Early Access Offer to New CellarTracker Drinkability Feature

Yes the “Wine Data & Insights” notification setting controls here. If you had that unchecked then you will not get the email. Checking it later won’t resurrect the email unless and until the team goes back and resends a private, new campaign just for you. I don’t know the feasibility of that.

Thanks everyone for the feedback and do keep it coming! We are working on refining this for broader release and incorporating these comments. I have a few specifics to address, but want to make an overall response on something in particular.

There are a lot of responses of the type, “I’d never open XXXX now!” That type of information is already encoded in the drinking windows. The logic behind these reports is exactly about confronting your priors about a wine. Maybe you are wrong and maybe the data is unearthing a new piece of information. Or not - that is fine as well. But the whole premise is provide at least parallel information to pre-formed ideas about a drinking window and ideally updated information. I think we at CT are not communicating that well.

On a few like the Cos 16 - there is a refinement we need to make to exclude bottles sold which can distort the trending data. In some cases like one of the vintage ports, I think there was a sale but the user didn’t record it as such. Nothing we can do in that case. The trending data is always going to have some lapses or oddities. eg - 94 Ports showed up for someone and I suspect there was a uptick because people like multiples of 10 (ie 30 years on those.)

The percent increase can be viewed like this : Say typically across time, 10 bottles per quarter of Wine X were consumed. But then in the last quarter, 15 were consumed. This would 1.5 times or a 50% increase.

Re tasting notes vs consumption data. These are two different sets of data that have less commonality than one might predict. A lot of users consume w/o tastings notes - this is what we are trying to capture here. And there are plenty of notes that are not from individually owned/consumed bottles (at tastings/etc.)

I agree the AI prompt crafting needs some work. I am wary of overly-crafting the prompt though as I think the distortions that might induce will be worse than the somewhat contradictory summaries it does give which might simply be an artifact of what I said above - notes are more discrete from literal bottle consumption that expected.

Is the goal to send out this report weekly or monthly for testing? Or was this a one-time experiment?

Is this just a high level example to help understand the %increase or is the current algorithm comparing Q1-2024 to the average (or similar) of the previous x quarters? I’m trying to cross reference with my own data to better understand the items in my Heating Up! list.

Can you elaborate on the “accounting for seasonality” mentioned in the email?

There are no Drink Now! suggestions in my report which I believe is to be expected. I already monitor the community consumption % of the wine in my cellar and high consumption% already influences how soon I’ll open a bottle.

“Accounting for seasonality” - we expect and the data shows seasonal variance like “rosé season” or sparklers at New Years. This has been smoothed out.

I will check what exactly what is being measured, but I am pretty sure you have it right - average of n bottles consumed in a quarter and we are reporting where the past quarter is significantly above n.

1 Like

My caveat is that folks aren’t drinking vintage Port like it’s going out of style. (One could argue that it’s already out of style.)

And I don’t know if “significantly above n” has been defined.

Heh. I ran some data analysis I think I posted here at the end of last year about how Port had the largest drop in rank of holdings across the lifespan of CT.

That ‘significantly above’ is something we are testing. It is less a pure statistic and more trial-and-error and utility. Make it too tight and the reports are anemic. Make it too loose and the reports are overly full. So we basically report the top 10 above par in your cellar. Some will end up with 4-5x and some more like 1.3.

Are you normalizing for total bottles purchased? A 5-bottle increase is very different in a base of, say, 100 bottles versus 1,000 bottles. Wondering if that’s why some big production wines showed up in my “heating up” list. It was a big increase of people trying the wines (e.g., 2020 Montebello and people checking for smoke taint) but overall tiny relative to the base.

(And sorry if this has already been discussed)

A true data scientist did the work here. Everything was statistically significant.

1 Like

One other comment, nicely summed about by our data scientist :

The trends are tellings us that some noticeable number of people feel a need to drink this wine, but the contrary tasting notes are showing us that they were quite possibly wrong to do so. Seeing the notes and the consumption data as being in opposition is wrong. The data isn’t telling us that you should drink a wine, just the people are. The notes can be illustrative that the reasons for doing so aren’t always purely based on maximizing the wine enjoyment (eg “checking in.”)

3 Likes

It would be nice if there was a way to define the stage a wine is in, or how a user likes their wine. If one user likes younger, more primary wine and another likes older, more tertiary, theres going to be a wildly different impression.

That said, Im not sure where it gets to granular for the app and where users just need to read the tasting notes.

That makes perfect sense. The question is whether the feature could be improved by somehow identifying both an increase in consumption AND the resulting tasting notes being ready.

It’s fine as it is now so long as it’s clear to users that it’s just a “take a look at this wine and review the notes” but if you can combine both data points, it’s even more powerful - albeit very tricky to implement now that i understand your approach

Thanks for sharing all the insight and I love the idea either way.

Two of my wines on “shriveling up” list, both Cabot, last bottle was obviously the best of each.

  • 2011 Cabot Vineyards Pinot Noir Nash Mill - USA, California, North Coast, Anderson Valley (5/15/2024)
    Last is the best.

    20 cases. 10 years. Most poured while my wife was serving as a US Ambassador to Switzerland and Liechtenstein.

    This is just stellar. Bitter. Sweet. Sour. Earthy. Lots of mushroom and tilled earth. But an impressive amount of dark fruit, really multi-dimensional. Not a shy wine at 13.8%. But really quite balanced and pleasing. Very Pinot with a ton of twangy, funky fun. Rhubarb. I was worried this would be tired, but frankly it had barely budged in 7 years. Tons and tons of life in the tank. (91 points)
  • 2007 Cabot Vineyards Syrah Humboldt County - USA, California, North Coast, Humboldt County (5/10/2024)
    Last bottle is the best?

    This was EPIC. I drink a lot of Syrah, best in the world, and when you gat a Cabot that is on it twinges pretty much all of the feelings of some of the best.

    This is labeled at 14.4% but drinks so fresh. Blackberry and funk on the nose. Black pepper. Olive. These are always so reminiscent of a Cornas flavor profile, but is is weightier and fruitier. Strong black fruit elements. Mmm, this is so tasty, fully resolved, a nice streak of acidity, just such a lovely expression of Syrah.

    Absolutely fantastic. Humboldt County Syrah at 17 years?!? (93 points)

Posted from CellarTracker

1 Like

That is actively under discussion. The prompt engineering is a bit tricky and might not make it to the first implementation, but it is definitely something we’d like to do!

1 Like

This is the comments I provided to the team:

Thanks for including me in the beta for this new service.

In general, I agree (or at least understand why included) the drink up recommendations. The only comment I have is if bottle size is being considered. One of my bottles was a magnum and one is a 9L.

In general, the heating up category seems to be serving up many wines very early in the drinking range. I understand that some of this is due to the fact that I may hold on to wine longer than the average CellarTracker user. But in at least one case (2021 Carlisle Montafi Zin), while the wine had a 61% increase in consumption, only 7% of the overall bottles have been consumed so this seems like a little early to be heating up. In a few cases, the AI generated review and/or the actual recent CellarTracker reference the wine being young or will improve with age so I don’t look at them as showing signs of being ready to drink. The AI generated review for 2018 Carlisle Petite Sirah Palisades references a specific review and say it says something that isn’t in the review nor can I find it in any other review.

I was surprised by the limited number of tasting notes on the wines that had a big increase in consumption but I guess I am part of that problem since I write less and less reviews these days.

1 Like

Sign up closed

1 Like

Not a part of the beta, but just reading the comments. This comment struck me as another way to cut the data would be to look at the preferences of the users vs the particular bottle of wine in question. I have a taste profile, as do we all. It can be referenced by the bottles in the cellar, consumed, scores and notes. It can be reported for each user along with the earlier trends and charts reporting, and that taste may evolve over time. Then, when I opine on/score a bottle that information can be compared to my baseline to determine how much weight my opinion, statistically, should have on the global reporting. Just a data geek thought prompted by Max_S1.

2 Likes