A Scaling Sequence

The next pattern, Mitosis, is a scaling pattern. We often see Scrum Teams with a desire to scale, and the primary reason is to increase output. Increased production must, however, be matched by corresponding market consumption; otherwise, the excess productivity is just waste. In any case, the best way to grow a team’s output is by removing impediments and by improving the process.

Scaling legitimately occurs under just one circumstance: when the product grows organically to the point where the demand for new features outstrips even a high-performing Scrum Team’s capacity to deliver in a timely fashion. A 'high performing' team is one that has been a Stable Team for some while, maintains a consistent velocity, and continually delivers on business outcomes. If such a team, using Kaizen Pulse, has diminishing returns on its self-improvement efforts relative to the market’s required rate of delivery of business value, then it may be an option to add Development Teams.

Scaling is not about transforming an existing micromanaged company to an Agile one (that is a quite different question; see Involve the Managers). When someone asks, “We have over 500 developers: how do we scale to Scrum?” they are asking the wrong question. How do they know they need 500 developers? Nor is ‘scaling’ a proper response to a product which is late in delivery. Brook’s Law [1] — that adding people to a late project makes it later — still applies at team level. Scaling is about the piecemeal growth of the Development Teams in response to the growth of the product itself. Scrum has always scaled in that sense. The first multi-team Scrum was initiated by Jeff Sutherland at IDX in 1994. Exactly how Scrum scales is in large part situational, but the following patterns are common. Most of them fall within the scope of the Product Organization pattern language.

Reacting with the old muscle memory of waterfall, management’s knee-jerk reaction to a need for greater productivity is to add new people to existing teams, or to add new teams. In the big majority of cases this is a mis-step. Scrum has its roots in the Toyota Production System (TPS), and one of its architects wrote a whole book based on the contrast between its approach and “large-scale production.” [2] Coaching the Product Owner to focus the Scrum Team on the most important work to be done; helping the ScrumMaster to remove impediments; growing the Development Team’s ability to deliver quality-assured product offer the best approach.

In a minority of cases, the above may be insufficient. But it is still necessary to first improve the Scrum Team to its maximum. Large organizations grow piecemeal from small organizations that work. That’s what this sequence is about. One may need to scale across geographically remote markets to meet diverse market needs (extensive localization or dependence on local laws, customs, or usage patterns). One may need to scale simply because of the intellectual mass of a single product. Examples include systems such as those that track and identify military threats and guide missiles to intercept them, or telecommunications products that must provide all telecommunication needs — local phone, toll calls, data, video, emergency services, billing, CRM, mobile, voice messaging, text messaging, Centrex, and business services — to a single market.

The following patterns focus on scaling a development effort. The underlying message is that Scrum can be scaled only with Scrum itself. The alignment of multiple teams behind a single product Vision requires that they work from a single Product Backlog managed by a Product Owner supported by an appropriately-sized Product Owner Team. Other patterns (like Definition of Done) are about the same as for single teams, with the change being only in unified scope of application (i.e., all teams share a common Definition of Done). Above all, the Development Teams’ self-organization and autonomy is a precondition for success. We present the patterns in a typical sequence of application you may want to consider following if you plan to grow your organization.

§ 7  Scrum Team.  The starting point is always the single Scrum Team. Most organizations are tempted to ‘scale’ as a shortcut to fixing the team. In the absence of a high performing team you will only be scaling your dysfunctions and adding complexity to your organization.

§ 12  Product Owner Team.  A Product Owner Team works under the leadership of a single Chief Product Owner and helps him or her in Product Backlog management across Value Areas.

§ 19  Mitosis.  If the product grows beyond the capacity of a single, high-performing Scrum Team to deliver, new developers may be hired. With too many people in the Development Team, the team is split into two smaller Development Teams. Organize the teams so that they remain Cross-Functional Teams and work across all parts of the product. The teams must keep delivering end-user functionality from a single Product Backlog. Organize development so that the Cross-Functional Teams can keep delivering end user functionality, following Conway’s Law.

§ 42  Scrum of Scrums.  With multiple teams working from the same Product Backlog, alignment with the product Vision must be preserved. The self-organizing Development Teams collaborate to deliver common Sprint Goals, and coordinate with each other to resolve emergent issues and dependencies.

§ 45  MetaScrum.  Regularly coordinate among multiple products within an enterprise, and entertain management input through a forum comprising the Product Owner and executive management, which works on aligning the expectations of management stakeholders.

§ 55  Organizational Sprint Pulse.  All the teams work in the same Sprint cadence and schedule to maintain a consistent view on progress. All teams together develop an integrated single Product Increment each Sprint that is a Done contribution to furthering the Vision.

§ 97  Value Areas.  The product domains become too broad for all teams to understand, yet the product cannot be split into separate products. The teams specialize according to areas of customer value, each one continuing to be cross-functional across its value area. This is a temporary organization, put in place until teams again grow to be cross-functional teams.

§ 98  Value Stream Fork.  Successful Value Streams may grow in scope, and one problem with success is that the result can outstrip the original Vision. If so, the Value Stream is split into multiple streams, each with its own Product Backlog, bringing focus and a refined Vision to each one.

[1] Frederick P. Brooks, Jr. “The Mythical Man-Month.” In The Mythical Man-Month: Essays on Software Engineering. Addison-Wesley, 1975, Chapter 2.

[2] Taichi Ohno. Toyota production system: Beyond large-scale production. Cambridge, MA: Productivity Press, 1988.