Shared publicly  - 
 
My post comparing Spyfu.com data to my own data. Found some interesting insights...
1
Mike Roberts (mrspy)'s profile photoLuke Alley's profile photo
9 comments
 
Hey Luke. Good article. I dropped a comment. Let me know if it makes sense.
 
Thanks Mike. I didn't see your comment. Can you post it again?
 
Ah, damn. I'll bet it spam-filtered it. I included a screencast link. I'll post it later. St. Patrick's drinking needs to get done first ;)
 
Just tried to repost, and it still won't go. Maybe you could post it? Here it is:

Hey Luke,

Mike Roberts with SpyFu. This is cool stuff.

You're definitely right that our budgets are search only; no content network, no partner sites.

When we publish the budget numbers we post a range. Something like $165-$330. The fewer data points we have the wider that range is; it's an expression of uncertainty --> very much like presidential polling sampling errors (+-6%). The larger the sample size (inclusive of history), the more accurate the results.

I'm curious what you did to turn our budget into one number. Did you take the bottom number, the top number, the mean?

I'd be curious how often your client's budget falls within the actual range.

In any case, calculating someone's ad budget from what we see them advertise on is akin to predicting intelligent life on other planets based on the wavelength of their star. There are so many factors that determine budget (Quality score, bid strategy, position, cost per click [which we get from Google, but they don't report accurately], Shopping, Videos, Images)

Here's the thing: When we say someone advertises on a keyword, it's 100% true -- there's a cache page you can use as an audit trail going back to 2006. Ad budget is the biggest extrapolation we make; the farthest from the raw data. My goal when I first calculated that number was to make it so that given that I know my own Adwords budget, can I estimate how much bigger (or smaller) a player my competitors are in the market. Then, secondarily, I wanted to be able to roughly gauge the size of a client, competitor, or partner; are they spending hundreds, thousands, tens of thousands, millions? It's kind of hard to tell by looking at someone's website, right? Make sense?

Anyway, I love what you're doing. Thx.
 
Thanks a lot for taking the time to respond Mike. I hope I didn't come across as bashing Spyfu. We are actually happy users of your product! I tried to find an easy way for everyone to digest the data when compared to my own accounts and I plan on looking at other tools also.

As far as the budget goes we took the mean of of your range. I'll go ahead and add a column on my graphs to say whether or not my clients fell within that range.

If you don't mind, I would love to add your response to my post. Would you be ok with that?
 
Luke - dude, that's awesome, thanks for adding my response in there. Definitely doesn't come across as bashing; I like it a lot; hopefully I don't come across as overly defensive.
I love this stuff. It's cool the things you observe about the numbers that I can confirm in the algorithm (because I wrote them). For example, I just re-read the article and noticed you talking about accounts less than 3 months old being less accurate -- I've got to post another comment, because I can tell you exactly why that occurs.
Keep up the good work. Cheers.
 
Well, whatever. The blog still wouldn't let me post to it. Here's what I would have posted:


Hey Luke,

Thanks for posting my response; that's awesome.

I just re-read the article and noticed a part that I hadn't before:
"Numbers were more accurate for accounts with budgets that were consistent and unchanging for several months.

Seasonal clients and accounts with less than 3 months data were significantly off."

It's really interesting that you notice this. I'm impressed, because this speaks to a subtlety about the system that I didn't think anybody had ever noticed.

Sometime last year, we started smoothing our budget estimates using a weighted rolling average.

So, if we see your domain show up on a keyword this month, but not last month or the month before, we won't assign all of it's value to your budget. In fact, you aren't assigned a keywords full budget until you've been on it for 3 or more months.

So, that's why accounts that are less than 3 months old are inaccurate; we value historical bidding quite a bit, and it takes that much history for our estimates to reach "normal" levels.

I'm totally impressed that you were able to observe the data and conclude what was going on with so few data points. Very cool.
 
I apologize about the issues with the comments. I tried to do a post in the comments and it didn't work for me either. We'll get that fixed. I think it's important that people understand how the tool can be best utilized so I really appreciate you taking the time to explain this. Definitely has helped me. I'd like for people to see your explanations so we'll get on that!

If we figure something out I'll just add your comment manually. That cool?

I'm glad I guessed right. I definitely didn't have significant data to base my conclusions of off, so I'm glad I got it right! Thanks for confirming.

Thanks for making a useful SEM tool and all the feedback!
Add a comment...