Tom Graves / Tetradian

Just how much damage is the cult of specialisation doing to enterprise-architecture?

We’ve struggled for years with the problem that the domain-specific specialisation of enterprise-wide IT architecture has portrayed itself ‘as’ enterprise-architecture – causing serious difficulties for anyone who does need to work across a true whole-of-enterprise scope. Yet this same theme came up for me in several other ways this week – and it seems worth taking note of these various cross-currents, because they all seem to point the same way.

One was an article by Microsoft enterprise-architect Nick Malik, on ‘Enjoying the BPM 2010 Conference‘ and discussing business-process management [BPM] and the ‘democratisation’ of business-processes:

The keynote on the first morning, from Phil Gilbert of IBM (formerly of Lombardi) makes the case … that we will see the democratization of business process management and the disintermediation of the “experts.”

The notion of democratization is interesting to me.  I look forward to that possibility.  To be honest, I don’t think we are all that close, but my mind is open.  BPM is a highly specific field, requiring considerable training and experience.  The development of layers of indirection necessary to truly hide that level of complexity is not yet in evidence.  I suspect that the abstractions will be leaky, at least for a long time.  Perhaps with the development of more “plug and play” patterns, we can empower average business people to get value out of working with the tools directly.  Not sure.

In other words, specialisation is being viewed as ‘the solution’, when it’s clear that it’s in fact fast becoming a key contributing factor to the problems faced in current BPM. Following from that I had a great Twitter-discussion with Thierry de Baillon about the usual errors in current BPM paradigms – particularly the dangers of IT-centrism to which Nick seems to allude above:

• tdebaillon: @tetradian @nickmalik Automate them, so we won’t have to deal with them any more…
• tetradian: @tdebaillon re “Automate them”, I presume you’re being sarcastic? (because misplaced automation is a core source of BPM problems…)
• tdebaillon: @tetradian Sure I am BPM would represent a great way to improve businesses… if only humans were machines.
• tetradian: @tdebaillon “Sure I am” (I breathe sigh of relief ) – to me BPM should *include* people – IT-centrism is the problem, not BPM itself
• tdebaillon: @tetradian Should.. yes, but I am afraid that this view is irreconcilable with what processes are (RT @michaelido: @jhagel: “60-80% of headcount within large organizations is focused on handling of exceptions” #e2conf )
• tetradian: @tdebaillon ‘irreconcilable’ is true only for circular-reasoning of BPM=automation=BPM – e.g. see @sig on ‘Barely Repeatable Processes’ http://thingamy.com

(Sigurd Rinde’s ‘Thingamy’ is one of the very few process-automation packages that’s designed around the way the people work, rather than Taylorist assumptions about how to force people to behave like machines… – strongly recommended for anyone involved in BPM, if you don’t already know it.)

The real point here is the role of the generalist, to link all of the specialist domains together – which requires a very different type of skill, broad rather than deep, but able to translate between the separate worlds of each domain, and also to know which specialist to turn to at each point. Without those generalist cross-links, each domain will usually believe that its own small subset ‘is’ the whole of the context – with the dire results that we see so often, for example, in IT-centric ‘enterprise’-architecture.

But likewise in medicine. Both my parents were general-practitioners (‘family doctors’, in US parlance), and the British Medical Journal [BMJ] still arrives on the doorstep here every week. Serendipitously, there’s an opinion-piece by Des Spence in the current issue, asking whether specialisation is leading to bad medicine [cite: BMJ 2010;341:c4903 - most of the article is behind the BMJ's paywall, unfortunately]:

“You’re too good to be a general practitioner,” someone once told me. I never knew whether this statement was a slight or a compliment. …  But generalism is in decline, with the ascent of the specialists. Gone is the widely experienced general physician, and general surgeons are replaced by an ever expanding list of “ologists” who now seem to be almost single cell specialists.

This drive to specialism is mirrored across all the allied medical professions. … Why has this happened? The primary driver has been the attempt to improve the technical aspects of medical care; most of medicine, however, is not technical. There are secondary drivers too: generalism has a low status, there is more money in specialising, and modern society has come to venerate the specialist, a proxy for “better”.

We are passing the tipping point: increasing specialisation is harming care. … The time has come for an international moratorium and non-proliferation of medical specialisation. We are undermining the confidence and position of the generalist, and soon there will be no one left good enough to call a generalist.

Like medicine, architecture is a discipline that relies almost entirely on generalism; and here too we see exactly the same drivers as described above for medicine. To paraphrase Des Spence above, yes, there is a great need to improve the technical aspects of architectures; yet most of architecture, however, is not technical. It’s much more about people – how real people interact with the real context to make real choices: technical matters are important, of course, but they’re not what the architecture is actually about.

And as I know to my cost, generalism still has far too low a status in many people’s eyes. Years ago I used to find myself frequently coming second in contractor-interviews, because I didn’t have quite the specialist knowledge that the ‘successful’ candidate could display; and frequently I found myself being called back to do the same nominal role a few months later, when they’d discovered that actual need ran across multiple domains, of which the ‘winning’ specialist could only work in one small subset.

(Interestingly, one reason why the ‘small countries’ seem to be so far advanced in enterprise-architecture is that we simply do not have enough people to enable anyone to become too much of a specialist: the technical pool is so small – relative to ‘big countries’ such as the UK, Germany, the US, or, now, India or China – that we were forced to become become generalists, whether we wanted to or not. Back when I was doing code, I used to reckon on having to learn at least two or three new programming languages every year; web-development meant that we had to switch mental ‘hats’ from user-experience to SQL databases to server-side versus client-side trade-offs to network-routers and everything in between. By contrast, in the ‘big countries’ most roles seem to involve at most a couple of those layers: doing any more than that means that you gain the title of ‘architect’, but the pay-rate goes down. Bizarre…)

And Gartner’s Mike Rollings brought up much the same theme around misplaced specialisation, again drawing on his own family background to compare the enterprise-architect to the mediaeval barber:

My father was a barber but he was a specialist focused on grooming.  Back in medieval days you could go to a barber for just about anything.  You could receive a haircut, a shave, a blood letting, a tooth extraction, surgery… the barber could do it all.

… I’m sure barbers fought for a long time to keep surgery, and tooth extraction as part of the trade.  Factions must have formed around surgery and barbering. Barbering methodologies each proclaiming they were the true practice.  It is similar to conversations about enterprise architecture and how the IT industry discusses the role of an enterprise architect.

We now have many different types of hairstylists, surgeons, and dentists… I wonder how many others think history will repeat itself and distribute various aspects of the EA discipline across many business roles and professions?

To me another brilliant if somewhat poignant illustration of generalism in one of its highest forms is in the BBC documentary Spitfire Women [UK only, until Tuesday 21 Sept 2010]. It describes the wartime role of the women of the Air Transport Auxiliary – the ferry-pilots who delivered all types of aircraft from the factories to the front-line airfields, usually alone, often in appalling weather, at a maximum altitude of 2000ft, with no radio and no modern-day navigation-aids. But the key point here is “all types of aircraft”: in some cases they might deliver a simple single-engined trainer in the morning, go back for a four-engined Lancaster bomber in the early afternoon, and in the evening “the greatest aviation prize of all”, an ultra-high-performance Spitfire. All in one day’s work. All of it solo. With only a half-sized ring-binder stuffed down the side of a boot as a guide to the different flying-characteristics and constraints of all the different types. One woman casually mentioned that over last four years of the war she’d flown seventy-six different aircraft-types: few regular wartime pilots would fly even a tenth of that number. The skill-levels that that work would demand were immense: and yet throughout that whole period they struggled to receive appropriate recognition or appropriate pay. Some of that, yes, was due to the entrenched sexism of the time; but I suspect the equally entrenched denigration of the generalist played its part too.

Which brings us back to the present-day, and enterprise-architecture and other architectures. Which depend on the somewhat strange skills of the generalist. Are we at risk, as Des Spence says above about medicine, that “we are undermining the confidence and position of the generalist, and soon there will be no one left good enough to call a generalist”? – because if that happens, we’ll have no viable architectures left…

The IT-centric obsessions of the past couple of decades have been bad enough for enterprise-architecture; but if we’re not careful, the cult of the specialist will kill architecture entirely. As generalists we need to make it clear that, without us, the specialists could deliver almost nothing that is of practical use; that without us, everything would fall apart – literally.

It’s time to stand our ground: time to put the specialists back into their preferred pigeon-holes, and reclaim enterprise-architecture as our own.

Spotted this one in the BMJ again: kind of illustrates the power of the pilots’ simple ‘pre-take-off checklist’ – or in this case a similar checklist for surgeons:

Research by World Health Organization (“New England Journal of Medicine” 2009 Jan 14, doi:10.1056/NEJMsa0810119 ) shows that use of a simple 19-point peri-operative checklist – checks immediately prior to and during surgical operations – reduced complications by one-third (from 11% to 7%) and deaths by 40% (1.5% to 0.7%).

The results, from a large statistical base (c.7500 cases) were much the same in rich and poor countries. The English-language version of the checklist is at www.who.int/patientsafety/safesurgery/en .

On the same day, I came across this op-ed piece in the London newspaper The Independent, by columnist Dominic Lawson: “I blame computers for this crisis“:

The astonishing power of computers is a wonderful thing, enabling us to do things we would never have dreamed of even a decade ago. Electronic calculation, however, is not the same thing as wisdom; and there are great dangers in confusing the one with the other.

Figures which might cause a nervous collapse when analysed within the human brain, seem soothingly manageable when generated by the click of a mouse; and, of course, there is the usual tendency to think that because it is generated by a computer, it is in some sense ‘right’ – even if the assumptions were originally fed in by a human who had never spoken to, still less met, the end user of the financial instrument. … As the former banker Charles Morris notes, “In 1983, modelling the payout scenarios on a comparatively simple three-tranche CMO [Collaterised Mortgage Obligations]  took a mainframe computer a whole weekend. But by the 1990s, when Sun workstations were standard furniture, CMO shops gleefully turned out phantasmagorical 125-tranche instruments that no-one could possibly understand.” …

If that’s the case, then just how much of the present mess is the fault of enterprise architects, from failing to carry out even basic safety checks? Perhaps more useful, how much is it – or should it be – our responsibility to create some similar means to help get of this mess? Seems to me that if a simple checklist can have that much impact in medicine – 20-30 lives saved just in that relatively small sample, along with vast effective reductions in pain, suffering and financial and other costs – what impact would the equivalents have in enterprise architecture and business architecture?

So what equivalent ‘peri-operative’ checklists could and should we develop? What improvements might we achieve with such checklists? And how would we verify their business value, and value in general?

Came across this one in this week’s edition of the BMJ (British Medical Journal).

You’ve heard of the Palm Pilot: well, now we have the Palm Pistol – “an ergonomically innovative single shot double action only defensive firearm”.

It gets worse. It’s being specifically sold as “ideal for seniors, disabled or others who may have limited strength or manual dexterity”. In other words, a gun specifically for the aged (or demented, more like).

It gets worse again. The manufacturers have formally applied for “medical equipment coding from the Center for Medicare and Medicaid services”, so that the gun can be prescribed by doctors as an aid to “extension of life expectancy” and “prevention of sudden onset death”, fully paid for by the state.

And yes, it gets worse again, because, as reported at MedGadget, some lunatic at the FDA (US Food and Drugs Administration) actually approved the thing:

The handgun will be listed as a Class I Medical Device, exempt from 510(k) Pre-Market Notification in accordance with 21 CFR 890.5050 “Daily Activity Assist Device.”

Only in America, perhaps, could anyone be so insane as to describe an instrument of intentional murder as a “Class 1 Medical Device”.

Yikes…

[Update: seems there is some sanity in the world after all: in a later blog-entry, MedGadget reports that the FDA have now rescinded their initial approval, with a rather lame assertion that they'd been misinformed by the manufacturers. Hmm...]

As you’ll see from the comments to my previous post on Cynefin, Dave Snowden again kindly came back with an appropriate critique:

The fact that order exist in nature (a constrained system that prevents agent action independent of the system) does not entail the statement that therefore all things can be reduced to order.

If you look at this from the perspective of constraints, relaxation and imposition then a lot of the problems you state above go away and there is no need to see order as an abstraction it is a phase state.

Looks like I need to do a fair bit of study on the theory of constraints…

But in some ways that isn’t related to the points I was trying to make. I fear we may be talking at slight cross-purposes here: surely the idea of a ‘phase state’ is an abstraction? a way of trying to shoehorn complexity into the knowable space?

Perhaps I could try tackling this a slightly different way…

Two quotes come to mind. One is a phrase that I think comes from James Gleick’s Chaos: “it turns out that behind apparent order lies an eerie kind of chaos; and behind that chaos lies an even eerier kind of order”. The kind of ‘order’ in the known-domain is not so much simple as simplistic, and to try to make it work in the real world we rapidly find ourselves making things more and more complicated, to cover more and more ‘special-cases’ – in other words, we find ourselves in the knowable-domain, but still clinging to a strict concept of cause-and-effect, a linear notion of ‘order’. Hence Dave’s descriptions of those two domains as the ‘ordered’ side of the framework. But at some point we realise that even that isn’t going to work – usually because we run out of time to deal with the inevitable ‘analysis paralysis’ – which is when we switch over to the ‘unordered domains’. But even here we still cling onto the idea of ‘order’, although now it takes the form of Gleick’s ‘even eerier kind of order’, as attractors and constraints and so on. So what I’m suggesting here is that ‘unordered’ in Cynefin is perhaps a misnomer: it’s still ‘ordered’ – a different kind of abstraction, but still an abstraction. With all of the problems that that entails.

As I said in the last post, it’s essentially a philosophical position: we can choose to believe that “order is real”, or we can choose otherwise. I’m with Feyerabend: I don’t believe ‘order’ has any inherent reality – it’s a tool that’s useful for sensemaking, but that’s it.

In essence, it comes down to a question of ‘truth’ versus ‘usefulness’ – otherwise known as science versus technology. (Technology is not ‘applied science’ – though that’s something to be argued in another post!)

Which brings me to the second quote, or editorial headline, rather, from the current issue of the BMJ: “Cardiovascular risk tables: estimating risk is not the problem, using it to tailor treatment is”. Risk-tables are classic examples of artefacts of complex-space: we can model the attractors, the constraints and the rest of the respective collective risk. Insurance people then move sideways into the knowable-domain to calculate required profit-margins and the rest, in terms of the collective risk. But the moment we think that these tables tell us much about individuals, we’re in deep trouble – straight into the Gambler’s Fallacy, in fact. The notion of ‘order’ fails us, because it assumes that an abstraction about a generalised collective (i.e. pattern, law, etc) will apply to any individual which might appear to belong to the set described by that abstraction. But in terms of set-theory, if the individual only intersects with the abstraction’s set – in other words, is not entirely enclosed as a subset – then other factors may come to play, which may take priority over the expectations of the abstraction.

(To give another medical example, there was an article many years ago in World Medicine with an allegorical tale of “Ulbricht the Badger’s Guide to Immunology”. The key point was that whilst we understand extremely well why people fall ill, we still understand very little about why they don’t fall ill – and even less about how they have the temerity to fall well when we don’t expect them to do so!)

Many people make the mistake of thinking that chaos-theory and attractors and the like make the unpredictable predictable. They don’t: they make the type of unpredictability, or degree of unpredictability, more predictable, but that’s it. The ‘order’ they describe is that, ultimately, there is no order. Kind of a paradox, really, but there ’tis.

Which comes back to the practice of business, and business-consultancy. It’s not an abstraction: we have to find a way to get real results, in the real world. Which happens to be made up of individual instances. Which damn well ought to tell us that, at best, we are always dealing with some of the ‘unorderedness’ of the Cynefin chaotic-domain, and probably some of the Cynefin ‘unknown’ as well. If we start to believe that ‘order is real’ – that the real world is somehow ‘wrong’ if it fails to match our expectations – then we’re on the slippery slope to the inane world of Victorian medicine, in which surgeons routinely reported “operation successful, but patient died”…

Science expects ‘fail-safe’ conformance to its ‘truth’; but technology aims for useful approximations – hence ‘safe-fail’. (Therein lies another post I ought to write, about ‘Inverse Murphy’.) In technology, we take an occurrence which is probably rare in nature, and provide conditions under which it becomes more and more probable – but as any maintenance engineer would warn us, we don’t delude ourselves into thinking we have any real ‘control’ over what happens. However probable it may become, it is never more than probable: it never becomes certain. The delusion of ‘control’ is useful when we’re dealing with incidents en masse – as in risk-tables and the like – but it’s not useful once we come down to the infinite complexities of individuals.

Hence, in Cynefin terms, we start off in the ‘unknown’ space; we’re usually wisest to start sensemaking in the ‘unordered’ space, working orderwards (clockwise) from chaotic, to complex, to knowable, to known. Dave describes this as “from exploration to exploitation”, but I’d suggest it’s better described as “from exploration to planning for exploitation” – because to do the exploitation, we need to work back to unordered in order to deal with individuality, the ‘market of one’ and suchlike.

So I don’t see how descriptions of phase-states and the like actually help that much in this – true, they’re not the same kind of ‘order’ as in the known and knowable domains, but it’s still ‘order’, still an abstraction. Useful for planning, for sense-making – but sometimes dangerously misleading at ‘the coal-face’.

Sure, attractors and constraints and the like are a heck of a lot more useful than Taylorism when we’re dealing with the messy complexity of the real-world: but if we’re not careful, the notion of ‘order’ itself – in whatever form it may take – can lead us straight back to what is really nothing more than a subtler form of the Taylorist trap. Guess that’s all I’m saying, really.