Let's not reinvent the wheel and use a framework that has been developed since the seventies. Anderson's ACT-R framework is based on cognitive theories and models, and can deliver pretty accurate results.

For a more user-friendly tool, take a look at CogTool. You can build or import a prototype UI, create tasks (sets of actions: click this, type that), and let the software evaluate or automatically optimize the interface. Last time I checked they used ACT-R as the cognitive engine, so the results should be accurate.

i think these are still pretty complicated. i could understand russel's link in 2 minutes, but I still do not know how to use ACT-R and cogtool.

Sure his method sounds easy, but are the results actually worth anything? It looks like an over-simplified version of ACT-R, with some made-up cost values. To me, this has no advantages over my gut feeling on usability; it's just that the numbers add some false sense of reliability.

For actual design decisions, I'd much prefer a method that has developed for thirty-some years, is based on actual cognitive models and has been empirically validated.

Besides, it's not that difficult to use. Building and testing your interface in CogTool would take you about an hour the first time, but it is easy to learn. You can even import HTML and record your actions. The results are pretty accurate, and you get automatic interface optimization (e.g. on widget size, location, types) too boot.

So, the author had a nice idea, but it has been done already. (At least for the user testing, the accurate results, the "codification" and the books; I don't know about the consulting company.)

A neat idea, AI rules applied to UI design. But some the suggested costs are a bit harsh. Of course these are just suggestions. I'm looking for this to be applied on our daily UI's and compare them.

actually me too. it seems like a reasonable way to compare UI's