Why Did Dell Beat Compaq When Both Bought the Same Parts?
Compaq had the same suppliers, the same parts, and by 1997 the same plan. Dell passed it anyway in 1999. A plan can be announced. A record cannot.
Compaq built computers and hoped the right people would want them. Dell built the ones people had already bought. Today every business can rent the same AI, so the record you keep is the one part nobody can copy.
What did Dell do differently?
Both companies made the same kind of machine out of the same parts anybody could buy — Intel chips, Microsoft Windows, screens and disks. The difference was when they built. Compaq built first and hoped the right customers turned up. Dell waited for the order, then built that machine. Dell's own 1998 annual report says its computers were "manufactured one at a time, as ordered," and that it "ended the year with seven days of supply in inventory, versus 80 days or more combined for indirect companies and their resellers." In plain words: Dell had about a week of parts in the building. The old way tied up months. Compaq announced the same plan in 1997. Dell passed it anyway in 1999. The parts could be copied. The plan was public. The record of who ordered what could not be — and that kind of record is what a business memory is for.
| Building from a guess | Building from the record | |
|---|---|---|
| What starts the work | A guess about next season | A customer who already bought |
| Who wrote it down | Your own people, months early | The customer, today |
| How you find out it was wrong | The stock will not sell | You do not — it was never built |
| What sits in the warehouse | Machines nobody asked for | Parts for machines already sold |
| What a rival needs to copy it | Nothing. Anyone can guess. | Years of real orders |
What is the difference between a forecast and a record?
A forecast says what might happen. A record says what did.
That sounds obvious. Here is the part that is not: inside a company, both of them come out as paperwork. The guess gets typed up, printed, passed around and signed off. By Friday it is being obeyed exactly like a fact.
So the factory is not guessing. The factory is following instructions. The guess happened weeks earlier, in a room, and then it put on a suit.
That is how a well-run company fills a warehouse with machines nobody asked for and never feels careless doing it.
Did Compaq know?
Yes. And it said so out loud.
In July 1997, Bill Ramsey, Compaq's vice president of manufacturing strategy and technology, named the trouble with the old way: "The disadvantages of an indirect model today is that you have a lot of computers out in the channel and the value of that inventory drops quickly over time."
Those are his words. Here they are in mine: Compaq built the machines first, sent them out to shops and dealers, and every week they sat there unsold they were worth less.
The same announcement said Compaq would start selling straight to customers, and start building machines only after somebody ordered them.
So Compaq had the same suppliers, the same parts, and now the same plan.
Then came 1999. Dell sold 7.02 million PCs in the United States and took 16% of the market. Compaq sold 6.86 million and took 15.7%. Dataquest and IDC — the two firms that count computer sales for the industry — both put Dell in first place for the first time.
A plan can be announced on a Monday morning. A record of who ordered what only arrives one real customer at a time.
What does this have to do with AI?
Every business on your street can now rent the same AI you can. Same models, same price, same week.
That is the parts bin, and it is open to everybody.
So the machine stopped being the edge. What you hand it is the edge.
An AI reading your real record — what you quoted, what you were paid, what you promised, who pays late — is answering about your business. The same AI with nothing to read gives you a guess instead. A confident, well-written guess.
Honestly, both come back sounding equally sure of themselves. That is the whole problem.
Where does this happen in a small business?
Every week, and it never looks reckless.
You price a job from what you think the last one cost. You order stock from what you think sold last spring. You tell a customer yes because you think there is room in the schedule.
Every one of those starts with I think. And every one of them already happened once and got written down somewhere nobody opened.
Now the honest part, so you hear it from me. Not every business can wait for the order. If you bake bread, you bake before anybody walks in, and guessing is part of the job — it always will be. The question is narrower than "stop guessing." It is whether you are guessing on top of what already happened, or instead of it.
So here is what I do now, and I am still learning it. Before I decide something that costs money, I ask my own records what happened the last three times. Not what I remember. Not what I expect. The answer comes back with the file behind it, and it argues with me more often than I would like.
That is what a business memory is for. A business memory is a private record of everything your business knows — what was agreed, paid, promised and decided — kept on your own computer as plain text files any AI can read, so you can ask anything and get the answer with the exact reference behind it.
Dell and Compaq reached into the same bin of parts.
Only one of them built what a customer had already asked for.
Records were never meant for storage. Give your business a memory.