At 5am on a Tuesday, we went in opposite directions.
I drove west, out toward the farms, in the dark, with coffee I'd made badly because I'd made it at 4:30. Fruit, vegetables, cut flowers. I want to be honest that this was the best part of my week. The sun would come up somewhere around the Quabbin and I'd get to a farm just as the crew was finishing their all-hands, a dozen people standing in a circle in the cold working out who was doing what. I'd load the van while they wrapped up. Nobody there needed me to be impressive. There was produce, and it had to get in the van.
My cofounder went the other way, into the city. Bakeries first, because bread waits for nobody, then the coffee roaster, then the butchers. Packaged goods and frozen came in out of band on their own schedule.
We ran a company called Natured, an online farmers market. Customers ordered from local producers on our site, and every Tuesday the two of us went and got it. We built it because we thought people should be able to buy food from the people who actually grew and made it, forty miles away, without that being a whole project. We believed that a lot. We still do.
The first thing we automated
Once we were working with enough producers to matter, the problem wasn't demand. It was that every producer had a different ordering cutoff, and there were more cutoffs in a week than two people could keep in their heads. We missed things. Missing a cutoff meant telling a customer that the thing they'd ordered wasn't coming, which is a small failure that feels much larger when you know the person's name.
So we built a just-in-time ordering system. It watched what was sitting in customers' carts and which item-level subscriptions were set, and it knew every producer's cutoff, because we'd typed each one into an admin tool from whatever price list they'd last sent us. When a deadline got close, it emailed us the suggested quantity and what to add, timed to land an hour before the cutoff.
This was 2019. There was nothing clever about it. Rules, a database, scheduled jobs. But I remember the first time an alert landed and it was simply right, and how good that felt, and it felt good every week after that. Two people cannot hold eleven deadlines. A cron job can.
It also had a manual heart, which is the joke of it. The system only knew a producer's cutoff and pricing because one of us had sat down and entered it by hand, and it only stayed true for as long as the last price list did.
Tuesday, in parallel
The split between our routes wasn't geography. It was perishability.
Produce is harvest-timed, which meant the long early drive west. Bread and roasted coffee are same-morning, which meant a tighter, denser loop around the city. Anything shelf-stable or frozen fell out of that rhythm and got handled separately, on the reasonable theory that it could wait.
At each stop the order would be staged and an invoice printed, sometimes sitting on a shelf inside the walk-in cooler, and I'd step into that cold to get it. Grab the paper, load the van, next stop. Cut flowers rode in the front with me, because they rode better there and because after four hours I liked having them in the front with me.
We'd converge at the warehouse around midday and eat lunch, which was often the first time either of us had sat down. Then we packed. We'd built ourselves a pack tool for this, which sequenced orders so we could move through the warehouse without crossing our own path forty times, and I was proud of that thing. It worked. There was always something more we wanted it to do, and there always would have been.
Then back into the vans for afternoon deliveries.
Where it stopped
The invoices came back to the office attached to the warehouse and went into a pile. They sat there until we needed to enter them into QuickBooks or check a due date, which meant that every so often one of us lost half a day to paying invoices and marking the books.
That contrast still bothers me. We had a system precise enough to give us a one-hour buffer on a produce cutoff, and it fed a stack of paper on a desk that got dealt with whenever somebody got around to it.
And the ceiling was never demand. It was two people, two vehicles, and one day. Every deadline that ordering system tracked converged on a single Tuesday, and there was no slack in it anywhere.
We couldn't get Natured to the volume it needed. We worked at it for a couple of years, hard, the kind of hard where you stop having other things, and then we shut it down. It wasn't for lack of tooling. We had the demand sensing, we had the pack tool, we had our own marketplace and customers who genuinely loved it. The work per producer just never came down, and past a certain point no amount of cleverness on our end changed that.
That was a bad few weeks. I'd recommend the 5am drives to anyone. I would not recommend the last email.
What we were actually looking at
Here's what took us far too long to see, and what makes me think those two years weren't wasted.
We weren't a peer of the producers on those routes. We were one of their wholesale accounts. Every Tuesday we were the customer placing the order, which meant we placed it however each of them preferred: email for some, a phone call for others, a form on a website. We adapted to every one of them, because that's what a customer does.
So we had a clear view of the other side. A producer with a dozen accounts was doing our version of Tuesday a dozen times over, in a dozen different formats, and without the option we had of going and building themselves a database of everyone's cutoffs. They had a phone, an inbox, and a spreadsheet per customer. We were watching people we admired spend their evenings on data entry instead of on the thing they were actually extraordinary at.
And we had been building them a new sales channel. That was our whole plan. For people whose real constraint was the channels they already had, another storefront isn't relief. It's one more inbox.
So we stopped, and started turning our code into something else.
What Foodstack is
Foodstack is the operations layer we built for ourselves, pointed at a producer's own wholesale business instead of at ours.
Your customers order from your storefront, on your order windows and your terms. Orders arrive already grouped into a pack list. Invoices go out when the order does, so there's no pile and no half day. Your customer list is a list, not something you carry around.
The reason this matters isn't tidiness. The producers we bought from weren't trying to survive, they were trying to grow, and they kept hitting the same wall we did: the tenth account cost more attention than the first nine put together. Not because making more was hard. Because tracking more was.
I still think about that circle of people in the cold at 6:30 in the morning, about to go do the hard part of their day. None of the hard part should have been paperwork.
We ran the manual version, for two years, badly enough that it closed. We know exactly which parts we'd want handled.