The Problem
An average is a single number standing in for a whole distribution, and it can look reassuring while concealing the thing that will actually hurt you. Five units a day for six days running gives a mean average of five. One, then ten, then fifteen, then two, across the same week, can give a very similar mean. The first product is predictable and can be stocked to its average. The second cannot, and stocking it to its average guarantees you will be short stocked on its best days.
The same blind spot appears wherever an average is used as a summary. A mean average dispatch time can look healthy because most orders go out same day, while a small tail of orders takes a week; invisible in the average but extremely visible to the customers who experience the long tail delays.
A mean average margin can look identical across two products.
1. Item A earns it’s average margin consistently
2. Item B swings between losing money and making a great deal of it on any given day
For a buyer, the practical consequence is that decisions made from mean averages alone are systematically wrong in one direction: they under-stock volatile products and over-trust stable-looking averages.
How It Works
What a percentile actually tells you that a mean cannot. A statistical percentile answers a different question to the calculation of a mean average. The 90th percentile of daily demand describes roughly what demand looks like on the busiest 10% of days. If that figure sits far above the mean average, the product has real spikes the average is masking, and the safety stock calculation needs to know about it. If it sits close to the mean average, demand is stable and a smaller buffer is genuinely sufficient.
Both facts are commercially useful, and neither is visible from an average.
The same method applied to margin, not just demand
Two products can show an identical average margin while behaving completely differently. On one, the cost from the supplier is stable and the margin is much the same on every sale. On the other, supplier costs move around and the margin swings between very little and a great deal. Percentiles on margin surface that difference, which matters when choosing between two lines that look equivalent on a summary report.
If the mean average margin is the same on both items, the more consistent item (calculated daily mean margin) is probably the less stressful item to hold because your daily margin outcome is more certain.
Applied to available inventory as well
Percentiles are computed on available stock across each rolling 30, 60 and 90 day window.
This is what makes it possible to distinguish a product that has been comfortably in stock all quarter from one that has spent much of the quarter hovering close to zero, even where both currently show stock on hand. The first is well managed; the second has been quietly at risk.
Why this is a design decision, not a reporting feature
Percentiles are calculated each night and stored for all SKUs, not calculated on request. They exist as pre-computed values in the nightly snapshot alongside the mean averages and standard deviations, which is what allows the AI to reason about distribution shape in a conversation without performing statistical work in the moment. It is the same principle that underpins everything else in the data lake: calculate in advance, interrogate instantly.
If speed and accuracy are important to you, then understanding how an AI model delivers calculated data to you on demand is a critical question to ask during the software buying cycle.
Trust your stock. Transfer with confidence. Control orders. Fulfill every day like clockwork.
G2 Customer Reviews
What do you like best about Stok.ly – Inventory-Centric Cloud ERP?
“we have been a user for the last 4 years and we cannot think of life without it for our retail business. it makes listing to shopify and other marketplaces easy, keeps inventory accurate online and the POS is easy to use. accurate inventory across all sales channels is the big win for us.”
Order management and inventory ERP software for growing retail, wholesale and distribution businesses.
The Numbers, and How They Are Calculated
| Metric | Windows held | Statistics held | What the distribution tells you |
|---|---|---|---|
| Daily demand | 30, 60, 90 days | Total, mean, standard deviation, P30, P60, P90 | Whether demand is stable or spiky, and how large a buffer is genuinely required |
| Gross margin | 30, 60, 90 days | Total, mean, P30, P60, P90 | Whether margin is earned consistently or swings with supplier cost |
| Revenue | 30, 60, 90 days | Total, mean, percentiles | Whether revenue concentration sits in a few large days or spreads evenly |
| Available inventory | 30, 60, 90 days | Mean, P30, P60, P90 | Whether the product has been comfortably stocked or hovering near zero |
| Inventory value | 30, 60, 90 days | Mean | Feeds the GMROII calculation as the average capital employed |
What Actually Changes
| Decision | With the mean only | With mean and percentiles |
|---|---|---|
| Setting a stock buffer | One buffer rule applied to everything | Buffer scaled to that product’s actual variability |
| Choosing between two products with equal average margin | Coin toss, or gut feel | The consistently profitable one is identifiable |
| Assessing service level | Looks fine because most orders are fine | The tail of bad days is visible and quantified |
| Judging whether a product has been in stock | Current stock on hand only | Whether it has been near zero for much of the quarter |
| Explaining a purchasing decision | Because it averages this much | Because on its busiest tenth of days it needs this much |
Questions the AI Answers From Pre-Calculated Data
-
- What is the 90th percentile (top 10% of extreme behaviour) of daily demand for this SKU, and how does it compare to the mean average?
- Which products have a 90th percentile demand more than double their mean?
- Show me two products with similar average margin but very different margin consistency.
- Which SKUs have spent most of the last 90 days close to zero available stock?
- What is the demand standard deviation for these products, and which is the most erratic?
- Which products are stable enough that a smaller safety stock would be safe?
How This Compares to Cin7, Brightpearl, Orderwise and Linnworks
Positions below reflect vendor-documented capability as at 27 July 2026.
-
- Cin7. Standard reporting is built on totals and averages. ABC segmentation classifies products by value contribution, which is a different analysis to distribution shape. No percentile-based demand, margin or inventory analysis is documented as exposed to the buyer.
- Brightpearl. Reporting and the Inventory Planner connection work from sales history and averages. No published percentile analysis across demand or margin.
- Orderwise. Historical sales reporting and seasonal trend analysis. No documented percentile or distribution analysis.
- Linnworks. Channel and order reporting. No statistical distribution analysis of this kind.
How to Test This in a Demo, Including Ours
Use the same anonymised data set and the same questions with every vendor, and insist on a live workflow rather than slides.
- Ask for the 90th percentile of daily demand for a named SKU alongside its mean, and ask what the difference implies for safety stock.
- Ask for two products with a similar average margin and different margin variability, and ask which the system would recommend buying and why.
- Ask how the percentile figures are calculated and what the error margin is.
- Ask which products have spent the largest proportion of the last quarter near zero available stock.
- Ask whether the percentiles are stored or calculated at the moment of asking, and check the response time on a large catalogue.
- Ask the same question twice and confirm the figures are identical, which tells you whether they are pre-computed or re-derived.
Score on time taken, number of clicks, whether the answer is traceable to source records, and whether the insight becomes an editable draft without rekeying.
Frequently Asked Questions
Does Stok.ly use percentile analysis rather than just averages?
Both. The nightly data lake computes the mean, the standard deviation and the 30th, 60th and 90th percentiles for demand, margin, revenue and available inventory across rolling 30, 60 and 90 day windows, so the AI can reason about the shape of a distribution rather than only its centre.
Why does a percentile matter more than an average for a buying decision?
Because an average can describe a stable product and a volatile one with the same number. The 90th percentile tells you what the busiest tenth of days actually looks like, which is the figure that determines whether you are in stock when it matters. Stocking a spiky product to its mean guarantees being short on its best days.
How accurate are Stok.ly’s percentile calculations?
They are computed using a t-digest algorithm with a documented error margin below 0.1%, which is indistinguishable from exact percentiles for planning purposes. Exact recalculation across full history for every SKU at every location every night would be computationally impractical at this scale.
Are percentiles applied to anything other than demand?
Yes. The same methodology is applied to gross margin, revenue and available inventory. Margin percentiles are particularly useful when choosing between two products with a similar average margin but very different consistency.
What does the standard deviation get used for?
It is the direct input into the statistical safety stock calculation. A high standard deviation means erratic demand and earns a larger buffer automatically; a low one earns a smaller buffer and releases capital.
Can percentile analysis tell me if a product has been quietly running out?
Yes. Percentiles on available inventory across the trailing windows show whether a product has been comfortably in stock or hovering near zero for much of the period, which current stock on hand alone cannot reveal.
Is this available outside the AI assistant?
Yes, via API including for Power BI, reading the same pre-computed fields, so dashboard figures and conversational answers agree.
Contact Information
All our sales, support and development team are located in Hereford and Cheltenham in the U.K. Please submit the contact form and we will contact you within the same business day.