What inspires me to bring this up now is the press around Microsoft Sharepoint. Apparently, there’s long been a SaaS version of Sharepoint for big firms, and now Microsoft is rolling it out for everybody. Now, I haven’t read the press releases, which weren’t sent to me by anybody at Waggener-Edstrom and are not easy to find on Microsoft’s web site. But the reporting doesn’t seem to mention partners, except in the negative. I.e., this seems like yet another significant direct-sales SaaS business.

If you follow this logic through, it suggests that a large part of the SaaS market will wind going to large companies with global reach — whether or not the rumors are true that Salesforce.com is currently being shopped around.

1. I need a mail host that can stand up under the kind of mailbomb/DDOS attacks that shut me down twice in the past year.
2. Similarly, I want to diversify my email addresses among two providers, rather than leaving them all with my general web hosting company.
3. David Ferris first wrote up Google Mail outsourcing, with a favorable view, last July. And some of his criticisms (e.g., lack of IMAP support) have already been rectified.
4. What’s more — as I remarked last night, David and his associate Richi Jennings have been voting with their feet, and moving their own email to Google. That’s an impressive endorsement. Ferris Research is a serious rival to Gartner as an analyst firm covering email, and Richi — who evidently LOVES Gmail — has also carved out a non-trivial identity as an expert in his own right.
5. Free sounds good, compared with the alternatives.

Now that I’ve gone ahead with the move to Google Mail, here are some scattered thoughts:

• Some terminology: Technically, Google Mail is part of the Google Apps service. And the terms “Google Mail” and “Gmail” are pretty interchangeable (you even have two choices of server name when setting up POP3 access).
• Google’s UI to get started can be a bit confusing. But googling on Google Apps will get you to the right place, namely this link.
• Particularly confusing is dealing with the MX records. My domain registrar didn’t seem to offer a way to redirect them at all. The cpanel interface for redirecting MX at my hosting company wasn’t very good; Google asks for about 7 entries of declining priority, but cpanel only makes it easy to enter 1. (I wound up asking my hosting company’s support to make sure all the entries were listed that should be.)
• If there’s a way to sort messages by subject or sender in Gmail online, I haven’t found it. That’s a major oversight if it can’t be done, or a minor one if it’s merely too hard to figure out how to do it.
• Setting up POP3 access has some steps that aren’t present in, say, setting up POP3 via a typical hosting company. You need to go into “Settings” and explicitly enable POP3 access. You also need to explicitly enable SSL in your mail client (on Eudora, the default setting did NOT work). Basically, you need to open this page or something similar, and actually look at the steps for your client. Chances are one or two will be non-obvious.

That’s about it for now. In particular, I haven’t done anything yet with Google Mail’s search capabilities. More on that down the road, perhaps.

We market our service to businesses on a subscription basis, primarily through our direct sales efforts and also indirectly through partners.

Looking back, I should have quoted that in support when I wrote:

By the way, I think the assumption SAP needs to sell ByDemand via indirect channels is an erroneous one. (Dennis Howlett seems to be at least partway to recognizing this. He also reports that SAP realizes that this is truly a sales issue.) Hence my stress on SAP’s internal sales management issues.

For 40+ years, application-oriented services have been sold in large part by direct sales forces. That goes back to the other payroll processors, and to time-sharing in general. Why would it change now?

1. SAP sure has a lot of resources to bring to bear – and as previously noted, I think the company is dead serious about this initiative.

2. On the back end, the business-service granularity SAP has been implementing is well-suited to deal with the unique challenges of SaaS, both the very real (e.g., short upgrade cycles) and the largely imaginary (e.g., multi-tenancy).

3. SAP recently hired Dan Rosenberg away from Oracle to head its UI efforts, and Release 1 of a Dan Rosenberg user interface is likely to be very good. I know Dennis Howlett has a contrary view, and he’s actually seen the product. Even so, I’m optimistic about SAP’s claims to have designed the UI with an open mind, for maximum ease and simplicity, and validated by many rounds of testing.

As for sales and marketing – that’s a solvable problem. Indeed, SaaS and high-end packaged software can be synergistic businesses. True, a ByDemand salesman will need to earn much lower compensation than a high-end software sales team leader. But that’s a management challenge companies like SAP already know how to meet. Inside salespeople, product specialists, and so on can already tend to be compensated a lot less than the classic account-manager types are.

By the way, I think the assumption SAP needs to sell ByDemand via indirect channels is an erroneous one. (Dennis Howlett seems to be at least partway to recognizing this. He also reports that SAP realizes that this is truly a sales issue.) Hence my stress on SAP’s internal sales management issues.

All this assumes, of course, that SAP’s ethics and organization aren’t as screwed up as I sometimes fear.

Keep getting great research about enterprise applications, analytics and related technologies. Get a FREE subscription by RSS/Atom or e-mail!

Technorati Tags: SAP, ByDemand, SaaS, software as a service, Peter Zencke, Dan Rosenberg

Salesforce.com is the biggest SaaS company. Salesforce.com is making next to no profit. Therefore, SaaS is currently not a profitable business.

But that’s nonsense. Here’s why.

If you look at Salesforce.com’s second quarter 10K, 6-month revenues were $339 million, up from $223 million the year before. Marketing and sales costs were $174 million, slightly over 50% of revenue. Profits were negligible. The company says the churn rate is negligible, so that 50% of revenue was spent increasing sales by 50%.

Now let’s suppose that the SaaS (Software as a Service) industry becomes more mature. 1% would no longer be a realistic churn rate. Let’s suppose it instead goes to 10%+, based on both true replacements and client disappearances. Let’s suppose the revenue growth rate settles down to somewhere in the teens. Bam. To a first approximation we can whack marketing and sales by a factor of 2, and take pretax margins well over 20%.

Of course, things aren’t really that simple. It’s also necessary to market and sell for customer retention. And true marketing cost (as opposed to sales) isn’t closely tied to the number of opportunities you have. But on the other hand, as you grow there are all kinds of economies of scale too. So to a second approximation, 25 – 40% pretax margins don’t seem unrealistic.

So how profitable can a recurring-revenue computer services company be? Two examples come to mind quickly – Automatic Data Processing and Paychex. ADP’s main business and Paychex’s whole business is services in the area of payroll processing and human resources. Paychex reported 39% pretax margins on its latest 10K, while ADP’s figure was in the 25% range. (Incidentally, it seems that ADP is a partner for ByDemand.)

ADP and Paychex actually tend to get much less revenue per client than, say, SAP’s ByDemand “please don’t call it ERP” system would, yet they manage to generate substantial levels of profitability. So even if we didn’t have the hard evidence from Salesforce.com cited above, it would be a good bet that nicely profitable business models exist for SaaS offerings sold to smaller clients.

Keep getting great research about enterprise applications, analytics and related technologies. Get a FREE subscription by RSS/Atom or e-mail!

Technorati Tags: SaaS, Software as a Service, SAP, ByDemand, Paychex, Automatic Data Processing, Salesforce.com

]]> http://www.monashreport.com/2007/09/25/the-unprofitability-of-the-saas-business-is-an-illusion-caused-by-growth/feed/ 0 http://www.monashreport.com/2007/06/09/technology-of-guild-wars/ http://www.monashreport.com/2007/06/09/technology-of-guild-wars/#comments Sun, 10 Jun 2007 02:00:23 +0000 Curt Monash http://www.monashreport.com/2007/06/09/technology-of-guild-wars/ Being an analyst has its perks, the main one being that you get to have some really interesting conversations. And so I recently had the chance to interview Mike O’Brien and Pat Wyatt, two of the founders and lead programmers for ArenaNet, makers of the Guild Wars MMORPG (Massively MultiPlayer Online Role-Playing Game).

If you play games of this sort, it’s surely obvious to you why you should care. But if you don’t, maybe you should be interested anyway. After all, Guild Wars is a graphics-intensive SaaS offering that easily supports 100,000 simultaneous users, while managing a gig or so of fat client even over dial-up speeds. Every user is a potential hacker, whether for fun or actual real-world cash profit, although we didn’t actually talk about security very much. And ArenaNet provides all this on a relatively shoestring budget; in particular, Guild Wars subscription fees are precisely $0.

Note: If you’re familiar with video RPGs but not MMORPGs, please just read on; the main salient difference is that in an MMORPG lots of people play at once. If you’re familiar with other MMORPGs but not Guild Wars, there’s one GW-specific design feature you should know of – Guild Wars is instanced. That is, a group of 1-24 players will get their own private copy of the game world. Any changes to the environment (e.g. player movement, monster-killing) are made only by them.

And if you’re not familiar with video RPGs at all, I suggest you stop working quite so hard and go play.

Here are some of the technical highlights I managed to glean in a brief half-hour conversation:

• The client doesn’t do much except graphical presentation and input preparation (including encryption). This is a deliberate security choice. They assume the game client is infinitely hackable, and hence limit its power to sending a small number of well-defined messages.

• Their servers run on blades – originally all IBM, now more varied. Generally these are 2.8 gigahertz dual-processor Xeon IBM blades with 2.5 gigs of RAM.

• There are a number of different categories of servers. These include: Authorization/log-in, game update download, actual gameplay, database cache, DBMS, several secondary game functions (e.g. guild membership, tournaments), and watchdogs overseeing the rest.

• Gameplay servers can support 2500-3500 users each, with the main limitation being addressable memory. (They run on 32-bit Microsoft Windows.) 3500 users is a bit uncomfortable. This matches well against the fact that character records are several 10s of kilobytes, for reasons I explain in this companion post on Guild Wars data management.

• There are 4 ½ million total lines of code.

• Their connectivity design goal is .0045 megabits/sec/user, and they’re close to achieving it. Dial-up is a fairly realistic way to play. (That said, as a player I can tell you it creates some lags that make a dial-up user less than idea to be partnered with.)

• Reliability is not just a nice thing for users, but an actual game integrity concern. There are possible exploits in MMORPGs that involve deliberately crashing a server when a goods-trade transaction is partway through. ArenaNet fights this both by focusing on transaction integrity (but not in the way you may think; I refer you again to the database-focused companion post) and by generally trying to make their database cache servers extremely reliable. Database cache servers typically stay up 150-200+ days at a time.

*Note: There is a high level of motivation for cheating in MMORPGs. One of the biggest is that professionals in low-wage but well-wired countries acquire fictional gold, so that they can sell it for real money. Gold sellers are a huge concern to ArenaNet.

That’s just some of what I heard. For more on the technology of Guild Wars, please see my companion post on the database technology of Guild Wars and Guild Wars game notes. This interview is also just one of several I plan with game and virtual world companies, so I hope to in the future address other kinds of issues, such as rules and AI.

• Crossbeam has two main hardware platforms – the high-end X-series and the midrange C-series. The X-series is the one with the architecture I previously praised, and about which Paul himself is “really excited.” The less remarkable C-series, however, is the one Check Point’s UTM-1 products are actually based on.
• There are three UTM-1 models. Two of them use hardware that exactly duplicates Crossbeam’s C2 and C6. The most powerful of the three – the 2050 – is based on a modified C6 design. Paul isn’t 100% sure in his recollection of what the modification was, but thinks it’s probably extra RAM.
• The hardware is actually manufactured by an unnamed Asian outfit. Crossbeam currently buys the boxes and resells them to Check Point. It is anticipated that this will change over time, and Check Point will take care of procuring its own boxes (from the same manufacturer). At least, that’s the plan if the Check Point and Crossbeam hardware specs significantly diverge.
• The Crossbeam C-Series — and hence also the new Check Point UTM-1 – are indeed classic Type 1 appliances. The biggest difference vs. generic Dell/HP/whatever servers is the density of Ethernet ports (4-8 per box, depending on model). In particular, Check Point is very proud of the work it’s done optimizing for Intel processors.*
• Notwithstanding anything above, the UTM-1 machines really are Check Point appliances. Check Point does 100% of the support, it has some administrative software pieces that are different from Crossbeam’s, etc.

*Indeed, as the focus of security processing shifts more and more to the application layer, they contend security processing is more and more like any other kind – rather than, say, low-level network processing.

What seems to be going on here is that Check Point is cannibalizing Crossbeam’s C-Series business, and Crossbeam is being gracious about giving it up while focusing on the much more differentiated and strategic X-Series. Crossbeam self-identifies as a high-end player anyway, so this all makes perfect sense. The real issue for Crossbeam going forward has little to do with whether it can squeeze a few more commodity dollars out of the midrange. Rather, it’s whether Crossbeam can hold its technical lead when the large server manufacturers finally figure out the need to create virtualization-friendly, networking-friendly, blade-based systems. The key point here is “networking-friendly”; many servers just need more data movement capability than conventional systems now provide.

Or to put it another way: The computer is a network.

• Virtualization has overhead, but probably a lot less than the 43-50% sometimes claimed.
• Just to be safe, don’t virtualize apps that are already I/O-bound or otherwise running flat-out. (So there’s no contradiction to my support for dedicated security, networking, and data warehouse appliances.)
• Big enterprises have lots of production servers that are old, unreliable, and/or idle most of the time. Virtualize those.
• If a server’s use is particularly spiky, it may be a great candidate for virtualization.
• Most development servers can and should be virtualized.

Makes sense to me.

• Business Objects and Netezza have announced a mid-range BI appliance.
• Ingres is headed in the same direction.
• QlikTech is enjoying great growth for its fast-deploying BI technology.
• KXEN and Verix offer “easy” data mining technology.
• Search-based BI is trying to circumvent the data warehouse deployment process.

It’s about time.

However, that turns out to be wrong in two interesting ways. First, it was slightly incorrect all along; Check Point’s “Edge” product line has been Type 1 for almost five years. Second and more important, a few weeks ago Check Point announced that it was finally entering the Type 1 appliance mainstream market itself.

“Edge” products almost don’t count; they’re limited-functionality perimeter devices that only work well if managed and strengthened by bigger Check Point deployments at remote locations. But the new Check Point UTM-1 appliances absolutely compete head-on with Juniper (nee’ Netscreen) and other mainstream firewall – excuse me, “unified threat management” — appliance vendors.

Check Point says its motivation for introducing physical appliances is their ease-of-deployment benefit. That sounds right to me. The primary other alternative would be performance, and that’s not what’s going on. While custom-manufactured, these Intel-based boxes seem to be quite generic, with the biggest non-standard aspect being the number of high-speed Ethernet ports. (Configurations vary among three models, targeted at 100- to 1000-user installations.)

So how does this affect Nokia and Crossbeam, Check Point’s two most important hardware partners? At the moment, it hardly affects them at all; they sell to a higher-end market than is served by these new appliances. Longer-term, it’s harder to say.

Technically, Crossbeam isn’t dependent on Check Point at all. But in practice, Crossbeam would be in a world of hurt should Check Point decide to compete directly. On the other hand, it’s not at all obvious that there are enough hardware margins to make it worthwhile for Check Point to turn against its partners. As for software – well, “unified threat management” is impressive in either its unification or its management. There’s a ways to go before we’ll know whether Crossbeam’s “best of breed” software mix will outdo Check Point’s “unified”-but-actually-involving-multiple-partners own umbrella offering.