Thank you for the great feedback on Project Sage. After thinking about your comments, I have iterated a bit (in my head) and wanted to share my latest thinking with you.
The key objective of Project Sage is to capture collective wisdom on key assumptions made in spreadsheets — turning an individual spreadsheet into a wiki-spreadsheet, if you will.
So, here are some updated thoughts on Project Sage:
1. Allow users to upload any existing spreadsheet (Excel, Numbers, 1-2-3, Google Docs by URL). Like Slideshare, but for spreadsheets (Google and Zoho allow you to upload and share a spreadsheet today, but the use patterns we are focused on here are closer to the way Slideshare is being used).
2. Present a web preview of the spreadsheet to the creator. In creator view, show the data values in one color (blue) and formulas in another color (black). Allow the creator to turn any data value into a user-editable field. For example, a creator could upload income statement projections for the Fortune 500 into Sage. He could then make forward revenue and expense growth values / growth rates editable [by any registered user] fields.
3. Provide some basic visual editing tools, similar to those available at the major blogging providers. Provide templates, color and font selection, and allow the creator to edit his URL — use sage.com/date/name_of_spreadsheet as the default. Encourage the user to enter tags and related URLs for SEO. And provide basic analytics to the creator.
4. In advanced settings allow the creator to change the statistical function applied to the data munging process — arithmetic mean, geometric mean, exponential functions, Bayesian probability. Also allow third party developers to insert their own treatment with a snippet of Python or some other scripting language.
5. Insert a comments pane below each spreadsheet and give the creator comment editing and anti-SPAM tools — again, very much like what you would find on a blog.
That’s it. Now anyone with a spreadsheet can tap The Wisdom of Crowds.
4 Comments
May 24, 2008 at 5:29 pm
Looks like there have been some experiments in this direction, but not with the same focus:
http://wikicalc.org/products/wikicalc/index.html
I still think that you want more people to participate than would look at the actual spreadsheet. So using something like Google/Zoho’s polls to input data into the spreadsheet would be better than editing it directly.
Sam
May 24, 2008 at 5:51 pm
I thought Google Docs would have easily done this but lo and behold it falls short (but is still a GREAT tool).
http://spreadsheets.google.com/pub?key=pBcjTSZ0zXJ5gTjcHiDxYNA
I wanted to make the sheet much “prettier” but could not
Couldn’t find out how to lock the test formula down (either just because I don’t want new users to mess it up at first or because is my secret magic quotient)
There’s no publishing URL API like docs.google.com/me/meal-size
Some people might want to map their creations to their own URLs on their own servers
Wish I could style the published doc with regular CSS
Live starter feeds would be fun/cool/useful like wunderground, popclock, the dow, etc
May 24, 2008 at 9:29 pm
Make it a stock forecasting contest. Populate a bunch of spreadsheets with publicly available financials from companies (the fact that companies are slowly moving to XBRL will help but even if you had to use manual labor for the data entry, it’s not that big a deal, just do it for the, say, 500 companies in the S&P 500). Let the crowds collectively tweak the numbers, fiddle with the models, etc. See if the masses can beat the “analysts”. Motley Fool CAPS does something similar but at a coarser level. It would be an interesting experiment to see if the wisdom of crowds can accurately predict specific numbers like revenues by segment, expenses by category, margins, etc.
May 24, 2008 at 11:36 pm
Now you’re cooking with gas Jolly. Showing people you could beat analysts on projects is what I had in mind with the Fortune 500 projects bit. I think your idea of providing some sort of incentive (e.g., forecasting contest) is very smart. Or perhaps you could have some points system or something like that.
In business school we built crazy complicated five-year models that projected the income statement (revenue, R&D expense, tax rates, etc) and the balance sheet (cash, payables, receivables, etc) and used those projections to build discounted cash flow models (use income statement and balance sheet to project free cash flow, then make a ton of assumptions to calculate a discount rate and terminal value). After all of those assumptions, we would divide the aggregate discounted cash flows by fully diluted shares outstanding to get a stock price. I would often hold a single variable constant and use goal-seek (or would build a macro to do that) to impute what market expectations were for something like long-term revenue growth. The problem with this approach is that there is far too much complexity for one person to do this. It’s crazy to believe that such an approach could reasonably compete with the market. But, using market-like forces to attack the inputs that determine share price… Using a wiki-spreadsheet approach would definitely be better than analyst projections.
Leave a Reply