A San Francisco company called Feeva has launched a new ad service it says targets Internet users with unprecedented geographic and demographic precision.

Advertisers have long lusted after ways to target users — they can sell more if they reach the types of people likely to buy their products.

One of the sexiest sources of such data — the Internet browsing behaviors of individual users obtained directly from Internet service providers (ISPs) — has been off-limits for privacy reasons. However, Feeva says it has found a way to harvest geographic and other information directly from these service providers — but in a way that doesn’t identify individual users.

With this kind of service, Feeva wants to rake in a portion of the multi-billion dollar online advertising market, which so far has been dominated by search engine giants such as Google, Yahoo and Microsoft.

Feeva is distancing itself from other players. Controversy erupted around the early efforts of two companies, Phorm and NebuAd, to access such data from Internet service providers. Their actions brought a storm of protest from privacy advocates, who say the data on browsing patterns of users sitting at their PCs is dangerous because it personally identifies people and violates their privacy. Phorm’s plans to track users of large UK service provider BT has brought scrutiny from European regulators. Silicon Valley-based NebuAd was sued in November after it tested its software with service providers. Congress investigated the privacy implications, and the company has backed off its plans — it’s now liquidating its assets.

Feeva, meanwhile, says it is safe from such criticism because it doesn’t access personal information. Through relationships with service providers, Feeva merely obtains the precise location (often detailed to the Zip code plus 4) of users as they request pages from the Web. Feeva does so by placing tags on Web page requests (called HTTP header requests) as they pass through an Internet router. Thus, if, say, a McDonalds in the area has a promotion, it can use Feeva data to target users in the immediate region with coupons for a discounted meal.

Today, Feeva will announce a partnership with Internet router giant Cisco to start distributing its service.

The company, which also has offices in Arkansas and North Carolina, has been working secretly on the technology for nine years. The founders have a lot of experience in the industry: Jaz Banga (CTO), the technical guy behind it all, is a surfer with a background in WiFi and cable. Nitn Shah (CEO), a 25 year veteran of the industry, started his career at Cambridge/Bell Labs, and served as an advisor to the FCC. The two have financed the operations from their own money, some angel money and then a venture capital round in mid 2007 of undisclosed amount.

True, advertisers can target users in other ways, but they’re unreliable. If a user signs into a service like Facebook or a Google account, and provides personal information, these services can serve very targeted ads. However, people often use the Internet without signing into such services. Services such as Quova try to get data by tracking a user’s IP address, but that method is far from precise — the corresponding region can be as wide as 20 to 50 miles. Another service, Skyhook, tracks user locations in relation to WiFi spots (less reliable, because WiFi nodes aren’t consistently located). Hitwise, another company that accesses ISP data, gets ISP log data, but is static — unlike Feeva’s realtime data.

Once Feeva has the Zip code, it gets interesting. Feeva can match it with other information it knows about people living there (see diagram above, and below). For example, Feeva can check things like housing, census and other data compiled by third-party data services, and know that one area of town has say, people who are older with higher income levels than people residing in another area of town.

Advertisers can use the tag to can call upon this demographic information. That way, they can send one ad (for Rolex watches) to users in higher-income areas, and another ad (say, for free banking) to the other. In all, Feeva says it can categorize user regions according to about 60 to 100 characteristics, giving advertisers a much more granular way to target people.

By contrast, the service of Phorm in the UK, slated to begin later this year, uses its relationship with BT to insert a cookie onto people’s browsers. Despite the criticism it has received, Phorm says its method doesn’t actually identify individuals either; instead, it assigns a random number to the cookies browser, thus avoiding identifying a person’s name with the cookie. Phorm says that once it matches the random number with a user’s interests, it deletes any record of Web sites a user visited. However, since Phorm is tracking user behavior, it’ll continually have to convince the public it’s not keeping their records.

Feeva’s business model is straightforward: If an advertiser (or ad agency representing it) pays Feeva, it gets to access the information on the Feeva tag. Feeva distributes a cut of the revenue it gets from advertisers to the service providers.

Shah said he has solicited feedback from privacy experts, and they’re comfortable with Feeva’s methods.

Still, Feeva has its work cut out for it. It’s dependent on ad networks like DoubleClick (and its competitors — AOL’s Platform A, WPP’s 24/7 Real Media, Microsoft and Yahoo’s Right Media) to educate big advertising agencies so that they’re comfortable planning campaigns with Feeva data.

Shah says mobile carriers have also expressed interest in the product now that smartphones like the iPhone are letting users browse the Internet more frequently.

The Feeva-Cisco product will be demonstrated at the Mobile World Congress in Barcelona next week.