In the previous two posts on the Economics of Analytics (I), I laid out an overview of the problems I was having – specifically with the Hourly Model.
In the Economics of Analytics (II), I argued that the agent-client relationship suffers from two horrible asymmetries in knowledge that cause the process of project estimation to become muddled and complex.
In this post, I’ll write about the scalability of Analytical methods, which is just another way of saying “how much benefit can a client derive from a project given a specific agent, depending on the size of the client and the size of the optimization benefit”.
The more I thought about scalability, the more I realized that analytical approaches themselves can scale, and the nature of the client-agent relationship could scale accordingly.
The most linear way of thinking about the scalability of analytics would run like this:
Client A has a landing page that has N visits at a conversion rate of X. By hiring an agent, conversion increases to X + z, resulting in Y more conversion, and P more profit. The price scale that said Client A can afford to pay the Agent is thus somewhere between the minimum that the Agent is willing to accept, G, and what Client A is willing to pay, E, so long as E>G.
In calculating that out, you might estimate that the minimum volume of visits that a client needs before engaging an Analytics Agent would be some fairly simply equation, say, 60000 visits per duration of the range of N, and you’re done.
In other words, there would be a vast segment of the business base that would have no effective incentive to hire an Agent, at least based on a semi-valid attribution model.
I think this line of logic is dangerous, because it doesn’t account for other types of client-agent relationships.
At the highest level, ‘in-sourcing’, you have a client that hires an agent full time and counts them as headcount. In effect, the agent is selling 1920 to 2000 labor hours a year (at least in Canada) in exchange for a salary, benefits, taxes and, back in the olden days, pension benefits. We don’t traditionally think of hired labour as being this kind of client-agent relationship, but I think it is. It’s a different kind of one.
Then you have this size and scale of mass where the client ‘outsources’ the work to an agent. The client pays the agent for a fixed number of hours, or a piece of work for a fixed fee. This is where concepts like risk estimation comes into play – and has been the subject of the bulk of these posts.
Beneath that – what?
Then you have people who train business people and entrepreneurs how to do analytics themselves. It’s the whole ‘teach a person to fish’ kind of a thing – and I suppose this could be treated like a one-time social capital cost. There are limits to what people can do without base skills and training – anybody who argues that there are no limits are simply unaware of powerful, actionable, techniques.
Beneath that, you have people who write books, and you have people who buy the books and learn how to do analytics themselves.
And beneath that, you have people who write about analytics on blogs.
As such, there really isn’t a ‘minimal scale’ when it comes to the application of analytics. A person with a small website, taking in 100 visitors a month can learn how to optimize their websites to a point. As that small website grows into a big one, the sophistication of analytics that is required scales as well – in part because even getting the basics of optimization right is much harder, and because, presumably but almost never, all the easiest optimization methods have been implemented. At higher scales it makes sense to take on an agent that can efficiently and effectively handle those types of sophisticated techniques.
There are, rather, thresholds where different client-agent relationships make the most sense.
While I think that many experienced web analytics analysts have a solid idea of where these thresholds lie, I couldn’t find a clearly defined list.
I think such a threshold list would be worthwhile and would be subject to change based on the progression of both social technology and physical technology.
The past 3 posts attempted to break down what I initially perceived as a wicked problem into core components and relationships.
The cost of Analytics suffers from information and trust asymetries – which I don’t think that the hourly model necessarily cures – and sometimes incents us to ignore.
The benefit side of Analytics doesn’t necessarily suffer from scalability issues, as the nature of engagement should scale with the perceived benefit.
The Economics of Analytics isn’t completely unique in terms of how the client-agent problems, but there are opportunities in there to fundamentally innovate that model.
If you have comments on any portion of these threads, please do.