The Analyst: Which of These Orders Are Real?

Monday morning of the second week. Tomasz’s allocation went out on Friday and is already the most-read document in the building: 520 boards, named customers, dated tranches. Holding up every line of it sits a number nobody has yet examined — 960 firm orders in the twelve-week window.

Yusuf Rahman has spent the weekend on that 960 rather than on the allocation. In a shortage, every unit of demand that turns out not to be real has already been paid for in boards taken from somebody whose demand was.

What lands on Yusuf’s desk

From Tomasz comes the allocation and its logic: rank by margin per board, tempered by contractual and strategic reality, phased rather than all or nothing. His protected list — Harlow 180, Nordfoods 240, Bellwether 60, St Chad’s 40, Acme’s single oven — needs 521 boards against the 520 that exist.

None of that is Yusuf’s to change, and nor is Ravi’s instruction to protect first the relationships that cannot be bought back at any price. The denominator is. The 960 came out of the order book the way it always does — a monthly report exported to a spreadsheet, never interrogated, because until 3 March Calder could build everything it was asked for. Quoted since in a supplier call, a leadership meeting and a draft board note, it is load-bearing and untested.

What a demand analyst actually does

Most months the job is forgiving: build a demand plan, feed the schedule, watch the error, adjust. A shortage inverts that. March is already decided, so nobody needs a better forecast of it; they need a truthful account of what is promised, to whom and by when, because that is the ledger the boards are drawn against. The work turns forensic — and every colleague whose orders he examines has an interest in them looking firm.

The software on Yusuf’s desk

The order book in the ERP is where the trouble starts, because in most companies firm orders, forecast and salesperson optimism end up in a single column that nobody separates while supply is plentiful. SQL is how Yusuf separates them — a precise question about purchase order numbers and confirmations rather than a judgement about who is exaggerating.

The history comparison is the module’s real technique: Peninsula’s request only looks suspicious next to what they ordered in the same weeks last year and what they actually sold through. That is a chart, not an accusation, which is exactly why it works. The published file then carries the softness openly, because a demand number with its uncertainty attached survives challenge far better than a confident one that turns out to be wrong.

The decisions

Firm order, forecast, and hope

A firm order has a customer’s name, a purchase order number from that customer’s finance system, and a delivery commitment Calder confirmed in writing. A forecast is Calder’s assumption about what will probably be ordered. A pipeline opportunity is a salesperson’s belief that a deal will land.

All three get added together in one column headed “Demand”, and the reason is not incompetence: that column was built for capacity planning, where roughly right is adequate, and while supply is sufficient the error is invisible. Cap the supply and the same column starts allocating scarce parts — every overstated unit now takes a board from a real one.

Yusuf applies the test line by line, and 74 of the 960 fail it. Forty-six are call-off volumes under an annual agreement with a supermarket group — a real contract, but one committing to a price and a yearly total, not to units inside these twelve weeks. Twenty-eight are opportunities two salespeople entered as orders because the quoting workflow made that the quickest way to reserve a build slot. Nobody lied; the system rewarded it, and until March it cost nothing.

Double-counting, and the phantom demand a shortage manufactures

Peninsula Catering Group is on the book for 90 units. Being a distributor, they reach the order book by two routes: stocking their own warehouse, and end customers who negotiate with Calder’s sales team but transact through Peninsula for credit reasons. The second kind gets entered twice, once by the account manager and once by Peninsula’s purchasing system.

Reconciling them, 34 of Peninsula’s 90 have a matching end-customer line in the tail of roughly 200 smaller orders — same ovens, same kitchens, counted twice. Yusuf keeps the end-customer line, because it carries a real installation date.

That leaves 56 units of stocking order, and here the harder problem lives. Peninsula’s sell-through is about two and a half ovens a week — 30 units across twelve weeks — and in the same weeks last year they ordered 28. This year, 56, skewed towards Meridian, a line they barely sell. Their selling has not moved; their buying has doubled.

That is rational of them and destructive in aggregate. Expecting to be rationed, they order ahead of need — but everybody does it at once, each rung amplifies the one below, and the book claims the market wants more ovens just as it wants fewer. Those 26 phantom units, if honoured, are 26 boards that never reach Bellwether or St Chad’s.

He finds it by comparison, not by asking: a distributor denies inflation sincerely, since its own planner wants the stock. The order is measured against the same weeks last year, against sell-through, and against the mix they genuinely sell.

What customers need, as opposed to what they ordered

Every order carries a delivery date. Almost none carries a need-by date — the date after which somebody actually suffers. Normally the two are treated as identical, because nobody has had reason to separate them. In a shortage the gap is the most valuable unrecorded information in the company, and the only way to get it is to ask.

Bellwether is easy: sixty units against a dry dock booked with a third-party yard, where missing the window cancels the order outright. Yusuf still asks for the booking reference, because a date backed by somebody else’s contract is a fact and a date asserted by an account manager is an opinion.

Harlow is the instructive one. Their 180 Meridians serve a rolling refurbishment, floor by floor. The contractual schedule front-loads 120 units into the first six weeks; the programme itself needs 60, because the remaining floors do not close to guests until later. No units disappear, but the pressure changes shape — and the contractual date drives the liquidated damages, at 0.5% a week capped at 5%, while the need-by date drives the harm. Moving the first needs Harlow’s agreement, which is Marcus’s conversation and not Yusuf’s.

St Chad’s has 40 Compacts feeding a new kitchen with a handover date in week eleven, so they need them by week ten. Not “as soon as possible” — week ten, which means they can follow Bellwether without anyone suffering.

Asking when a customer truly needs something is uncomfortable, because it signals you may be late, and unusual, because most order books have never been asked. In a crisis it beats any forecasting technique: forecasting infers a date, asking obtains one.

The number he publishes, and its uncertainty

Seventy-four unfirm units, thirty-four duplicates and twenty-six inflated units come out, and demand restates from 960 to 826. Yusuf could sell that as good news and does not, because of where the reduction sits: almost all of it came from the unprotected tail and from Peninsula, and the protected 521 is untouched. The shortage is not smaller where it hurts, only more honestly described — and that sentence goes above the number, so nobody reads 826 as a rescue.

He publishes a range too, 790 to 870, naming the judgements that move it: whether Peninsula’s stocking claim is 30 units or 56, whether the twenty-eight pipeline lines convert inside the window, and whether two contract releases land in these twelve weeks or the next. He publishes the method as well — the firm test, the reconciliation rule, the sell-through comparison — so people argue with the logic rather than lobby for their own accounts.

He refuses a single confident number and gets pushed back on for it. But one number invites a debate about its last digit while hiding assumptions that could each move it by twenty units, and if it turns out wrong the method goes with it. A range that contains the truth earns the right to be believed next time. False precision is not more useful than an honest range, only more comfortable.

Being the person who says the number is wrong

Sales colleagues experience this work as being called liars, and handled badly that is what it becomes: a note circulates showing whose orders were softest, and the next time Yusuf needs a straight answer about a delivery date he does not get one. So he inverts the order of the work. The definition goes out before any findings, account owners classify their own lines against it rather than having them classified for them, and the aggregate is published without names. He runs the same test against Calder’s own promised lead times, quietly the most overstated numbers in the building.

Distributors overstate to secure allocation, salespeople to hold build slots, customers to get earlier dates, Calder to keep customers calm. All four are rational and all four pull the same way, which is why he never says your orders were wrong; he says we never agreed what an order was. The fix is a shared definition, not a blame conversation.

Where this goes wrong

The classic failure is not rejection. It is that 826 gets published while 960 carries on living in a board pack, three slide decks and the memory of everyone who heard it in week one. Within a fortnight the company holds two demand figures and each audience gets the convenient one — the big number for the supplier call, the small one for the allocation meeting. Nobody decides this, and it discredits both.

The defence is dull. Retire the old figure explicitly, in a dated note explaining the difference; make the new one the only figure feeding the allocation; re-run it weekly, because demand data does not rot slowly in a shortage.

What Yusuf hands on

To Tomasz goes a revised demand file: 826 firm units, each with a need-by date and the evidence behind it. Peninsula’s claim drops from 90 to 30, the unprotected tail falls from 349 to 275, and the protected 521 stands at the quantity he set. What changes is sequence, not size — Bellwether earliest and complete, St Chad’s forty by week ten, Harlow’s 180 re-tranched around which floors actually close when. The constraint he inherits is blunt: 521 against 520 is real, not an artefact of dirty data.

To Claire goes a smaller number that is harder rather than easier. Some of the original $5.3m exposure was never going to be booked in this window at all — but every remaining dollar sits against a customer with a dated commitment, so none can be waved away as timing.

To Marcus goes the list for module nine: accounts ranked by real urgency, with the evidence for each date beside it, so the argument about who gets protected is conducted in facts rather than volume. He also inherits the errand nobody wants — Peninsula’s order has been re-based from 90 units to 30, and somebody has to tell them.

Spot the decision

Read each situation and decide how an analyst should handle it, then tap a card to check.