The Stok.ly Ai Data Architecture: Calculate Ever Night. Retrieve Values On Demand.
Every night a batch process reads every transaction, purchase, stock movement and cost change for every SKU at every location made that day and computes the full statistical picture — means, standard deviations, percentiles, margins, forecasts, risk scores — once, in a controlled process. The result is a stored value with a calculation date attached – there is an audit trail back to every value retained in our database.
When you ask the Stok.ly AI Pro a question, it is not deriving an answer through reasoning.
It is looking up a number that already exists in the database.
A number with an audit trail that you can rely on and trust.
The Breadth: 91 Fields, Eight Categories, Every SKU, Every Night
The scale of what is pre-calculated is the part most competitors’ “AI-powered” claims can’t compete with. This is not two or three headline metrics. It is a complete statistical picture of every product’s operational and commercial reality, rebuilt from scratch every single night.
Identifiers — 6 fields
| What is calculated | What it tells you |
|---|---|
| A unique reference for every SKU, location, account and date | Every one of the 91 fields below can be traced back to the exact SKU, the exact location, and the exact day it describes; nothing is aggregated in a way that loses that traceability. |
| Days of tracked history, capped at 365 | Qualifies how much evidence sits behind every forecast and trend figure for that specific SKU-location. A new product is treated differently to an established one. |
Inventory Position — 6 fields
| What is calculated | What it tells you |
|---|---|
| On hand, on order, available, pre-allocated, reserved and allocated | Six distinct states of stock, not one blended “in stock” number; the difference between what physically exists, what is already spoken for, and what is genuinely free to sell. |
Valuation — 5 fields
| What is calculated | What it tells you |
|---|---|
| Weighted Average Cost, recalculated as costs change | A stable, non-spiky cost basis for every unit held. |
| Mean inventory value across 30, 60 and 90 day windows | How much capital has genuinely been tied up in this product at a location over time, not just its value today. |
Demand — 27 fields
| What is calculated | What it tells you |
|---|---|
| Daily demand, and days since demand was last recorded | The most basic signal, held precisely enough to identify a genuinely dead line rather than one that simply sells in bursts. |
| Rolling totals, means and standard deviations across 30, 60 and 90 days | Not just how much sold, but how consistent or erratic that selling pattern actually is. |
| Zero-demand day counts and non-zero day counts, per window | Distinguishes a product with no demand at all from one with occasional but genuine demand; a distinction a single average cannot make. |
| 30th, 60th and 90th percentile of daily demand, per window | What a bad day and a busy day actually look like for this specific product, not just its average day. |
Stockout History — 12 fields
| What is calculated | What it tells you |
|---|---|
| Days per window where available stock was zero | Separates a product that did not sell because nobody wanted it from a product that did not sell because there was nothing left to sell. |
| Percentiles of available inventory itself, across 30, 60 and 90 days | Whether a product has been comfortably in stock all quarter or has been hovering near zero for much of it, even if it currently shows stock on hand. |
Profitability — 24 fields
| What is calculated | What it tells you |
|---|---|
| Revenue and cost of goods sold, recognised together at fulfilment | One consistent margin figure, because both revenue and cost of goods sold are recognised at the same moment; not two numbers from two systems reconciled by hand. |
| Gross margin totals across 30, 60 and 90 day windows | Whether a product’s profitability is improving or deteriorating over time, not just what it was today. |
| Percentiles of daily gross margin and daily revenue, per window | Whether a product earns its margin consistently or swings between losing money and earning a great deal; invisible from an average alone. |
GMROII — 3 fields
| What is calculated | What it tells you |
|---|---|
| Gross margin return on inventory investment, across 30, 60 and 90 day windows | Whether the capital tied up in this specific product at this specific location is actually working hard enough to justify holding it. |
Forecasting & Replenishment — 8 fields
| What is calculated | What it tells you |
|---|---|
| Lead time, demand trend ratio, and two forecast models | A moving-average baseline and a trend-adjusted version, so an accelerating or decelerating product is caught early rather than after the fact. |
| Statistical safety stock at a 95% service level | A buffer sized to actual demand variability and lead time, not a number a person set once and forgot. |
| Reorder point, days of cover and a 0 to 1 stockout risk score | The three decision-ready numbers a buyer actually needs: when to order, how long stock will last and how urgent the situation is. |
Every field above is recalculated every night, for every SKU, at every location, in every account.
Nothing on this list is a report you have to remember to run.
Every number is auditable and retained for 365 days.
Trust your stock. Transfer with confidence. Control orders. Fulfill every day like clockwork.
Order management and inventory ERP software for growing retail, wholesale and distribution businesses.
Why Percentiles Sit Alongside Every Average
A recurring design choice across all 91 fields is worth calling out on its own: wherever a mean is calculated, statistical percentiles are calculated alongside it — the 30th, 60th and 90th percentiles, across each of the three rolling windows, for demand, margin, revenue and available inventory.
A mean can describe a stable product and a highly erratic one with an identical number.
The 90th percentile reveals what the busiest 10% of days actually looks like, which is the figure that determines whether stock, cash or margin held up when it mattered most.
Calculating this precisely, every night, across this many SKUs and locations delivers a documented error margin below 0.1%.
These are numbers you can trust when you need to make critical commercial decisions.
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.”
What “Audit Trail” Actually Means Here
This is not a marketing word. It means, specifically, that every one of the 91 fields traces back to three things: the formula that produced it, the rolling window it was calculated across (30, 60 or 90 days), and the date it was calculated — sitting in your own account’s database, not in a third-party model’s reasoning that was never recorded anywhere.
If a number looks wrong, it can be checked.
If a buyer asks why a safety stock figure is what it is, the answer is the formula and the inputs, not “the AI thinks so.”
| A generic AI tool calculating live | Stok.ly’s pre-calculated data lake | |
|---|---|---|
| When the number is produced | Inside the conversation, at the moment you ask | Overnight, in a nightly batch process, before you ask |
| Ask the same question twice | May return a different answer | Returns the identical stored value both times |
| Can the calculation be audited | Rarely — the reasoning happened inside a model and was not recorded | Yes — formula, data window and calculation date are all inspectable |
| Breadth of what is covered | Typically a small number of headline metrics | 91 fields across 8 categories, per SKU per location |
| History retained | Rarely specified | 365 days per SKU-location pair |
| Confidence on a new product | Often answered with the same certainty regardless of history | Explicitly qualified by tracked history depth |
How to Test This Yourself, With Any Vendor
This is not a claim to take on trust either. Ask any AI-in-ERP vendor — Stok.ly included — to prove it, live:
- Ask any AI-in-ERP vendor to show you the exact formula behind a number they just gave you, not just the number itself.
- Ask the same question twice, a few minutes apart, and check whether the figure is identical both times.
- Ask when the underlying data was last calculated, and whether that timestamp is available at all.
- Ask whether the answer came from a live calculation inside the conversation or a value that already existed before you asked.
- Ask for the percentile behind an average — the 90th percentile of demand or margin — and see whether the concept exists in their system at all.
- Ask what happens on a new product with five days of history: does the system state reduced confidence, or answer with the same certainty as an established line?
Score the answers on whether a specific formula was named, whether the figure was identical on repeat, and whether a calculation date existed at all.
Frequently Asked Questions
What is Stok.ly’s pre-calculated data lake?
Every night, Stok.ly’s data lake processes every SKU at every location for every account and pre-computes 91 separate statistical fields — inventory position, valuation, demand, stockout history, profitability, GMROII and forecasting signals — storing each one as a value with a calculation date attached. None of these figures are calculated live when you ask a question; the AI retrieves a number that already exists.
How many data points does Stok.ly track per SKU?
91 fields per SKU per location per day, retained for 365 days. That includes rolling means, standard deviations and 30th, 60th and 90th percentiles across 30, 60 and 90 day windows for demand, margin, revenue and available inventory, alongside forecasting and replenishment signals.
Are these numbers calculated by the AI when I ask a question?
No. The AI retrieves a pre-calculated value and explains it. The calculation itself happens overnight, once, in a batch process, using a fixed and documented formula. Asking the same question twice returns the identical figure, because both times you are reading the same stored value.
Can I verify how a specific number was calculated?
Yes. Every field traces to a specific formula, a specific rolling window (30, 60 or 90 days) and a specific calculation date, sitting in your own account’s database. This is what an audit trail means in practice: not a claim you have to take on trust, but a value you can trace back to its inputs.
Why does Stok.ly use percentiles as well as averages?
An average can describe a stable product and a highly erratic one with the same number. The 90th percentile of daily demand, margin or revenue reveals what the busiest or most extreme days actually look like, which a mean alone conceals. Both are calculated and stored, across three window lengths, for demand, margin, revenue and available inventory.
How accurate are the percentile calculations at this scale?
Percentiles are calculated using a t-digest algorithm at accuracy level 1000, with a documented error margin below 0.1% — indistinguishable from an exact percentile for any real planning purpose, at a scale where exact recalculation every night for every SKU at every location would be computationally impractical.
What does history_days actually protect me from?
A silently confident answer on a product with almost no sales history. Every SKU-location pair carries a count of how many days it has actually been tracked, capped at 365. A new product with five days of history still gets a forecast, but the AI states the limitation rather than presenting the same confidence as a product with a full year of evidence behind it.
Is Weighted Average Cost recalculated live or is it also pre-calculated?
WAC is recalculated as inventory and costs change, which keeps the valuation always current, and the 30, 60 and 90 day mean inventory values built from it are part of the same nightly pre-calculation as everything else, feeding directly into GMROII.
Does this data only cover financial figures?
No. Financial fields (valuation, profitability, GMROII) are 32 of the 91 fields. The remaining 59 cover inventory position, demand statistics, stockout history and forecasting and replenishment signals — the full operational and commercial picture, not only the numbers accountants care about.
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.