Article Info

San Francisco MuniFi Economics

San Francisco gave the green light to Google (GOOG) and Earthlink (ELNK) to offer wi-fi throughout the city. You’ll get free 300kb/s ad-supported service from Google, or $20 a month for 1Mb/s ISP service from Earthlink.

Total capex is $15MM to install, though I bet that number will at least double to deliver full coverage. Assuming Earthlink spends $15MM on it’s own, they would need around 8,000 subscribers at $250/year to justify the investment and deliver a 15% return. This seems very, very achievable in a city like San Francisco. If Earthlink can win peering agreements with guys like T-mobile to allow roaming outside of SFO, it looks even better.

San Francisco is always on the vanguard of both good and bad.

Correction I neglected to include any provision for Opex in the above calc. Anyone care to take a guess at what that might be?


Comments are disallowed for this post.

  1. Minimum staffing would easily cover that, say 30 people average 50k =1.5m
    Triple it for overhead and equipment maintenance = 4.5m. Say they need 30k subscribers.

    But don’t ignore the advertising. Say each customer sees 200 ad pages per day at $5cpm (both very reasonable numbers if its their default screen). e.g. 1-2 clicks a day at 50cents to 1 dollar per click. Thats easily enough to cover $20 per user per month on the free advertising sponsored service.

    So can they get 30k users for free? Easily.

    Yellow Pages directory is a huge book, with mostly free listings, high printing costs, requiring a lot of staff to maintain it, delivered to the door free to everyone and the local advertising (the large ads) pays for it all. Here they have much lower overhead and automated systems so far fewer staff, so they’ll find it trivial to profit on that.

    Posted by njjf | April 6, 2006, 12:51 PM
  2. This is a subject I have no small amount of experience with, so let me take a crack:

    For each 1000 users, you need about 6 (overworked) people, 2 for support (40k each), 1 for administration (this could be outsourced, though – about 25k maybe?), and 2 field crew types (40k each) to replace physical hardware and that sort of thing, as well as 1 network administrator type (75k in SF). That’s about 260k or so per 1000 users, who are generating 250k of revenue each year in terms of subscriptions. If you add advertising to the mix, the numbers would look a lot different.

    This venture has some pretty hard problems. First, your employee cost is pretty high, relative to the wireless industry at large. Do this in Milwaukee, and you’re spending 30-40 percent less on staff. More importantly, you’ve got an extremely varied terrain that makes ubiquitous coverage extremely difficult at 2.4 ghz. People are going to expect 100 percent coverage, or something very close to what their cell phones provide. Count the number of towers in SF from a single provider, and multiply by at least 8 to get roughly the same coverage.

    Sprint lets you see their cellular towers, so this isn’t all that hard to see. AND, they have NO interference. Intel lists over 8000 (public) hotspots in San Francisco. I don’t even want to know how many private AP’s there are in that city. The end result is that you have to overpower your distribution to communicate over that horrendous noise floor. Meaning you need more AP’s at a higher cost.

    As a private venture, I can see ways this venture makes sense, the same way EV-DO makes sense. But, I think it’s going to be a huge money drain.

    Just my 2.5 cents!

    Posted by Jason Powell | April 6, 2006, 4:15 PM
  3. Jason,

    How many users do you estimate per hotspot? and how many hotspots?

    Surely the situation is like one of my websites, I have about 2-3000 users daily on a typical site, it gets 1 or 2 technical problems a year. They buy things, but its a purely electronic transaction and the computer handles it.

    So admin costs are zero, and free Wifi, or website purchased wifi would be the same. Automated by computer and nearly zero cost.

    Hardware just doesn’t fail so often. Say 20 hotspots per thousand customers, MTBF of 20,000 hours, thats 1 failure every 41 days.

    I also don’t see why 1 network admin per 1000 customers, since the system would be completely automatic its just not needed.

    Posted by efwef | April 6, 2006, 4:54 PM
  4. Technically, a single 802.11b AP with 1 meg plans can support around a real world maximum of about 50 or so concurrent users, with 2 or 3 doing something at any given moment. There are some methods of tuning you can use to maximize for more users, but at the expense of bandwidth. I’ve never worked with the Tropos system Google want to use, but 802.11b is a pretty well known standard at this point, and I wouldn’t expect a whole lot of improvements to my estimate. The real question would be how many people are using the network as a fixed wireless method as opposed to a transitional hotspot application.

    I agree with you on the admin costs – I forgot that Earthlink is doing that, and they’ve already paid for all the infrastructure on that end. My mistake.

    Now, with regards to hardware failure, I think you are wrong. You’ve got sensitive electronic gear out in the environment, being exposed to rain, sleet, heat, etc. Not to mention power surges, brownouts, power outages, lightning strikes, car crashes, general human idiocy, and the like. Wireless gear in outdoor environmentscan really only be counted on to last a year or 2. Lots of gear lasts longer (particularly the expensive licensed gear which lasts forever), but the cheaper, mass deployed stuff will fail often. Multiply that by the 1000 or whatever AP’s they’re going to need, and thats a lot of failure. And then you’ve got maintenance costs of going back to each POP every so often and replacing cable and checking the physical stuff. You can skip this, but I certainly wouldn’t.

    As far as network administration, you need folks who are working around the clock. To fully staff a NOC 24/7 requires 7-10 people, which is in line with my 1 per 1000 estimate. I think that Earthlink has all the staff already to do this, so this isn’t a “new” expense, but it is ongoing expense. You can only automate so much. The biggest difficulty here is interference mitigation, which is almost impossible to do automatically without disconnecting each user every time something is encountered. I’m speaking from experience here, trust me.

    I could go on, but suffice it to say, I think that this enterprise is going to be a break even or worse venture until the ad revenue model is worked out.

    Posted by Jason | April 6, 2006, 7:50 PM
  5. Jason –

    I have zero experience in this area… but the numbers sound awful high per 1k people. Network Admins will scale, maybe 1 for the first 1000, a second for the next 5000, a third for the next 5000, etc.

    A support person should be able to take 4 calls an hour. Figure users average 3 calls a year. 10k users is 30000 calls/year. That’s only 7500 hours of support time, or 187 man weeks. That’s four people for 10k users.

    I think the coverage issues are the biggest problem. The key thing will be setting peoples expectations low and giving them a free trial.

    I wonder how Google is setting up the business side of things. Do they commit to cover a fixed portion of capex and opex and get to keep all of the ad revenue? Or does Earthlink front the opex and take a cut of Googles ad revenue? Interesting problem to analyze….

    Posted by Andrew Schmitt | April 6, 2006, 7:52 PM
  6. “A support person should be able to take 4 calls an hour. Figure users average 3 calls a year. 10k users is 30000 calls/year. That’s only 7500 hours of support time, or 187 man weeks. That’s four people for 10k users.”

    Right, but it doesn’t work like that in the real world. Say some drunk guy runs into a pole where there’s a particularly heavily utilized set of AP’s. That’s 100 people calling in at the same time. And you’ve got to be prepared to handle that sort of thing 24/7, which requires a lot of staff because you need 3 shifts + 2 days off for each shift member. Earthlink should have this problem solved – and I’ve certainly never run a wireless network at this large of a scale, so it’s very possible I’m wrong.

    I do agree with you that the key thing is expectation setting. They have a lot of challenges – coverage, interference, maintaining bandwidth in the absence of clear spectrum, supporting users with laptops (and viruses), etc. – which are all going to impact service, probably at the same time. It will be interesting to see how it will shake out.

    Posted by Jason | April 7, 2006, 7:16 AM
  7. I see what you are saying – you need surge capacity.

    The best way is to have a pool of support shared among 30 muni-fi installs, which is exactly what Earthlink is trying to do.

    Thanks for the comments.

    Posted by Andrew Schmitt | April 7, 2006, 8:01 AM
  8. Trackbacks / Pingbacks

  9. IP Democracy | April 6, 2006, 9:34 PM
  10. Philidelphia Municipal WiFi Analysis - Nyquist Capital | April 25, 2006, 9:40 PM