Data Operations

Share this post
Building a better decision model
www.finddataops.com

Building a better decision model

"Everything Starts Out Looking Like a Toy" #78

Greg Meyer
Jan 24, 2022
Share this post
Building a better decision model
www.finddataops.com
Photo by Alina Grubnyak on Unsplash

Join smart, curious folks to get the Data Ops šŸ“Š newsletter each week (it’s free!)


Hi, I’m Greg šŸ‘‹! I publish this newsletter on finding data products and interesting data observations with the goal of finding patterns and future product insights. (Also, it’s fun.) If you need a background on how we got here, check out What is Data Operations?

This week’s toy:Ā a very minimal typing tester to help you get rid of those pesky errors. It’s not Mavis Beacon, but it does give you a quick way to get better (maybe you can even stop looking at the keyboard as you type now.) Edition #78 of this newsletter is here - it’s January 24, 2022.


The Big Idea

A short long-form essay about data things

āš™ļø Building a better decision model

Photo by Clint Adair on Unsplash

What can you do with a spreadsheet?

Yes, I’m talking about Excel, Google Sheets, and all of the other Software as a service tools that let you put numbers and formulas into a grid. Because the logic is expressed in a model allowing you to plug in your own data, it’s an easy way to share a functional solution to a problem.

Why use a spreadsheet? To some, it is a simple calculator. To others, it is a way to create financial models or other decisions. Spreadsheets have lots of limitations. It’s very hard to use for people who don’t know spreadsheets. It’s even harder to use spreadsheets as a multiplayer game.

In the hands of Tyler Robertson, spreadsheets are a programming canvas. Here are a few examples he’s built without resorting to external programming:

Twitter avatar for @aTylerRobertson
Tyler Robertson @aTylerRobertson
Finally updating my spreadsheet portfolio a little bit šŸŽØ docs.google.com/spreadsheets/d…
Image
10:20 AM āˆ™ Jan 20, 2022
3Likes1Retweet

Tyler Robinson’s Google Sheets suggest a future where Spreadsheets behave more like deployable applications. The ā€œgridā€ of rows and columns is a low-resolution vision of what’s possible.

Why Use Spreadsheets?

Spreadsheets let you express logic on a canvas. Because the information is displayed on a grid, we tend to think of it as bound to x and y dimensions. But perhaps we should be thinking of spreadsheets as the predecessors to a new kind of functional programming.

For example:

Excel supports 1,048,576 rows by 16,384 columns (about 17 million cells)

Google Sheets? Up to 5 million cells for spreadsheets that are created in or converted to Google Sheets with 40,000 new rows at a time, and a maximum of 18,278 columns

Spreadsheets, if you consider them analogous to digital cameras, have hit an information level of 17m items. This is a lot of information, and most people don’t come close to exercising the format to its limits like Tyler does in the tweet above.

But where could they go? The human eye has a theoretical limit of 576 megapixels - meaning that even at current limits it’s probably 10-20x as sharp as the best digital photos. Our current information resolution in spreadsheets is like an 8-bit video game. If you zoom out things look pretty cool but when you focus they are pretty chunky.

If the brain is an analog to a spreadsheet, we’ve got a long way to go before we can store the depth of information (number of rows), the complexity of model (intelligence?), and the speed of recall (search).

Functional Programming in a Spreadsheet

We don’t need to invent a whole new model to improve spreadsheets along these dimensions. We do need to improve spreadsheets in areas where they have low information resolution.

What would this look like?

A better spreadsheet might:

  • store and retrieve many more rows (like a database)

  • search in flexible ways (like a graph database or a highly tuned index like Elastic or Algolia)

  • provide suggestions for next actions based on the schema, structure, and model of the spreadsheet (exploring ā€œwhat ifā€ scenarios based on a known structure)

All of these improvements seem straightforward. You want to avoid having a practical limit to rows, unless you are storing some unimaginably large set of data. You also want to create a way to search and filter for information that lets multiple people use a spreadsheet model without breaking each other’s logic. And you want to provide ā€œintelligentā€ next action suggestions based on what the user is doing.

But implementing any one of them is fraught with peril.

What is a ā€œlargeā€ number of rows? How do you let multiple people use a model without breaking it and also without prescribing the way that they do things? Who knows when intelligent next actions deliver any real value?

Even a smarter spreadsheet with many fewer limits requires the right logic, placed in the right sequence, and a user experience that makes it easy for the user to complete the job they tried to do.

AI will not save the day

Pointing an AI model at many many spreadsheets probably won’t give you the right answers, either. Here’s a humorous example that Janelle Shane created with GPT-3 where she asked it to make new breakfast cereals from existing examples. One can only imagine the weird spreadsheets GPT-3 might make out of spreadsheet templates.

Twitter avatar for @JanelleCShane
Janelle Shane @JanelleCShane
I used GPT-3 to generate new breakfast cereals and descriptions. The cereals from Ada, the smallest GPT-3 model, were a bit questionable. aiweirdness.com/new-breakfast-…
OrbCrumpets - Bright yellow biscuit made with only high quality, raw ingredients.
Foam Squares – Chewy and colorful fun fun fun foam squares deliver the fun flavor of muffins and covered train cars !!!
Gudgetz Tallow Rolls - a science developed recipe.
Non-Stick Carbonated Waffle Balls - Gold standard for fun waffle ball innovation using wheat-based flavors and crispy, aluminum foil
4:32 PM āˆ™ Jan 21, 2022
418Likes105Retweets

Machine learning is not an inherently bad idea. But it needs to be focused on small changes across many observations to deliver solid results.

A new way to spreadsheet

Tyler Robertson’s spreadsheets suggest a new way to build decision models in Spreadsheets that looks more like functional programming.

Spreadsheets need reusable building blocks. These would work like pluggable models that define their own logic and work independently.

What if …

  • you could compose a spreadsheet model from other functions?

  • in doing so you could identify the Inputs needed to run that model and automatically connect them to the corresponding values in your current spreadsheets

  • and then connect that model in turn to another?

This would give you the basic model that we use today in spreadsheets – a row-based way to manipulate numbers with formula functions – and a more flexible, pluggable model to add logic that helps with particular problems.

ā€œFunctionsā€ then are run within individual spreadsheet modules, taking as input the output from other modules, and delivering the results down the line.

The application (because functionally, this is an application) is able to do much more than a simple spreadsheet model. And it retains the basic abilities of a spreadsheet.

Voltron: Legendary Defender LEGO set is truly legendary | EW.com
Like Lego Voltron, Spreadsheets combine to make a legendary defender

What’s the takeaway?Ā Spreadsheets are logical building blocks to assemble into applications that run in context of the data you provide. Most of the time we think of these things as disconnected and starting to think of them as logical data pipelines provides a lot more flexibility, power, and utility.


A Thread from This Week

A Twitter thread to dive into a topic

Calls would be easier to screen with context (people sometimes text for permission to call). What if there were a feature for this?

Twitter avatar for @danmall
Dan Mall @danmall
My wife has wanted to feature on her phone for a while where people have to let you know WHY they’re calling you, so I decided to sketch it out for fun. 🧵
2:27 PM āˆ™ Jan 19, 2022
4,830Likes665Retweets

Links for Reading and Sharing

These are links that caught my šŸ‘€

1/Ā Do not adjust your television -Ā why do certain color palettes, especially in movies, look so strange? Emily VanDerWerff takes a look at digital color painting in all kinds of video formats and comes up with some reasons why they look, well, dull. Makes you long for the days of Technicolor.

2/Ā Tipping point for evs? -Ā Porsche’s Taycan outsold the 911 brand last year, despite being more expensive. When electric vehicles are faster, more luxurious, and smarter than their ICE contenders, why wouldn’t people want them more? Now the task begins to bring EVs down to the same cost as a ā€œregularā€ car, and improving the ubiquity of chargers so that you don’t have to worry where to ā€œfill up.ā€

3/Ā Create the future, or re-create the past? -Ā this fascinating story of Japanese craftspeople creating the perfect American blue jeans is another good reminder of the importance of iconic design. When you keep removing features that are unnecessary, you get a result retaining only the features or facets that people value. But copying a design (5 pocket pants with rivets) does not create a classic. There are certain product attributes that become timeless when they just work.


On the Reading/Watching List

Some things to watch or read, in no particular order

Watching: Picard. Back to the Future. I love Star Trek retcons … especially when they tease a Star Trek multiverse. (The series starts March 3.)

Reading: Re-reading Clayton Christensen’s classic article from 2016, ā€œKnow your customers’ ā€˜Jobs to be Doneā€™ā€ reminds me that features, deliverables, and design don’t really matter if your customer doesn’t identify that your product is the one to solve the problem they have. The core of Jobs to be Done as a design philosophy is the belief that you are making the customer better by helping them achieve what *they* want to do. The tricky part is that they might not know what they want to do - a great product or feature leads them to the same place they would get to if they knew the answer.


What to do next

Hit reply if you’ve got links to share, data stories, or want to say hello.

I’m grateful you read this far. Thank you.Ā If you found this useful, consider sharing with a friend.

Share

Want more essays? ReadĀ on Data OperationsĀ or other writings atĀ gregmeyer.com.


The next big thing always starts out being dismissed as a ā€œtoy.ā€ -Ā Chris Dixon

Share this post
Building a better decision model
www.finddataops.com
Comments
TopNewCommunity

No posts

Ready for more?

Ā© 2023 Greg Meyer
Privacy āˆ™ Terms āˆ™ Collection notice
Start WritingGet the app
SubstackĀ is the home for great writing