Syndicai
Contact
SolutionsSolution 02

Stock Ordering

The factory needs 3–4 months. Your customer needs it Tuesday. Stock ordering is the bet between those two dates — and today it rides on one person’s memory.

We join your sales history and your warehouse into one monthly recommendation: which configurations to order for stock, how many, with the reasoning attached. Not what sold yesterday — what should be arriving a quarter from now.

Book a discovery call

First recommendation from your real exports in 14 days

  • 3–4 months

    Typical factory lead time on imported machines. Every stock order is a forecast of a market that doesn’t exist yet.

  • ~1 in 7

    Units on a typical dealer’s warehouse list that are not actually available — sold, rented out, or reserved in a field nobody reads. Recommendations built on that list are fiction.

  • 2 systems

    Sales live in the ERP; stock lives in the warehouse tool. The ordering decision needs both — today a person joins them by hand, or not at all.

The core problem

You are not ordering for today’s market.

A machine ordered today arrives in a different quarter — often a different demand climate. The question is never “what is selling” but “what will be missing.”

  1. TodayOrder placedConfiguration and quantity decided on current knowledge
  2. Month 1–3Production & sea freightCapital committed. The market keeps moving without you
  3. Month 3–4ArrivalUnits land in a quarter you had to predict
  4. VerdictSold — or parkedThe right guess sells in weeks. The wrong one ties up cash for a year

Both failure modes are expensive: the missing bestseller sends your customer to a competitor with stock; the misjudged configuration stands in the yard, financed, losing value.

The deliverable

A recommendation, with its homework shown.

Once a month, one table: configuration by configuration, what sold, what is on hand, what is reserved, what is already on the water — and what to order net of all of it.

Configurations, not just models

“Twenty 3-tonners” is not a decision. Drive type, mast, cab — demand differs per combination, and so does the recommendation.

Confidence, stated honestly

A model with deep history gets a firm number. A model with three data points gets a hedge and a flag — the system says “thin evidence” instead of inventing certainty.

The double-order guard

Units in production and on the water are subtracted before the number is proposed. The classic failure — reordering what is already coming — becomes arithmetically impossible.

Order recommendation · monthly cycleNet of stock · reservations · in transit
ConfigurationSold 3MFreeTransitOrderReasoning
3.5 t · electric · triplex · cab142010Steady seller, cover below one month — priority order.
2.0 t · electric · duplex9560Six already on the water — ordering now would double up.
3.0 t · LPG · duplex6104Demand shifting to electric — order below trend, flagged.
1.5 t · reach truck2001Thin history — low-confidence suggestion, buyer’s call.

Illustrative rows. Every number links back to the underlying records, so “why 10?” has a one-click answer.

Applied AI

The hard part isn’t the forecast. It’s the data underneath it.

Dealer exports were written by busy people for other busy people. AI does the reading a person would need days for — code does the arithmetic that must never be creative.

  1. 01

    Phantom stock detection

    Cleaning

    Warehouse tools without a status field push people to type SOLD, RENTAL, or RESERVED into whatever column accepts text — often the engine/battery field. AI reads those fields and removes the phantoms, because a recommendation built on units you cannot sell is worse than no recommendation.

  2. 02

    One model, one name

    Normalization

    The same machine appears as a bare model code, a code with a year suffix, and a code with a serial number glued on. AI maps every variant to one base model so history aggregates instead of scattering across five spellings of the same truck.

  3. 03

    Stock sale or customer order?

    Classification

    The single most important flag for demand planning usually doesn’t exist in the ERP. AI infers it from delivery terms, free-text comments, and timing — “ASAP” smells like stock; a date four months out smells like a factory order — and marks the uncertain cases for human review.

  4. 04

    Fields that are secretly comments

    Parsing

    Date fields holding “end of Feb / early March”, “urgent!”, or “already at customer’s site”. AI turns each entry into a usable date, a range, or a labeled comment — a quarter of a typical export parses no other way.

  5. 05

    Gap analysis

    Analysis

    The configurations customers keep buying that you keep not having — and the inverse: what has stood in the yard for two quarters and why. Surfaced monthly, before the pattern costs a season.

Decision discipline

The system recommends. Your buyer decides.

Ordering discipline is a management asset. The tool’s job is to make the decision auditable — not to take it away.

  • Reasoning attached

    Every proposed quantity carries its evidence. Six months later you can still see why April’s order looked the way it did.

  • Overrides are recorded

    The buyer changes a number, the change and its reason are kept. Over cycles you learn which instinct beats the data — and which doesn’t.

  • Market corrections, labeled

    Electrification, fuel prices, a competitor’s exit — entered as explicit adjustments, visible in every affected row.

  • Cash exposure on one screen

    Stock value, aging, and committed orders — the finance view of the same data the buyer sees, updated every cycle.

  • Survives personnel change

    When the person who “just knew” the ordering rhythm leaves, the rhythm stays — encoded, documented, running.

  • Your data stays yours

    Runs in your cloud or ours, in the EU. Sales history is commercial intelligence — role-gated, audited, never used to train public models.

Dealer machinery yard seen from above, rows of units awaiting sale

Engagement

Shadow first. Trust is earned on record.

  1. Discover01

    Two weeks on your exports

    We take a season of sales exports and a current warehouse export — as they are, statuses in wrong fields included — and return a cleaned, joined dataset plus the first recommendation draft.

  2. Design02

    Your buying logic, encoded

    Lead times per manufacturer, minimum batches, the models you never gamble on — the rules your buyer applies by instinct, written down and agreed before automation touches them.

  3. Build03

    Shadow cycles

    For one or two monthly cycles the system recommends in parallel while your buyer decides as always. Recommendations are scored against what they would have done — trust is earned on record, not claimed.

  4. Operate04

    The monthly ritual

    Fresh exports in, recommendation out, one working session to decide. Optional retainer covers data drift, new manufacturers, and extending history for seasonality.

Frequently asked

Questions buyers and owners ask first.

  • Our sales data is messy — statuses typed into random fields, model names written five different ways. Is that a blocker?

    It is the starting condition of every dealer we have looked at, and it is exactly what the first two weeks address. AI normalizes model names, extracts statuses from free-text fields, and classifies ambiguous records — each with a confidence flag, so a person reviews the doubtful ones instead of all of them.

  • Does the system place orders automatically?

    No, and it should not. It produces a recommendation with the reasoning attached; your buyer decides. The economics are asymmetric — a wrong automated order costs real money for months, while reviewing a recommendation costs an hour. The system earns influence, not authority.

  • We only have a few months of clean history. Is that enough?

    Enough to start, not enough for seasonality. With one quarter the system recommends conservatively and says so. Each monthly cycle extends the record, and around the second year the seasonal pattern — what sells before harvest, what moves in Q4 — becomes part of the arithmetic.

  • How is 'in transit' handled? That's where we've been burned.

    As a first-class column. Units ordered but not yet arrived are subtracted before any recommendation is made, which is precisely the arithmetic that prevents ordering twenty units that are already on the water. If that register lives in a spreadsheet today, the system ingests it; if it lives nowhere, the system becomes it.

  • Can it recommend against the trend — say, more electric than history suggests?

    Market corrections are explicit inputs, not silent model opinions. If you decide the fleet is shifting to electric faster than history shows, that adjustment is entered, labeled, and visible in every affected recommendation — so you can tell afterwards which calls came from data and which from judgment.

Where this runs

Anywhere machines cross an ocean before they cross a yard.

Talk to us

Bring one quarter of exports. Leave with a recommendation.

Book a discovery call