Thought I'd share some behind-the-scenes product development, ops, and science stuff we're working on.
So, over at Ultraworking we launched Clockwork Nutrition on the 4th of July.
100% automated basically-perfect-nutrition. Algorithms with a human touch, two large meals show up once per day to hit body composition and performance goals with basically no time required thinking about, purchasing, or preparing food. Then a very simple software interface to rank recent performance and enjoyment that we use to further dial things in —
We had a great reception to it. The first full week started last Monday and it's going well.
I imagine most people have seen the general concept and theory behind it, but if not, you can read all of that here.
The people who bought tended to be very scientifically literate and interested in performance, so I wrote up a little informal update.
I thought it might be interesting from a business, technical, and scientific standpoint in general, hence posting it here.
Subject: Clockwork - unofficial Sebastian Week 1 thoughts
Friends, Romans, Countrymen —
Unofficial note from Sebastian here. No obligation to read this or reply — just wanted to share thoughts and invite dialogue. Feedback is always welcome!
[Note: If you ever need/want customer service, email@example.com is the right email; I'm a slow email correspondent at times. To have the whole team read your message, reply-all instead of just to me.]
So, this was our first week of Clockwork — and my first week of letting the team entirely run my meals; I wasn't involved in selecting meals or scheduling them in any way.
Quick thoughts and observations —
*I really enjoyed literally not having to think about meal choices. The difference between some thinking/scheduling and none was terrific.
*We recognized pretty quickly that a lot of people didn't remember what time they put in for deliveries. In fact, I myself forget — and wasn't sure if dinner was going to come at 5PM or 6PM. We'll remedy this going forwards; either with calendar invites or possibly Twilio messages. Those little things are easy to forget, eh?
*Adaptation: I've been hungry, to some extent. I requested the most aggressive caloric deficit for myself. I'm thinking the magnitude of that effect is primarily just going through an adaptation period — it's not really effecting me when I'm working or my performance, but I do sometimes notice it when I'm between activities. I've been doing some research and trying to figure out how long the adaptation period towards equilibrium would be... but that depends on what mechanisms are causing this. I suspect the biggest factor here is probably that "blah sugar" has dropped tremendously from my diet and, after an adaptation window, I think the hunger will decrease substantially even at this (moderately aggressive) caloric deficit level.
*"Ranking shyness" - despite being rather hungry (by design, of course) - I was only marking "Higher than normal" for hunger rather than "Exceptionally high." This is because, more-or-less I know that "exceptionally high" means "hey team, consider increasing those calories" and I'm fine with where I'm at. But I wonder if people will be shy about ranking hunger high or noting chance of breakdown... something to be mindful about as we do interface design.
*Bottles of water: In one of those "grrr argh" moments, one of the delivery services we're using advertised free delivery if you had a premium account and the order is above $10, but their actual threshold got changed (and not documented) to $15 for free delivery. So, sometimes an order would come out to $13.50 + $4 delivery charge, but if we added a cheap no-calorie item (bottle of water) then the total price would drop. This is kind of stupid on many levels, we know it's stupid, and we'll keep investigating this. This is why you might be getting bottles of water from time to time. Over time, if this holds, we might actually be better off scouting for slightly more expensive items that offer larger volume of leafy green vegetables, or more "interesting" foods, or whatever. We'll keep working on it.
*Variety: Relatively low, and I'm actually not sure how I feel about it. There were days that was I like "oh, this is similar to what I had before" and was bored. But then I thought, "Is that good or bad? Do I actually care about variety?" Furthermore, when I got a steak on one day, it was like a special treat and I enjoyed it more than I otherwise would have. Still thinking and musing on this.
*Tradeoffs, Reflections, etc: We're thinking about how to put together a good (but still super fast) way to debrief and review a week. Maybe some levers like "give me more variety"? The challenge with interface design is keeping things simple and fast while giving disproportionally useful benefits for the time it takes to reason through and make adjustments. We're still thinking about this.
*"Heads up" A/B similar meal varieties: it occurs to me that it might be a good idea to flag you, from time to time, that two meals are quite similar and ask which variety you prefer. Obvious example - Burrito vs burrito bowl. I prefer the bowl a little, but not consistently every time necessarily. Likewise, for side dishes to a main course, "sweet potato (~115 cal) + more carby salad" vs "baked potato (~165 cal) + less carby salad"... these little local optimizations are possible if we figure out an elegant way to do it. Even then, there's still some benefit to variety and rotation of food choices, but I wonder if there's some way to flag "heads up" tests once or twice per week to dial things in towards your preferences. The challenge here is setting it up so that (1) it's worth the time it takes to think and rank heads-up on things and (2) some way to rank/evaluate/settle learned lessons in terms of algorithms/databases/profiles, so it doesn't result in excessively more complexity as we do these things. We might be able to just infer heads-up preferences from Meal Enjoyment rankings but that might need a longer sample size than just asking you? But then, complexity management and minimizing time/friction required. And storing all the preference things in some schema so ordering complexity doesn't go up. We're thinking about this.
*Business/ops side of things: waaaay more complicated than we'd like with all kinds of stupid details, but no catastrophic failures. Pleased with week one logistically/operationally, hats off to the team.
I haven't sat and gone through all the stats and patterns yet with the team and will do so soon, just wanted to share my quick thoughts and how things are going.
If this sort of thing bores you, let me know and I won't send unofficial updates — just, as a cool thing, 100% of people who bought in the first wave were either a business owner, programmer, or both — with a pretty high level of scientific literacy and curiosity about the topics. Hence, a little informal uptake from over here. Thought it might be interesting for you.
Feel free (but not at all obligated) to reply with any thoughts or preferences; do please reply all if you want the whole team to get those points.
Appreciate being able to serve you! Salutations and warm regards,
Overall, it's been pretty cool — lots of details to get right and refine, but enjoyable so far. Very excited about the future of Clockwork Nutrition — and hope you enjoyed a little inside look in how we're thinking about product development, science, performance, ops, etc.