By Mark Ward BBC News Online technology correspondent

Websites are using increasingly sophisticated methods to watch what surfers are doing and to make e-commerce pay.

Many use the basic logs generated by web servers to find out how many visitors they get, how long those surfers stay, the pages they look at and the adverts or graphics that catch their attention.

But some websites are finding that the raw data is not a good guide to user behaviour.

Instead, some are adopting more complex tools to work out whether their efforts to make people stay longer are working and, more importantly, if these surfers are spending more money.

Model users

Mark Wilding, spokesman for UK user analysis firm Intellitracker, said the scrutiny often turned up information that was a real surprise to those running a site.

It's not just about snapshots of users

John Thompson, WhiteCross

Mr Wilding said his own company's work with the Teletext Holidays website revealed that visitors were not looking to book breaks as far ahead as Teletext thought.

This meant that some of the adverts and holiday offers being run on the site were of interest to only a small proportion of visitors.

Many more were simply coming to the Teletext site looking for a quick getaway.

Mr Wilding said Intellitracker discovered the mismatch by analysing which adverts people reacted to, what types of holiday the surfers searched for and at what point these users made a booking or asked for more information.

Tuning the site

Ashley Mealor, research manager at Teletext, said the company could now track visitors from the home page to the final offer page and analyse exactly what people did with the offers.

Teletext Holidays is using visitor behaviour to tune its site

A computer model of the behaviour helped Teletext Holidays work out what the majority of people wanted and pushed the company into providing more of the offers people were likely to buy.

The analysis also revealed the problems people had when searching for particular flights or offers, and subsequent work helped tune results to provide more help to users.

The model and database of behaviour would also be used to test whether changes to the site had their desired effect, said Mr Wilding.

"Teletext can see how the changes impact the behaviour of people using the site," said Mr Wilding.

Simulated shoppers

Many other sites are likely to turn to computer modelling to get a better idea of what works on the web.

John Thompson, marketing director for user statistics firm WhiteCross, said its customers were demanding more in-depth information about what their visitors were up to.

Mr Thompson said many WhiteCross customers were using simulations to find out what effect changes had before they were tried on the live site.

Web firms can see how virtual populations of users react to changes instead of risking alienating loyal visitors by using them as guinea pigs.

"It's not just about snapshots of users," said Mr Thompson, "you need to understand and drive your business through real facts and figures."

Mr Thompson said the modelling techniques were proving very useful for marketing and strategy departments who wanted to plan how best to develop a business.