Issue No. 04 - July/August (2007 vol. 24)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/MS.2007.113
Hakan Erdogmus , National Research Council Canada
Enterprise software is plainly problematic, no matter whose perspective you take—that of the vendor selling it, the buyer acquiring and operating it, or the people using it. It's especially a nightmare to deploy and upgrade. Everybody suffers. Software as a service promises to address some of the ills afflicting product-based enterprise software. But a few flies might be lurking in the ointment.
The virtues of on-demand
In a keynote address at a recent major conference, the executive of an up-and-coming IT company evangelized the virtues of delivering enterprise automation as an on-demand service. I believe he made a good case.
He recounted the several benefits of a common, scalable platform catering to multiple subscribers through a unified underlying engine and data model. The underlying platform is shared, but a customer-specific metadata layer lets individual subscribers generate and deploy applications without the nightmarish worries of traditional enterprise systems. Ease of management and tailoring, faster release cycles, painless customer provisioning, and maintenance-free upgrades were among the benefits he touted. He admitted that platform upgrades that are completely transparent to subscribers are a challenge for service providers, but once overcome, this challenge translates into handsome payoffs for customers when they instantly inherit any improvements made.
Ease of application customization and integration is intended to empower customers. They can create virtual applications by defining and specializing layouts, data objects, workflows, reports, and so on, which persist as part of their private metadata. While wizard-like features can guide simple tailoring, the real power would come from a programmatic interface in which concepts such as database tables and workflows are first-class objects. Such an interface, in addition to customization, would let customers glue separate applications, hopefully with less complexity than current product-based alternatives require.
What about security and privacy?
In this context, security and disaster recovery aren't as disconcerting as they might initially seem. Really. A common multisubscriber platform must build in security and redundancy from the ground up. And service providers are better positioned than individual organizations to offer effective related capabilities at low cost—with defendable assurances of protection from intrusions, attacks, and failures.
The same should apply to reliability and performance. Maintaining and monitoring a single underlying platform lets the provider quickly identify bottlenecks and glitches and deploy global fixes to address them.
Privacy is touchier than security. The thought that their organization's critical data are physically stored on an external party's server, intermingled with many others' data, can't be comforting to executives. As individuals, we might no longer be concerned about keeping our personal communications in an email account owned and maintained by a faceless service provider together with millions of other users. Whether rational or not, the perceived benefits override any privacy and ownership concerns we might still harbor. The situation is more complicated for organizations, where such matters are business decisions, subject to explicit justification and contingency planning. Nevertheless, several companies are apparently willing to put their concerns aside to reap the promised benefits—for now.
All appears well so far. So what's the catch? Here some old ghosts resurface. One is about guaranteed continuity and data ownership. Another pertains to the promised utopia of effortless end-user application development.
Closed for business
On top of platformwide internal redundancy, service providers sometimes encourage customers to further protect themselves by offering an external back-up capability. Customers may then regularly back up their hosted data in some standard database format to a server that they control. The actual physical data the provider uses internally for its multisubscriber platform are of no value to the customer. Neither may the provider make the raw data available without compromising other subscribers' sensitive information and its own intellectual property. The idea is that the platform assembles a snapshot of the subscriber's private data on demand and makes this snapshot available for external backup. This strategy should theoretically offer a reasonable assurance to the customer, but in practice, it does not.
Suppose the provider goes out of business. The customer is left with a static database and no metadata. The end users don't directly interact with a database but with a virtual application that implements a set of business processes encoded within the metadata. So, the machinery that brought the whole thing alive has effectively disappeared with the provider. This machinery consists of the metadata (created and owned by the customer) and the engine (created and owned by the provider) that interprets the metadata. Putting new machinery in place could amount to nearly the same enterprise automation problem the customer faced before the system existed. Worse yet, now the organization relies on the system for its day-to-day operation.
As a first-order solution, the backed-up data could also include the metadata, a representation of the underlying business processes, including custom workflows, reports, layouts, dashboards, and so on in standard formats. At least the customer would possess all its invested corporate assets, with some hope of recovering them.
A second-order and more complete solution is for the provider to offer an easy-to-deploy, specialized, single-subscriber instance of the engine to be coupled with the external backup. Just like the private database, the hard-coded engine is automatically generated. In response to an interruption in the provider's existence, the customer would have the right to deploy the instantiated, complete system as a short-term solution to keep operations up and running—that is, until a more permanent solution becomes available.
Although developing such a second-order capability is quite hard, the provider must do it only once at the platform level. The question is whether the risk-averse customer would be willing to pay for it in return for the added protection of a significantly reduced downtime.
Who performs application development?
In many situations, the customers' illusion of partial control over data and processes might be powerful enough to automate organization-specific business processes with reasonable fidelity. Although such control is theoretically both complete and real in traditional product-based enterprise systems, the cost of change is often so prohibitive that this ultimately results in very little actual control. So we may indeed be better off with the service approach. In any case, the customer still deals with the residual complexity of application customization and integration after the provider has neatly dealt with all the platform issues.
The customer's private metadata is key to achieving control. But exactly what kind of people on the customer side would be responsible for creating, evolving, and maintaining the metadata? End users would in all likelihood be a couple of levels removed from these activities, as is the case with traditional systems. If robust application customization and integration hinge on leveraging programmatic interfaces (likely to be supported by plenty of modeling activity), such responsibility would naturally fall on the shoulders of IT departments (and consultants). The IT folks won't be absolved from sorting out complex requirements, mapping them into data models and weaving them into interacting workflows.
On the upside, the IT person's job becomes easier. Liberated from dealing with daunting and recurrent platform issues, the focus can now be shifted to the real goal: determining and implementing vital business processes for end users. As a side benefit, usability in enterprise software might finally receive the bandwidth it deserves.
The concept of enterprise software as an on-demand service doesn't eliminate customer-side software development, as some of its proponents claim. The type of software produced might be different and oriented toward the right kind of abstractions, but the dream of end-user enterprise application development without coding (or without the equally end-user-inaccessible modeling) remains, well, a dream.
An uncertain outlook
In sum, the on-demand, multisubscriber paradigm promises a sensible sharing of risks and responsibilities between enterprise service providers and consumers, enabling each party to focus on their core problems and strengths and take advantage of scale economies when doing so. While the provider maintains the platform and creates generic capabilities in a transparent manner, customers create and maintain artifacts related to their core corporate assets. This synergy is preferable to frictions resulting from clunky product-based solutions, but it's not a panacea. Nor is its success, technologically or in the market, guaranteed.
It will be exciting to see how the enterprise software industry evolves in response to emerging service-based business models. Traditional players have already been incorporating "services" in their vocabulary. Will additional obstacles and opportunities arise? How will providers and their customers innovate in turn? What will the impact be on software development practices? And how will end users be affected? I'm interested in your perspective. Reach me at firstname.lastname@example.org.