There is an evolutionary process in industrial design, and it seems to me that this is relevant to software. The stages are:
1. Design for purpose This is where the objective is simply to make something that works, something that satisfies the business requirements. Design for purpose is a cottage industry at best, and often more like a craft. This is where most in house software development is stuck, and Tim Bray is quite right to call it out in his recent ‘doing IT wrong‘ post.
2. Design for manufacture This is where the objective is to build an efficient and repeatable process that achieves some local optimum for cost/quality. When Ford came up with the Model-T the primary purpose was to make the car easy to make. Operational issues are somebody else’s problem. This is where most packaged software development is stuck – because the manufacturer (the software house) doesn’t bear the cost of operations [in fact many software houses have transformed into services companies because there's money in those hills dealing with the operational issues]. Where operational issues get pushed back into the manufacturing process (by incumbent liabilities like software patching) then we see the emergence of things like Microsoft’s Security Development Lifecycle, which is a step along the road to…
3. Design for maintenance This is where the designer of a product or service has to consider the full operational cycle and costs associated with what they’re doing. They don’t just want to make it cheap to make, but it also has to be cheap to own. Cloud computing forces a shift to this because the everything as a service model is about through life operations rather than one off delivery. Successful SaaS (and PaaS, and IaaS) is fundamentally different to what came before because intentionally or otherwise it has been designed for maintenance. We may talk about better management/automation/scaling, but ultimately these things come about from a different design paradigm.
I thinks this helps to explain why many software houses are struggling with a transformation to SaaS and the Cloud. Design ethos becomes very tightly intertwined into an organisation’s structure and culture, and a shift from design to manufacture to design for maintenance is clearly a hard one to make – just look at what happened to the car industry.
I first came across the concept of a virtual resource market at the Grid Today conference in 2004 when Steve Yatko (who later became my boss) made his keynote presentation on ‘Service Oriented Computing’. Apparently we were not alone in thinking about this convergence of grid computing (as we called it then) and economics, with JP Rangaswami and Sean Park discussing similar concepts back in 2002. A few years later I found myself presenting on the same topic at OGF20, and a short while later Steve, Vlad and I applied for a patent (which still remains pending).
It was only a couple of weeks ago that I was discussing how soon this would come to pass with a friendly cloud startup CEO at IGT2009, and he thought it was some way off. He can join Simon Wardley in the camp that thought this would happen but take longer. I personally was always optimistic that this would happen sooner than later, though I must confess to some recent confusion over Amazon’s capacity management for AWS – just what is ’spare’ capacity, and what happens when Amazon itself is busy (in the run up to Christmas)?
Today saw those questions answered as Amazon announced the introduction of spot instances, which allows Amazon to auction EC2 capacity (in addition to the existing on demand and reserved instance prices). Right now the market is closed[1], and people can’t sell their reserved instances to others, but it’s reasonable to expect that these things will come to pass. I’m not sure at this stage whether the VRM will become a reality first in the Amazon public cloud (and I guess I’m with Simon on thinking that there’s a great opportunity for them there) or in the private data centres of large enterprises (which is the problem we were looking at 5 years or more ago); but as of today we’re a whole lot closer to that vision.
[1] and there’s some speculation about transparency – do different people see different prices?
To protect the guilty I’m not going to name names, but when I saw an M&A deal last week for almost a quarter of a billion dollars for a firm that I consider poor in its space I could hold my peace no more.
The problem seems to be one of timing. New sectors to the technology marketplace open up all the time. Sometimes the early entrants to those sectors are outrageously successful, led by the smartest people on the planet, who figure out straight away the best way to solve a problem. Most of the time though there is a first mover advantage in gaining early market share, but a second mover advantage in doing things right.
Why then does good money go on buying bad tech (when presumably better tech could be bought for less from a startup company that hasn’t grown as large/expensive)? The obvious answer is that valuations come from revenue multiples, and reflect upon market success. But hang on a moment, the acquirers we’re talking about here are industry behemoths - they don’t need to buy their way into customers.
The comparison that I’ve used in the past is the humble internal combustion engine. For the first century that we had such engines fuel got into the combustion chamber using a carburettor. Carbs do a pretty reasonable job of mixing fuel with air, but if you’ve ever done any engine tinkering then you know that they can be somewhat fiddly to set up, and require some degree of ongoing tuning. These days you won’t find a carb on a new car engine, as they have been almost entirely replaced by electronic fuel injection, a system that requires much less set up and tuning as the electronics are self calibrating.
The acquisition that grabbed my attention last week was the equivalent of GM buying Weber just as Bosch were starting to move down market from range topping high performance specials (that would have special ‘injection’ or ‘i’ badges on the back). This is a big mistake, because technology that has a greater setup and admin overhead (like a carb) will also be harder to integrate into an overall product strategy. It might look like a good opportunity for synergy with professional services, but in the long run the market will move to the simpler, better automated product that requires less people to care and feed it.
A wise man once told me ‘there’s no such thing as a successful technology company, there are only successful sales companies, some of them sell technology’. He’s dead right, but my question here is why would successful sales companies prefer to buy a successful sales company with bad technology over a good technology company that hasn’t (yet) done so well with sales?
Introduction:
Our Software and IT Services Demand Barometer is a composite index of several variables that we believe are leading indicators of demand for the European sector. It was first formulated by Benjamin Berenson back in 2001 and members of this team have developed it further to the index you see today. We have used it extensively in supporting our overall sector views over the years and it has served us well.
Our index is a “diffusion” index, calculated in a similar fashion to the US ISM indices. The final result reflects the number of variables that have increased, decreased, or remained unchanged on a month-to-month basis. Hence, a reading above 50 indicates expansion from the previous month, while a reading below 50 indicates decline from the previous month.
The IT Demand Barometer dropped back to a modest decline in IT demand in August, following a brief expansion in July.
Figure 1: The IT Demand Barometer from January 1998 to August 2009
Recent trends:
After falling to an all-time low of 27.3 in January 2009 (significantly worse than the prior trough of 32.3 in November 2001), the reading has revived: The IT Demand Barometer increased from 27.3 in January to 52.7 in July 2009, before dropping back to 44.8 in August.
Though this may be interpreted as a revival in demand for Software and IT services may be near, our experience after the bursting of the tech bubble showed there were a number of false dawns as fiscal policy stimulus delivered temporary respite (November 2001 to April 2002 and September 2002 to January 2003) before the sustained growth period from July 2003 to December 2007.
The recent rise in the IT Demand Barometer is mainly attributable to an improvement in IT staffing data in the UK and an improvement in the general economic conditions. However, US permanent IT staffing data has still not shown signs of improvement.
Constituent component movements:
Micro Indicators:
IT Staffing levels:
The US technology industry cycle is generally a few months ahead of Europe and any major trends in demand in the US technology sector are likely to be witnessed in Europe in the next few months, so it is important to look at data from the US technology sector for lead indications. For this reason, we look at IT staffing data in the US and UK (UK being the one of the largest IT services market in Europe and because relevant data for the UK is available in greatest detail).
The US IT staff and temporary help services data are taken from the US Bureau of Labour Statistics data (computer systems design and related) and modified to indices that reflect expansion (increase in employment) or contraction (decrease in employment). UK IT Staff data is available in the form of a diffusion index (an index that indicates expansion or contraction depending on whether it is above or below a steady-state level) published by Markit.
Figure 2 shows that while the Permanent and temporary IT staff indices appear to move together, the temporary index seems to slightly lead the permanent index. This can be explained by the fact that companies would prefer to lay off temporary staff ahead of its permanent staff. There also appears to be some correlation between US and UK IT staffing levels. But in the US, it is clear that the temporary staffing is much more volatile and any changes in staffing are likely to have the first impact here.
From this we can conclude that the temporary IT staffing data is a key lead indicator of IT demand, and these indicators forms a key component of the IT Demand Barometer.
Figure 2: Micro Indicators: Permanent IT Staff and Temporary Staff Diffusion indices for the US and UK (3 month moving averages)
After the steep falls during the 2008 recession, US temp help services staff growth seems to have reached a trough in Jan 2009 with permanent and temporary UK IT staff showing similar troughing in March. Recent trends show an improvement in staffing data, particularly in the UK, though not yet back to expansionary levels. US temporary staffing data, however, has still not shown signs of improvement while US permanent staffing data has been flat for the last few months.
Macro Indicators:
EU Corporate Profits growth forecast:
Generally, it is reasonable to expect that a firm’s IT investment plans will depend to an extent on its profits and profit outlook. A loss making firm is highly likely to cut back on its IT budget, while a profitable firm would look more favourably on investment in IT to support growth. We look at 12 month forward earnings forecast on an industry wide basis to get an indication of corporate profitability expectations. Our experience suggests that bottom -up aggregate forecasts, such as these, tend to have less lead over the real economy performance, but remain the best available measure of explicit profit expectations.
As we can see in Figure 3, during periods of recession, corporate profit forecasts have shown a drop – no surprises here. But we are particularly interested in this as an expectations measure and are focused on the changes in expectation.
Corporate Profit forecasts have bounced back after being negative from April to July 2009.
Figure 3: Macro Indicators: 12m forward EPS growth forecast for FTSE Europe (Source: Datastream)
EU Economic Sentiment indicator:
The economic sentiment indicator published by the EU commission, is a composite indicator that accounts for industrial confidence, services confidence, consumer confidence, construction confidence and retail confidence, thereby giving an overall indication of the sentiment in the economy. The EU economic sentiment indicator could give a useful indication of broad investment trends of which Software and IT Services will be part.
As seen from Figure 4, the most recent data shows a likely bottoming following an extremely steep fall.
Figure 4: EU Economic Sentiment Indicator (Source: EU Commission)
Appendix:
Key inputs include:
Changes in temporary IT staffing on a monthly basis. We believe this is an excellent proxy for the marginal demand for IT services projects – e.g. if the use of IT contractors trends up it suggests a more robust environment.
Changes in permanent IT staffing. Interviews with IT services providers and in-house IT teams suggests permanent staffing levels have greater latency vs. changes in the demand environment, but carry greater weight in assessing medium term as opposed to short term demand trends.
Changes in overall macro-economic leading indicators that are correlated with IT spending. Most of our analysis suggests that business confidence is the prime determinant of willingness to spend on IT; therefore we incorporate factors such as EU business confidence surveys, expected corporate profitability, and other leading indicators (though they are weighted less than our staffing variables).
Interpreting the results:
Our index is a “diffusion” index, calculated in a similar fashion to the US ISM indices. The final result reflects the number of variables that have increased, decreased, or remained unchanged on a month-to-month basis. Hence, a reading above 50 indicates expansion from the previous month, while a reading below 50 indicates decline from the previous month.
Predictive power:
We have designed the IT Demand Barometer to inject some quantitative rigour into assessing the very subjective issue of European Software and IT services demand. While it is not a substitute for channel checks with CIOs/IT purchasers or other anecdotal sources, we believe it provides a useful complement.
Assessing the predictive power of the index is difficult without an objective measure of IT services demand for historical “back-testing”. Measured against major policy and market turning points it shows good results.
While the Barometer is not designed to predict the performance of stock prices in the Software and IT services sectors, it does show good results over time especially at turning points, but sector volatility can be distracting.
Constituents:
Micro indicators:
US IT Staff Change Index (source: US BLS Data)
US Temporary help services Change Index (source: US BLS Data)
UK Permanent IT Staff Diffusion Index (source: Markit survey)
UK Temporary IT Staff Diffusion Index (source: Markit survey)
Macro indicators:
EU Corporate Profits growth forecast (source: Datastream, FTSE Europe 12m EPS growth forecast)
EU Economic sentiment indicator (source: EU Commission)
Earlier this year I characterised platform as a service as the filling in my ‘cloudburger’:
This isn’t meant to infer that PaaS is the tastiest piece of the cloud, though the recent acquisition of SpringSource by VMWare for $420m would seem to suggest this (well done Rod and team). What I was in fact trying to show with this illustration is that PaaS is a thinner piece of the overall value stack than either the underlying infrastructure as a service (IaaS) where the value comes mostly from the capital assets in the data center, or the Software as a Service (SaaS) that’s built on top – delivering the actual business functions.
My model seems to tally much better with the more modest price of $28m paid by Tibco for the acquisition of DataSynapse. So why the huge gap? Why is DataSynapse a salami PaaS, whilst SpringSource is a juicy fat gourmet burger? Here are some thoughts on what happened:
I started working with DataSynapse around the back end of 2003 just as ‘grid’ and ‘utility computing’ were becoming white hot in financial services IT. At the time it seemed like DataSynapse and Platform Computing would be the next big tech IPOs, but things didn’t work out that way. Grid struggled to escape from the world of High Performance Computing (HPC), despite all the promise that it held to revolutionise development and infrastructure management. I think part of the problem was around flexibility – grid platforms like DataSynapse had too much of it. They didn’t say to the developer ‘this is the right way to do things, and we’re going to help you get there’, but rather ‘you can run anything on this platform’, ‘you can bring out your smelly old legacy code, it will still work’; and by and large that’s exactly what happened – smelly old code got lifted off cobbled together home brew HPC platforms and dropped onto the shiny new grid middleware. The result of this is that there wasn’t anything compelling for developers to get their teeth into. There was also precious little engagement with the developer community on places like TheServerSide.com (which still mattered back then).
I do recall one app where the dev team were persuaded to use the grid rather than building a dedicated infrastructure. They saved months of development time by making use of the PaaS qualities of the grid, and also didn’t have to bother themselves with the usual hassles of buying hardware. I’m still somewhat stunned that more developers weren’t pushed in that direction (especially given the noise about utility computing that my CIO was making at the time), but there was a huge gap between rhetoric and governance.
So… grid failed to escape the HPC gravitation field because it failed to attract developers. Which is exactly what didn’t happen with Spring. Spring began it’s life in the full glare of TheServerSide, it was the community that made it what it became. It was, and remains, compelling to developers because it leads them to the right way to do things, which is also the simple way to do things. Adrian’s work on folding aspect oriented programming (AOP) under the hood of Spring was pure genius (I remain impressed to this day by a presentation he once gave where there were powerful code examples in a large readable font on PowerPoint slides). Open Source was a key factor in Spring’s popularity and success, though I feel that there were times when this didn’t guarantee commercial success. Rod has indeed done an awesome job of turning a consulting firm into a VC backed Open Source firm into the PaaS piece of the worlds most successful virtualisation company. I’m probably not alone in wondering where the rabbit came from when it got pulled out of the hat. Of course it remains to be seen whether VMW get their money’s worth from the deal, but at less than 4% of their present market capitalisation perhaps the PaaS piece is still a thin filling in their burger?
There are more cloudburgers to be made here. Obviously Microsoft’s Azure is going to have a significant impact, especially for all those with a .Net leaning, and makes them an unlikely acquirer of anything else in this space. Citrix now surely has to make a move to get some kind of PaaS story. DataSynapse’s old competitor Platform Computing is still independent and has a reasonably strong customer base. It’s also worth keeping an eye on Paremus, as their service fabric arguably beats SpringDM at its own game.
I was recently asked by a VC what our ‘house view’ was on when M&A markets will return to normal. I didn’t have a convenient canned answer for him, maybe one day something like that will appear on this blog – though don’t hold your breath – if normal is defined by the last ten years (and two burst bubbles) then we’ve got a long way to go.
What I did provide an opinion on is the state of the IT security industry. This may be a somewhat controversial position, but I think things are going to flip over the next few years. Here’s why…
IT security is Pareto inverted. Consumers of IT security products (mainly enterprises) are spending 80% of their budget on network based security, and the remaining 20% on other things like host security, data security and application security (aka secure software development). The trouble is that network security is only 20% of the problem:
I can’t believe that organisations will continue to allocate their budgets so disproportionately, and there’s really no way that the budgets are going to rise dramatically, so there’s clearly some large (and uncomfortable) change coming down the pipe. The losers will clearly be the incumbent network (and network based) security solutions, and and vendors that are too closely tied to that model. The winners (if they can be called that for getting a larger slice of the same sized pie) will be those that provide more data centric approaches. That’s why we’ve seen an explosion in the data leak prevention (DLP) market over the last few years, with each of the large security vendors placing a big bet on the table. This is however the end of the beginning DLP for unstructured data suffers the same (or worse) than enterprise rights management (ERM) in that it’s not fit for purpose. The other trouble is that controls over unstructured data as it moves around web, email and file server mediums doesn’t address all of the stuff that’s wrapped up inside applications. It’s easy to get distracted here and throw a bunch of controls around structured data in relational databases, and there has been a burgeoning market for enhanced monitoring and segregation of duties solutions there, but once again the data store is just part of the problem – with the applications themselves being where the real trouble lies.
So where should the next bets be placed? Secure software development practices, and the tools that support it, have been slow to take off in enterprise environments; though in the last few years this stuff has become like religion in some of the larger software houses. The move from packaged software to SaaS also changes the game – in terms of where the problem lies and where the solution comes from… Now where can I find a SaaS based service that will tell me how secure my SaaS based app is?
