Demand planning, on your premises

A forecast you can stand behind.

Stratos turns the sales exports your team already produces into a reviewed, auditable twelve-month plan. The models compete, the planner decides, and every number can explain itself.

A year of demand, one item RAMADAN TODAY HISTORY FORECAST · P10–P90 JFM AMJ JAS OND Fig. 1 — a year of demand, one item.

I — The problem

Every month, one planner faces two hundred products and a spreadsheet that answers to no one. The busiest weeks of the year arrive on a calendar the spreadsheet has never heard of; the formula that sets next quarter’s stock was written by someone who left; and when a number is challenged in the room, there is nothing behind it to show.

So the plan is negotiated instead of read. Sales pads it, production discounts it, finance re-does it — and the company quietly runs on three private versions of the truth.

Two hundred products. Twelve months. One person answerable for every number.

II — How it works

One cycle, every month.

Stratos does not replace the planner’s judgement. It gives that judgement a place to stand, a record, and a way to be checked next month.

01

Ingest

The same export your team already emails. Stratos maps it to a clean monthly picture, proposes the cut-off, and waits for the planner to confirm.

02

Review

Months that broke pattern are flagged with their evidence — a collapse, an event, a spike. The planner rules on each. Nothing is corrected quietly.

03

Simulate

Seven model families re-forecast months they never saw. The winner must beat doing nothing at all, or it does not win.

04

Publish

One plan is signed and becomes the record. Every publication is revisioned; nothing is overwritten, and amendments carry a reason.

05

Grade

Next cycle, last cycle’s promise is measured against what actually happened — in kilograms and in money, at every level of the portfolio.

06

Roll up

The published plan loads the production lines, sizes the buffer, and feeds the executive view of volume, revenue and gross profit.

Inside steps two and three: the pass that runs twice.

The first scan for unusual months knows nothing about the model. The second one is the model — and what it did not see coming comes back to you.

The loop: scan, rule, train, be surprised, look again WHAT THE MODEL DID NOT SEE COMING GOES BACK FOR A SECOND LOOK — advisory, never automatic 010203 040506 Pattern scan You rule Train & compete Model surprises Grid search Publish forward Months that break the product’s own shape — found without a model. Keep, correct or mark. Nothing changes on its own. Every family forecasts the twelve months it was never shown. The shipped model’s own surprises, scored in its own noise. Challenge the factors; every alternative re-fitted and ranked. Twelve months ahead, signed, revisioned, never overwritten. THE APP FINDS YOU DECIDE THE APP COMPUTES Every flag waits for a person. The loop never closes itself.
Fig. 2 — the detection pass runs before the model, and again from inside it.

Forecast Simulation.

Before a model is allowed to forecast your next twelve months, it has to prove itself on the twelve it was never shown.

The split, the test, the scoreboard 1 · THE SPLIT THE PRODUCT’S WHOLE HISTORY TRAINED ON EVERYTHING BEFORE HELD OUT THE LAST 12 MONTHS One split, one honest test. The model is never shown the year it will be judged on. 2 · THE TEST THE TWELVE HELD-OUT MONTHS EVERY GAP IS SCORED — ITS SIZE AND ITS DIRECTION WHAT HAPPENED THE WINNER THE OTHERS Every family forecasts the same hidden months. Only then are they scored. 3 · THE SCOREBOARD MUST BEAT DOING NOTHING SARIMAX · airline SARIMAX · sma2 ETS damped Seasonal naïve Seasonal MA Moving average Naïve WINS Shorter is better. Ties go to the model you can explain.
Fig. 3 — the split, the test, the scoreboard.

A model that fits your history perfectly has proved nothing; it has only memorised. So Stratos cuts each product’s history once, twelve months back. Every candidate family is trained on everything before the cut and then asked to forecast the twelve months it has never seen — the same twelve months, for all of them, so the comparison is fair. Each month’s gap between forecast and reality is measured, and the families are scored on the total weighted error and on bias, which says whether a model leans high or low rather than merely missing. The winner is the one that actually forecast best, with a preference for models a planner can explain when the margin between them is thin. One competitor is always in the race: the seasonal naïve, which simply repeats last year. Anything that cannot beat that is not worth running.

← Back to the instruments

Outlier scanning, with a Kalman innovations filter.

Not “which month looks odd” — which month actually surprised the model that is doing the forecasting.

The surprise, measured in the series' own noise THE SERIES, AND WHAT THE MODEL EXPECTED RAMADAN RAMADAN WHAT SOLD WHAT THE MODEL EXPECTED THE SURPRISE, IN UNITS OF THIS SERIES’ OWN NOISE 0 +3.2 SURPRISE −2.8 STOCKOUT? FLAGGED — THE PLANNER RULES. NOTHING IS CORRECTED AUTOMATICALLY. inside a known event, the bar is raised
Fig. 4 — the surprise, measured in the series’ own noise.

Most tools flag a month because it is far from an average. That catches every Ramadan and misses the quiet disasters. Stratos asks a sharper question: given everything the model knew a month ago, how surprising was what actually arrived? The filter carries a running expectation for each product and a running measure of that product’s own noise, and scores each month’s surprise in units of that noise. The tolerances are deliberately lopsided — a sudden collapse is more often a stockout or a delivery that slipped than a real fall in demand, so it is questioned sooner than an equally large jump. Inside a known event the bar is raised, because a surprise during Ramadan is usually the calendar rather than the market. What breaks tolerance is flagged, with its evidence, and handed to the planner. Nothing is ever corrected on its own.

← Back to the instruments

Calendar intelligence.

The biggest weeks of your year do not sit still. Stratos knows where they will be, per product, for every year ahead.

The peak that walks backwards through the year JANFEBMAR APRMAYJUN JULAUGSEP OCTNOVDEC Y1 Y2 Y3 Y4 Y5 THE SAME EVENT, ELEVEN DAYS EARLIER EACH YEAR — across five years it moves from May into March. WHERE STRATOS PUTS THE PEAK BEFORE IT HAPPENS ≈ 7 WEEKS OF DRIFT
Fig. 5 — five years of one product, aligned to the civil calendar.

The Hijri year is about eleven days shorter than the civil one, so Ramadan and both Eids march steadily backwards through the Gregorian months. A model that only knows “March” sees a peak that arrives early, then earlier, then splits across two months — and reads all of it as noise. Stratos places the events themselves on the calendar for every year ahead, along with the weeks that lead into them, the school terms and the paydays. It then measures, per product, how much each of those actually moves demand: the run-up to Eid matters enormously for gifting lines and hardly at all for a staple. The result is a forecast that expects a February peak to become a January peak, because it knows why.

← Back to the instruments

Feature scanning, selection & grid search.

Which signals actually move this product — and what happens if the planner disagrees.

From candidate signals to a defended shortlist 1 · THE CANDIDATES Ramadan · the month Ramadan · run-up Ramadan · the month after Eid al-Fitr · the month Eid al-Fitr · run-up Eid al-Adha · the month Eid al-Adha · run-up School term · in session School term · returns Trading days in month Every product is offered the same list. None is assumed. 2 · MEASURED ON YOUR HISTORY NO EFFECT Eid · run-up Ramadan · month Ramadan · run-up School · returns PULLS DEMAND DOWN Eid al-Adha · month Trading days School · in session DISCARDED — TOO WEAK Bars show effect; whiskers show how sure. 3 · CHALLENGED, THEN RE-RANKED TOO CLOSE TO CALL — WITHIN 2 POINTS OF THE BEST FACTOR SET PENALTY vs. THE BEST Eid run-up + Ramadan month + run-up Eid run-up + Ramadan month Eid run-up + Ramadan month + school Ramadan month + run-up Eid run-up only No calendar factors at all best +1.4 +1.9 +5.6 +9.2 +24.7 IN USE ← THE COST OF NO CALENDAR Fit, minus a penalty for every factor used. Lower wins.
Fig. 6 — from candidate signals to a shortlist that can be defended.

Stratos offers every product the same list of candidate signals and lets the history decide. Each candidate is measured for how much it moves demand and how confident that measurement is; the ones that cannot earn their place are dropped, and the survivors are shown with their strength, not merely their names. Then comes the part that matters in a review meeting. If a planner believes a signal is missing — or believes one that survived is nonsense — the grid search re-fits the alternatives and ranks them side by side, including the option of using no calendar factors at all. The argument stops being a matter of seniority.

How the ranking works, in plain English

The trap. Adding another factor always makes a model fit the past a little better — even a meaningless one. So you cannot choose by fit alone, or you would always choose the biggest set, and the model would be memorising your history rather than learning from it.

The fix. Each combination is scored on how well it explains the history minus a penalty for every factor it used. A factor only improves the score if it earns more than it costs. The penalty gets steeper when history is short relative to the number of factors, which matters here — four years of monthly numbers is not much to spend. Lower is better, and the table shows each set’s distance from the best rather than a raw number, because only the gap means anything.

The honest part. A gap of less than about two points is not a real difference — those sets are statistically indistinguishable from the winner. Stratos draws that band rather than hiding it, so a tie is presented as a tie. When several sets are level, the choice belongs to the planner, and the sensible rule is to take the simplest one you can explain to the room.

← Back to the instruments

Capacity check.

A plan the factory cannot build is not a plan. Stratos says so before the month begins, not during it.

The plan, pressed onto the lines EVERY LINE, EVERY MONTH JFMA MJJA SOND Oven 1 Oven 2 Wafer Pack A Pack B THESE TWO DO NOT FIT LIGHTER → HEAVIER LOAD BEYOND WHAT THE LINE CAN MAKE Two months on Oven 2 do not fit. Everything else does. THE GAP, MEASURED WHAT THE LINE CAN MAKE FEB 420 t HEADROOM MAR 310 t OVER THE CEILING Stratos measures the gap. What to do about it is the plant’s call, not the app’s.
Fig. 7 — the published plan, weighed against the lines that must make it.

Once the plan is published, Stratos routes it onto the production lines that actually make the products and compares each month’s load against what that line can produce. The result is a single grid: every line, every month, shaded by how hard it is being pushed, with the months that do not fit marked plainly. Where a month overflows, the shortfall is quantified rather than described — this many tonnes, on this line, in this month — and the neighbouring months show their headroom, so the size and the shape of the problem are both visible. What Stratos will not do is decide for you. It does not schedule the factory, it does not move volume between months, and it never quietly trims the forecast to make the arithmetic work. It shows the collision early enough that the people who own the plant can deal with it while there is still time.

← Back to the instruments

Safety stock from measured error.

How a buffer is actually sized — in five moves, each one visible in the application.

From measured misses to a defensible buffer FORECAST − ACTUAL C · 90% A · 98% trimmed SHELF-LIFE CEILING SERVICE LEVEL 95.0 94.6 PROMISED REALIZED 1 · THE MISSES, COLLECTED 2 · GATHERED INTO A CURVE 3 · THE SERVICE LINE, PLACED 4 · CAPPED BY SHELF LIFE 5 · GRADED IN PUBLIC
Fig. 8 — from measured misses to a defensible buffer.

Stratos never sizes a buffer from a textbook constant. It collects the shipped model’s real misses over the review window, gathers them into their own distribution, and places each item’s service target on that curve — higher for the A items that carry the business, lower where a miss costs little. The buffer is then capped by the item’s shelf life, because stock that expires is not safety. And the promise is graded in public: the service level the plan committed to, printed next to the service level the shelves actually delivered.

← Back to the instruments

IV — How it is delivered

On your premises. Under your name.

Stratos installs on a machine you control. There is no tenant, no shared cluster, and no copy of your sales history anywhere but your own building.

It runs where you are

A single install on your own hardware. Your sales data is read locally and never sent anywhere for processing.

Every cycle is kept

Each month’s run is archived whole — the data, the decisions, the published plan. Old cycles can be reopened and read, never rewritten.

Configured, not coded

Your product hierarchy, calendar, measures and service policy are settings. Adapting Stratos to your business does not mean a development project.

Plain language throughout

The application explains itself to a planner, not to a statistician. Every recommendation says what it is and why, in words.

V — The demo

Step inside.

The demo is the real application, running over a fictional client — John S Foods, a Gulf biscuits and snacks manufacturer invented for the purpose. Every figure on every screen was computed by the same engine that would run in your building.

  • A full planning cycle, from raw export to a published, signed plan.
  • The Forecast Simulation with its scoreboard, item by item.
  • Last cycle’s promise graded against what actually happened.
  • Capacity load and safety stock built on the published plan.

One email at most, to follow up on your visit. No mailing list, no third-party trackers.

Rather talk to a person first? WhatsApp us · sales@elunia.cloud