1.4 Benefits of Continuous Delivery
Continuous Delivery offers numerous benefits. Depending on the scenario the different advantages can be of varying importance—consequently this will influence how Continuous Delivery is implemented.
1.4.1 Continuous Delivery for Time to Market
Continuous Delivery decreases the time required for bringing changes into production. This generates a substantial benefit on the business end: It becomes much easier to react to changes of the market.
However, the advantages extend beyond a faster time to market: Modern approaches like Lean Startup4 advocate a strategy that benefits even more from the increased speed. The focus of Lean Startup is to position products on the market and to evaluate their chances at the market while investing as little effort as possible in doing so. Just like with scientific experiments, it is defined beforehand how the success of a product on the market can be measured. Then the experiment is performed, and afterwards the success or failure is measured.
1.4.2 One Example
Let us look at a concrete example. In a web shop a new feature is supposed to be created: Orders can be delivered on a defined day. As a first experiment the new feature can be advertised. Here the number of clicks on a link within the advertisement can be used as an indication for the success of this experiment. At this point no software has been developed yet—that is, the feature is not yet implemented. If the experiment did not lead to a promising result, the feature does not appear to be beneficial and other features can be prioritized instead—without much effort having been invested.
1.4.3 Implementing a Feature and Bringing It into Production
If the experiment was successful, the feature will be implemented and brought into production. Even this step can be conducted like an experiment: Metrics can help to control the success of the feature. For example, the number of orders with a fixed delivery date can be measured.
1.4.4 On to the Next Feature
The analysis of the metrics reveals that the number of orders is high enough—interestingly most orders are not sent directly to the customer, but to a third person. Additional measurements show that the ordered items are obviously birthday presents. Based on this information the feature can be expanded—for example with a birthday calendar and recommendations for suitable presents. This requires of course that such additional features are designed, implemented, brought into production and finally evaluated with regards to their success. Alternatively, there might also be options to evaluate the potential market success of these features without any implementation—via advertisements, customer interviews, surveys, or other approaches.
1.4.5 Continuous Delivery Generates Competitive Advantages
Continuous Delivery makes it possible to more rapidly bring required software changes into production. This allows enterprises to more quickly test different ideas and to develop their business model further. This creates a competitive advantage: Since more ideas can be evaluated, it is easier to filter out the right ones—and this is not based on subjective estimations of market chances, but on the basis of objectively measured data (Figure 1.1).
Figure 1.1 Reasons for Continuous Delivery in a startup
1.4.6 Without Continuous Delivery
Without Continuous Delivery the feature for the fixed delivery dates would have been planned and brought into production during the next release—this would likely have taken a number of months. Before the release, marketing would hardly have dared to advertise the feature since the long time up to the next release would render such advertisements futile. If the feature had not proven successful in the end, there would have been high costs caused by its implementation without creating any benefit. Evaluating the success of a new feature would certainly also be possible in the classical approach; however, the reaction would be drastically slower. Further developments such as the features supporting the buying of birthday presents would reach the market much later since they would require that the software be brought into production again and the time-consuming release process be run through a second time. Besides, it remains doubtful whether the success of the feature would have been analyzed in enough detail to recognize the potential for additional features supporting the shopping for birthday presents.
1.4.7 Continuous Delivery and Lean Startup
Therefore, optimization cycles can be passed through much faster thanks to Continuous Delivery because each feature can be brought into production practically at any time. This makes approaches like Lean Startup possible. This influences how the business end is working: It has to more rapidly define new features and does not have to focus on long range planning anymore, but can immediately react to the outcome of the current experiments. This is especially easy in startups, but such structures can also be built in classical organizations. The Lean Startup approach has, unfortunately, a misleading name: It is an approach where new products are positioned on the market via a series of experiments, and this approach can of course also be implemented in classical enterprises, not only in startups. It can also be used when products have to be delivered classically—for instance, on media such as CDs, with other complex installation procedures, or as part of another product such as a machine. In such a case the installation of the software has to be simplified or ideally automated. Besides a range of customers has to be identified who would like to test new software versions and be willing to provide feedback on them—that is, classical beta testers or power users.
1.4.8 Effects on the Development Process
Continuous Delivery influences the software development process: When individual features are supposed to be brought into production, the process has to support this. Some processes use iterations of one or several weeks’ length. At the end of each iteration a new release with several features is brought into production. This is not an ideal approach for Continuous Delivery because in this way individual features cannot pass through the pipeline on their own. This also poses obstacles for the Lean Startup approach: When several features are rolled out at the same time, it is not obvious which change influences the measured values. Let us assume that the option for delivery on a fixed date is introduced in parallel with a change of the shipment costs—it will not be possible to distinguish which of the two changes had a greater influence on the higher number of sold items.
Therefore, processes like Scrum, XP (Extreme Programming), and of course the waterfall are impedimentary since they always bring several features together into production. Kanban,5 on the other hand, focuses on bringing a single feature through the different phases into production. This fits ideally with Continuous Delivery. Of course, the other processes can also be modified in ways that allow them to support the delivery of individual features. However, in such a case the processes have been adapted and are not implemented according to the textbook anymore. Another possibility is to initially deactivate the additional features in order to bring several features together in one release into production, but still be able to measure their effects separately.
In the end this approach also means that teams include multiple different roles. In addition to development and operation of the features, business roles such as marketing are conceivable. Thanks to the decreased organizational hurdles, the feedback from the business end can be translated into experiments even faster.
1.4.9 Continuous Delivery to Minimize Risk
The use of Continuous Delivery as described in the last section goes together with a certain business model. However, for classical enterprises the business often depends on long-range planning. In such a case an approach like Lean Startup cannot be implemented. In addition, there are many enterprises for which time to market is not a decisive factor. Not all markets are very competitive in this regard. This can of course change when such companies are suddenly confronted with competitors that are able to enter the market with a Lean Startup model.
In many scenarios time to market cannot motivate the introduction of Continuous Delivery. Still the techniques can be useful since Continuous Delivery offers additional benefits:
Manual release processes require a lot of effort. It is no rare event that entire IT departments are blocked for a whole weekend for a release. And after a release there is frequently still extensive follow-up work to do.
Also the risk is high: The software rollout depends on many manual modifications, which easily leads to mistakes. If the errors are not discovered and fixed in time, this can have far-reaching consequences for the enterprise.
The sufferers are found in the IT departments: Developers and system administrators who work through weekends and nights to bring releases into production and to fix errors. In addition to working long hours they are subjected to high stress because of the high risk. And the risks should not be underestimated: Knight Capital, for instance, lost $440M because of a failed software rollout.6 As a consequence the company went into insolvency. A number of questions7 arise from such scenarios—in particular why the problem occurred, why it wasn’t noticed in a timely manner, and ultimately how such events can be prevented in other environments.
Continuous Delivery can be a solution for such situations: Fundamental aspects of Continuous Delivery are the higher reliability and the quality of the release process. This allows developers and system administrators to sleep calmly in the true sense of the word. Different factors are relevant for this:
Due to the higher level of automation of the release processes, the results become easier to reproduce. Thus, when the software has been deployed and tested in a test or staging environment, the exact same result will be obtained in production because the environment is completely identical. This allows largely eliminating sources of error, and consequently the risk decreases.
In addition, testing software becomes much easier since the tests are largely automated. This increases the quality further as the tests can be performed more frequently.
When there are more frequent deployments, the risk decreases likewise since fewer changes are brought into production per deployment. Fewer changes translate into a smaller risk that an error has crept in.
In a way, the situation is paradoxical: The classical IT tries to bring releases as seldom as possible into production since they are associated with a high risk. During each release an error with potentially disastrous consequences can creep in. Fewer releases should therefore result in fewer problems.
Continuous Delivery on the other hand advocates frequent releases. In that case fewer changes go live at each release, which also decreases the probability for the occurrence of errors. Automated and reliable processes are a prerequisite for this strategy. Otherwise the frequent releases lead to an overload of the IT personnel performing manual processes, and in addition the risk increases since errors are more prone to occur during manual processes. Instead of aiming at a low release frequency the relevant process are automated to decrease the release-associated risk. It is of course an added advantage that in case of a high release frequency the individual releases comprise fewer changes so that the inherent risk of errors is lower.
Here, the motivation for Continuous Delivery (Figure 1.2) thus profoundly differs from that of the Lean Startup idea: The focus is on reliability and a better technical quality of the releases—not on time to market. And the beneficiaries are the IT departments—not only the business domains.
Figure 1.2 Reasons for Continuous Delivery in an enterprise
Since the benefits are different, other compromises can be made: For example, investing in a Continuous Delivery pipeline is often worthwhile even if it does not extend all the way up to production—that is, when the production environment still has to be built manually. In the end the production has only to be built once for each release, but multiple environments are required for the different tests. However, if time to market is the main motivation for Continuous Delivery, it is essential that the pipeline include production.
1.4.10 Faster Feedback and Lean
When a developer modifies the code, she receives feedback from her own tests, integration tests, performance tests, and finally from production. If changes are brought into production only once per quarter, several months can pass between the code modifications and the feedback from production. The same can hold true for acceptance or performance tests. If an error occurs then, the developer has to think back to what it was she had implemented months ago and what the problem might be.
With Continuous Delivery the feedback cycles become much faster: Every time the code passes through the pipeline the developer and his/her entire team receive feedback. Automated acceptance and capacity tests can be run after each change. This enables the developer and the development team to recognize and fix errors much more rapidly. The speed of feedback can be further increased by preferring fast tests, such as unit tests, and by first testing broadly and only afterwards testing deeply. This ensures from the start that all features function at least for easy cases—the so-called “happy path.” This makes spotting basic errors easier and faster. In addition, tests that are known from experience to fail more often should be executed at the start.
Continuous Delivery is also in line with Lean thinking. Lean regards everything that is not paid for by the customer as waste. Any change to the code is waste until it is brought into production since only then will the customer be willing to pay for the modifications. Besides, Continuous Delivery implements shorter cycle times for faster feedback—another Lean concept.