Portfolio Management

As you consider your options for partitioning your product, in terms of the AKF scaling cube, a useful and widely adopted distinction is that between “features” and “components”; see Features versus Components.

Features are what your product does. They are what the customers perceive as valuable. “Scope as viewed by the customer” according to Mark Kennaley [Kennaley 2010]. They may be “flowers” – defined by the value they provide externally, and encouraged to evolve with some freedom. You may be investing in new features using Lean Startup, the Spotify DIBB model, or some other hypothesis-driven approach.

Components are how your product is built, such as database and web components. In other words, they are a form of infrastructure (but infrastructure you may need to build yourself, rather than just spin up in the cloud). They are more likely to be “cogs” – more constrained and engineered to specifications. Mike Cohn defines a component team as “a team that develops software to be delivered to another team on the project rather than directly to users” [Cohn 2009].

Feature teams are dedicated to a clearly defined functional scope (such as “item search” or “customer account lookup”), while component teams are defined by their technology platform (such as “database” or “rich client”). Component teams may become shared services, which need to be carefully understood and managed (more on this to come). A component’s failure may affect multiple feature teams, which makes them riskier.

It may be easy to say that features are more important than components, but this can be carried too far. Do you want each feature team choosing its own database product? This might not be the best idea; you will have to hire specialists for each database product chosen. Allowing feature teams to define their own technical direction can result in brittle, fragmented architectures, technical debt, and rework. Software product management needs to be a careful balance between these two perspectives. SAFe suggests that components are relatively:

than features. SAFE also recommends a ratio of roughly 20-25% component teams to 75%-80% feature teams [SAFe 2016a].

Mike Cohn suggests the following advantages for feature teams [Cohn 2009]:

He also suggests that component teams are justified when:

Ultimately, the distinction between “feature versus component” is similar to the distinction between “application” and “infrastructure". Features deliver outcomes to people whose primary interests are not defined by digital or IT. Components deliver outcomes to people whose primary interests are defined by digital or IT.

In the coordination Competency Category, we talked of one product with multiple feature and/or component teams; see One Company, One Product. Features and components, as we are discussing them here, are large enough to require separate teams (with new coordination requirements). At an even larger scale, we have new product ideas, perhaps first seen as epics in a product backlog.

Eventually, larger and more ambitious initiatives lead to a key organizational state transition: from one product to multiple products. Consider our hypothetical startup company. At first, everyone on the team is supporting one product and dedicated to its success. There is little sense of contention with “others” in the organization. This changes with the addition of a second product team with different incentives; see One Company, Multiple Products. Concerns for fair allocation and a sense of internal competition naturally arise out of this diversification. Fairness is deeply wired into human (and animal) brains, and the creation of a new product with an associated team provokes new dynamics in the growing company.

Because resources are always limited, it is critical that the demands of each product be managed using objective criteria, requiring formalization. This was a different problem when you were a tight-knit startup; you were constrained, but everyone knew they were “in it together”. Now you need some ground rules to support your increasingly diverse activities. This leads to new concerns:

Structurally, we might decide to separate a portfolio backlog from the product backlog. What does this mean?

The DEEP backlog we discussed in Work Management gets split accordingly; see Portfolio versus Product Backlog.

The decision to invest in a new product should not be taken lightly. When the decision is made, the actual process is as we covered in Product Management: ideally, a closed-loop, iterative process of discovering a product that is valuable, usable, and feasible.

There is one crucial difference: the investment decision is formal and internal. While we started our company with an understanding of our investment context, we looked primarily to market feedback and grew incrementally from a small scale. (Perhaps there was venture funding involved, but this document does not cover that.)

Now, we may have a set of competing ideas on which we are thinking about placing bets. In order to make a rational decision, we need to understand the costs and benefits of the proposed initiatives. This is difficult to do precisely, but how can we rationally choose otherwise? We have to make some assumptions and estimate the likely benefits and the effort it might take to realize them.

Fundamentally, we plan for two reasons:

We have discussed investment decision-making in terms of the overall business context, in terms of the product roadmap, the product backlog, and in terms of Lean Product Development and cost of delay. As we think about making larger-scale, multi-team digital investments, all of these practices come together to support our decision-making process. Estimating the likely time and cost of one or more larger-scale digital product investment is not rocket science; doing so is based on the same techniques we have used at the single-team, single-product level.

With increasing scope of work and increasing time horizon tends to come increasing uncertainty. We know that we will use fast feedback and ongoing hypothesis-driven development to control for this uncertainty. But at some point, we either make a decision to invest in a given feature or product and start the hypothesis testing cycle – or we do not.

Once we have made this decision, there are various techniques we can use to prioritize the work so that the most significant risks and hypotheses are addressed soonest. But in any case, when large blocks of funding are at issue, there will be some expectation of monitoring and communication. In order to monitor, we have to have some kind of baseline expectation to monitor against. Longer-horizon artifacts such as the product roadmap and release plan are usually the basis for monitoring and reporting on product or initiative progress.

In planning and execution, we seek to optimize the following contradictory goals:

Obviously, we want outcomes – digital value – but we want it within constraints. It has to be within a timeframe that makes economic sense. If we pay 40 people to do work that a competitor or supplier can do with three, we have not produced a valuable outcome relative to the market. If we take 12 months to do something that someone else can do in five, again, our value is suspect. If we purchase software or hardware we do not need (or before we need it) and, as a result, our initiative’s total costs go up relative to alternatives, we again may not be creating value. Many of the techniques suggested here are familiar to formal project management. Project management has the deepest tools, and whether or not you use a formal project structure, you will find yourself facing similar thought processes as you scale.

To meet these value goals, we need to:

Estimation sometimes causes controversy. When a team is asked for a projected delivery date, the temptation for management is to “hold them accountable” for that date and penalize them for not delivering value by then. But product discovery is inherently uncertain, and therefore such penalties can seem arbitrary. Experiments show that when animals are penalized unpredictably, they fall into a condition known as learned helplessness, in which they stop trying to avoid the penalties.

We discussed various coordination tools and techniques previously. Developing plans for understanding dependencies is one of the best known such techniques. An example of such a planning dependency would be that the database product should be chosen and configured before any schema development takes place (this might be a component team working with a feature team).

Agile and adaptive techniques can be used to plan larger, multi-team programs. Again, we have covered many fundamentals of product vision, estimation, and work management in earlier material. Here, we are interested in the concerns that emerge at a larger scale, which we can generally class into: