demand, so small and large items can be compared.
CV²CV multiplied by itself. This is the value compared with the CV² cut-off.
ADIThe average number of periods between demand occurrences. A higher ADI means demand happens less often.
Probability of demandk ÷ n. The share of periods that had demand.
ThresholdThe cut-off that separates one profile from another. Defaults: ADI 1.32 and CV² 0.49.

Which periods are analysed

Both measures are worked out per item, over the item's own history. They are not worked out over the whole range you selected.

  1. Start = the first period with demand. This is the first period where demand is greater than zero. Any months before it are ignored. In those months the item had no demand data, because it had not been launched or created yet. That is not the same as "zero demand", so those months must not count against the item. Counting them would make a young item look far more intermittent than it is.
  2. End = the end of the selected range. If the item is discontinued, the end is the period before the discontinue period, when that is earlier than the range end. Demand after discontinuation is excluded. The discontinue period is checked across all four dimensions.
  3. n = End − Start + 1. This is the number of periods used.
  4. Zeros after the start are real. Once an item has started selling, a period with no demand is a genuine gap. It counts in n but not in k. These gaps are exactly what ADI measures.
  5. Items with no demand at all in the range have no start period, so they cannot be analysed. Their analysis start, periods used, ADI, CV², XYZ values and both classes are left blank.

Example: you select 24 months of history, and an item is only 12 months old. It is judged over its own 12 months, not over 24.


How ADI is calculated

ADI = n ÷ k

n = number of periods from the first demand to the end of the analysis period
k = number of those periods that had demand (greater than 0)
  • Demand in every period gives ADI = 1.
  • Demand in 1 period out of 4 gives ADI = 4.
  • The higher the ADI, the less regular the demand.
  • Cut-off 1.32: below it, demand is regular. At or above it, demand is intermittent.

Why the start rule matters. An item has 24 months selected, but its first demand is in month 13, and it has demand in 10 of its 12 months.

  • Correct ADI = 12 ÷ 10 = 1.2 (regular).
  • Counting the 12 months before launch would give 24 ÷ 10 = 2.4, which wrongly marks the item as intermittent.

How CV² is calculated

CV² measures how much the size of demand varies. Only the k demand periods are used. Zero-demand periods are left out, because CV² describes order sizes, while ADI already describes how often demand occurs.

Step 1  Mean              μ  = (sum of the k demand quantities) ÷ k
Step 2  Deviations        d  = each quantity − μ
Step 3  Squared           d² = d × d
Step 4  Variance          σ² = (sum of all d²) ÷ k
Step 5  Standard dev.     σ  = √σ²
Step 6  CV, then CV²      CV = σ ÷ μ        CV² = CV × CV = σ² ÷ μ²
  • Cut-off 0.49: below it, demand quantities are stable. At or above it, they are highly variable.
  • If an item has only one demand period, there is nothing to vary, so σ = 0 and CV² = 0.

Demand profile classification

ProfileADICV²What it looks like
Smoothbelow 1.32below 0.49Demand in almost every period, with similar quantities. Easiest to forecast.
Intermittent1.32 or morebelow 0.49Demand is infrequent, but the quantity is fairly constant when it occurs.
Erraticbelow 1.320.49 or moreDemand in almost every period, but quantities swing widely.
Lumpy1.32 or more0.49 or moreDemand is infrequent and quantities vary widely. Hardest to forecast.

The 1.32 and 0.49 values are defaults. They can be changed on the Segmentation configuration screen.


Why CV² and XYZ CV are different

The Nine Box (XYZ) analysis also uses a coefficient of variation. It is not the same number as CV²:

CV² (Demand profile)CV (XYZ / Nine Box)
Periods usedOnly the k demand periodsAll n periods, zeros included
Meansum ÷ ksum ÷ n
Standard deviationPopulation σ of the k valuesPopulation σ of the n values
Answers the question"How much does the order size vary?""How variable is demand overall, including the gaps?"

So an item can have steady order sizes (low CV²) and still fall into a weaker XYZ class, because its gaps make overall demand variable. Example 1 below shows this.


Impact on XYZ analysis

The XYZ calculation follows the same start rule as ADI and CV².

  • XYZ starts at each item's first period with demand and runs to the end of its analysis period.
  • The average and the variability (CV) are worked out only over those periods.
  • Months before the item's first demand are not counted.
  • Items that existed for the whole range are calculated exactly as before.
  • Result: a newer item is no longer pushed into a worse XYZ class because of months when it did not exist.
XYZ mean    = sum of all n periods ÷ n
XYZ std dev = population standard deviation of the n values
CV          = XYZ std dev ÷ XYZ mean

X if CV ≤ X threshold,  Y if CV ≤ Y threshold,  otherwise Z

Worked examples

All examples use 12 months. Each one shows the full working. The examples cover each of the four profiles. The XYZ cut-offs are X ≤ 0.5 and Y ≤ 1.0.

Example 1: Intermittent (late start)

Monthly demand (Jan–Dec)

MonthJanFebMarAprMayJunJulAugSepOctNovDec
Demand000056075065
Counted?IgnoredIgnoredIgnoredIgnoredUsedUsedUsedUsedUsedUsedUsedUsed

Analysis period. The first demand is in May, so Jan–Apr are ignored. The end is Dec. n = 8.
Values used: 5, 6, 0, 7, 5, 0, 6, 5. k = 6 months have demand. Total = 34.

ADI = 8 ÷ 6 = 1.33. Probability = 6 ÷ 8 = 0.75.
(If the four pre-launch months had been counted: 12 ÷ 6 = 2.0, which overstates intermittency.)

CV² (demand months only: 5, 6, 7, 5, 6, 5)

MonthMayJunAugSepNovDecTotal
Demand56756534
Deviation from 5.67−0.67+0.33+1.33−0.67+0.33−0.67
Squared deviation0.440.111.780.440.110.443.33
  • Mean = 34 ÷ 6 = 5.67
  • Variance = 3.33 ÷ 6 = 0.556
  • Standard deviation = √0.556 = 0.745
  • CV = 0.745 ÷ 5.67 = 0.131, so CV² = 0.017

XYZ (all 8 values, zeros included)

  • Mean = 34 ÷ 8 = 4.25
  • Sum of squared deviations = 51.5, so variance = 51.5 ÷ 8 = 6.44
  • Standard deviation = √6.44 = 2.537
  • CV = 2.537 ÷ 4.25 = 0.597, so XYZ = Y (0.5 < 0.597 ≤ 1.0)

Result: ADI 1.33 ≥ 1.32 and CV² 0.017 < 0.49, so the profile is Intermittent.
Order sizes are steady (CV² 0.017), but two empty months push the overall CV to 0.597, which gives class Y.

Example 2: Smooth (full history)

Monthly demand (P01)

MonthJanFebMarAprMayJunJulAugSepOctNovDec
Demand10011090100105951001109010010595

Analysis period. The first demand is in January and demand continues to December, so n = 12 and k = 12.

ADI = 12 ÷ 12 = 1.00

CV² (all 12 months have demand)

MonthJanFebMarAprMayJunJulAugSepOctNovDecTotal
Demand100110901001059510011090100105951,200
Deviation0+10−100+5−50+10−100+5−5
Squared deviation010010002525010010002525500
  • Mean = 1,200 ÷ 12 = 100
  • Variance = 500 ÷ 12 = 41.67
  • Standard deviation = √41.67 = 6.45
  • CV = 6.45 ÷ 100 = 0.0645, so CV² = 0.004

XYZ. Every period has demand, so the XYZ figures equal the CV² figures. CV = 0.0645, so XYZ = X.

Result: ADI 1.00 < 1.32 and CV² 0.004 < 0.49, so the profile is Smooth.

Example 3: Erratic (regular demand, wild quantities)

Monthly demand (P05)

MonthJanFebMarAprMayJunJulAugSepOctNovDec
Demand2015035020010902518015060

Analysis period. The first demand is in January, so n = 12. Two months (Apr and Nov) have no demand, so k = 10. Total = 785.

ADI = 12 ÷ 10 = 1.20

CV² (the 10 demand months only; Apr and Nov are left out)

MonthJanFebMarMayJunJulAugSepOctDecTotal
Demand20150352001090251801560785
Deviation−58.5+71.5−43.5+121.5−68.5+11.5−53.5+101.5−63.5−18.5
Squared deviation3,422.255,112.251,892.2514,762.254,692.25132.252,862.2510,302.254,032.25342.2547,552.5
  • Mean = 785 ÷ 10 = 78.5
  • Variance = 47,552.5 ÷ 10 = 4,755.25
  • Standard deviation = √4,755.25 = 68.96
  • CV = 68.96 ÷ 78.5 = 0.878, so CV² = 0.772

XYZ (all 12 values)

  • Mean = 785 ÷ 12 = 65.42, standard deviation = 69.42
  • CV = 1.061, so XYZ = Z (above 1.0)

Result: ADI 1.20 < 1.32 and CV² 0.772 ≥ 0.49, so the profile is Erratic.
Demand almost always occurs, but the quantity jumps between 10 and 200, so overrides and promotional adjustments are often needed.

Example 4: Lumpy (infrequent and variable)

Monthly demand (Jan–Dec)

MonthJanFebMarAprMayJunJulAugSepOctNovDec
Demand05000200800100
Counted?IgnoredUsedUsedUsedUsedUsedUsedUsedUsedUsedUsedUsed

Analysis period. The first demand is in February, so January is ignored. The end is Dec, so n = 11.
Demand months: 50, 2, 80, 1, so k = 4. Total = 133.

ADI = 11 ÷ 4 = 2.75. Probability = 4 ÷ 11 = 0.364.

CV² (50, 2, 80, 1)

MonthFebMayAugOctTotal
Demand502801133
Deviation from 33.25+16.75−31.25+46.75−32.25
Squared deviation280.56976.562,185.561,040.064,482.75
  • Mean = 133 ÷ 4 = 33.25
  • Variance = 4,482.75 ÷ 4 = 1,120.69
  • Standard deviation = √1,120.69 = 33.48
  • CV = 33.48 ÷ 33.25 = 1.007, so CV² = 1.014

XYZ (all 11 values)

  • Mean = 133 ÷ 11 = 12.09, sum of squared deviations = 7,296.91
  • Standard deviation = √(7,296.91 ÷ 11) = 25.76
  • CV = 2.130, so XYZ = Z

Result: ADI 2.75 ≥ 1.32 and CV² 1.014 ≥ 0.49, so the profile is Lumpy. Demand is rare and the sizes are uneven. This is the hardest profile to forecast.

Example 5: Discontinued item

Monthly demand (Jan–Dec). The product is discontinued in September.

MonthJanFebMarAprMayJunJulAugSepOctNovDec
Demand567567569999
Counted?UsedUsedUsedUsedUsedUsedUsedUsedExcludedExcludedExcludedExcluded

Analysis period. The start is Jan. The end is Aug, the period before the discontinue period, so n = 8. The Sep–Dec values are excluded.
Values used: 5, 6, 7, 5, 6, 7, 5, 6. k = 8. Total = 47.

ADI = 8 ÷ 8 = 1.0

CV² (all 8 months have demand)

MonthJanFebMarAprMayJunJulAugTotal
Demand5675675647
Deviation−0.875+0.125+1.125−0.875+0.125+1.125−0.875+0.125
Squared deviation0.7660.0161.2660.7660.0161.2660.7660.0164.875
  • Mean = 47 ÷ 8 = 5.875
  • Variance = 4.875 ÷ 8 = 0.609
  • Standard deviation = √0.609 = 0.781
  • CV = 0.781 ÷ 5.875 = 0.133, so CV² = 0.018

XYZ. Every period has demand, so the figures match. CV = 0.133, so XYZ = X.

Result: ADI 1.0 < 1.32 and CV² 0.018 < 0.49, so the profile is Smooth, class X.

Example 6: No demand in the range

All periods are 0, so there is no first-demand period. The item has no analysis start. Analysis start, Periods used, ADI, CV², XYZ mean, XYZ standard deviation, CV, and both classes are left blank.

Quick comparison

ItemM1M2M3M4M5M6M7M8M9M10M11M12nkADICV²Profile
P011001109010010595100110901001059512121.000.004Smooth
P0310000120008000100001243.000.020Intermittent
P05201503502001090251801506012101.200.772Erratic
P0700100020000050001033.330.891Lumpy

P07 has no demand until March, so it is measured over 10 months (n = 10).


Where do I see these values?

The Demand Profile details table and the Nine Box details table both show two columns that explain which periods were used for each item:

  • Analysis start: the first period with demand for that item.
  • Periods used: n, the number of periods analysed.

The Average volume and Standard deviation columns reflect the item's own periods.

Demand Profile details table

[[IMG_PD]]

Nine Box details table

[[IMG_NB]]


What should I watch out for?

  • Young items. An item is measured only from its first demand. With very few periods, ADI and CV² can swing a lot, so treat the profile as a first indication.
  • A profile that differs from XYZ. CV² ignores zeros and XYZ CV counts them. An item can be "Intermittent" with steady sizes yet land in Y or Z. Both results are correct, because they answer different questions.
  • Zero-demand periods after launch count as gaps. If an item has genuinely stopped selling, the gaps raise its ADI.
  • Items with no demand in the range show blank values and are not classified.
  • Changed thresholds. If an admin changes the ADI or CV² cut-off, the same data can give a different profile. Check the configuration when you compare runs.
  • Stale results. Re-run the segmentation after major changes in the data so the values stay current.

  • Check the Analysis start and Periods used columns before trusting a profile, especially for items that are newer than the selected range.
  • Review Erratic and Lumpy items each planning cycle, because they carry the highest forecast risk.
  • Use the worked examples above to reproduce a value by hand when you need to explain or challenge a result.
  • Re-run the segmentation after the data or the thresholds change.

Was this article helpful?

That’s Great!

Thank you for your feedback

Sorry! We couldn't be helpful

Thank you for your feedback

Let us know how can we improve this article!

Select at least one of the reasons
CAPTCHA verification is required.

Feedback sent

We appreciate your effort and will try to fix the article