When I introduce a new product how much should I make? Here's an idea: zero.

You've got a brand new product. You've never sold it before. In fact, no one has ever really sold anything quite like it. Also, it's not really one new product. It has many variations—color, size and so on—so it's more like dozens of new products.

Here's your problem. How many of each kind should you make?

It's a classic manufacturing dilemma. The typical approach is to guess. No one calls it guessing of course. You dress up the process to make it seem like you're not guessing. You have math and formulas. You have spreadsheets. You title the spreadsheets "forecasts" which is another word for "guesses" but sounds way more scientific. And you are wrong. Always. You make too many of some versions, too few of others.

Another approach is to not make any—at first. You wait until one is ordered, then build it to order. Sure, you have to have all the components on hand so those aren't built to order, but most of the components are probably shared between the different variations so it's not that big of a deal. The big benefit is that you never have the wrong level of finished goods inventory. It's always zero. The only products that exist are the ones that are already paid for.

Apple introduces their watch this week. Build-to-order is the approach that some believe they are taking. There are demo versions of watches out there but, when you buy one, the order goes to China and someone in a pink dust suit starts making your watch. 

Will Apple keep doing this forever? I don't think so. For one, FedExing a single watch at a time out of China is expensive. Another reason is that build-to-order creates a long lead time for the customer—days, if not a week or more (though Apple doesn't seem to care much about long lead times when they introduce products, to some extent it appears to be their strategy). If Apple does build to order they're probably doing it to learn about the demand. Once they see which versions sell in what quantities they can begin to build inventory ahead of time.

Silicon Valley wants to deliver your food

Restaurants and small food shops have always been flustered by delivery. On the one hand they could help customers—and find more of them—if they took orders online and delivered. On the other hand there's the problem of how to price delivery, the logistics of delivery and the problem of setting up an online order system and making sure its inventory is accurate.

In the last year there's been a wave of new Silicon valley start-ups that try to help with the last part—the online order system. 

The most prominet are Grubhub and Seamless. They take orders for restaurants. The restaurants figure out how to make the food and deliver it. Grubhub and Seamless take a cut that's probably around 20%. Speaking personally, I've used Seamless a lot in Brooklyn and it's very good. The benefit to a restaurant here is that they only have to figure out the logistics part of delivery. They can put all or just part of their menu online—and make it available at times that make sense to them.  Take Prime Meats in my neighborhood, a fancy restaurant that's full almost every night. They are on Seamless but in order to prevent overburdening their kitchen they initially showed up on Seamless only between 5 and 7pm, when they were slow.

In the novelty arena, you can also order pizza on a smartphone, albeit in a ridiculous way. There's a one button app that, when you push it, delivers pizza in 30 minutes. From somewhere. Anywhere.

In a more interesting twist, Square, the payment processing software company, is buying a food delivery company called Caviar. This is the only Silicon Valley firm I know trying to do the delivery part of the delivery business. Presumably the idea here is that a restaurant can buy their POS system from Square and the delivery software—and delivery logistics team—will come along with it.

This is happening with grocery, too. I just spent time at Bi-Rite and talked with the GM Patrick (ex-Zingerman's Deli manager) and learned about Instacart, which Bi-Rite just joined. With Instacart you place your grocery order online and they find someone to go get it from you. It's not an employee of the grocery store, it's not an employee of Instacart, it's just some shmo who signed up to be a grocery store picker. (They call them pickers, just like we do for people who pick items for boxes on our production line). Like with Grubhub and Seamless, Instacart takes a cut.

How these all play out will be interesting. Short-sighted merchants, or ones that do their own order and delivery, may look at the cut these companies take and say they don't need to pay someone else for something they can do on their own. The problem there is they will be shut out of network effects. The more merchants sign on to Seamless the more common it'll be for customers to shop there. If you're a merchant and you're not there, you'll loose out. Merchants will be saving cost to give up sales, which is rarely a good move.

What's the downside for the customer? On the restaurant side there seem to be very few negatives. Seamless doesn't mark up for delivery so why not order online and get the same food you could have driven to pick up brought to you for free? On the grocery side, I can see inventory being a hard nut to crack. Right now Instacart has no database connection between what's for sale online and what's in stock at the merchant. If something is sold out, the merchant has to remember to go to Instacart's website and mark it sold out. Will that happen? Sometimes, but not always. That will mean upset customers. Instacart gives leeway to their pickers to choose subs or call the customer to see what they'd like, but either answer is a flawed fix, one that will frustrate customers and hurt sales.