The F-word: Failure. It is development agencies’ worst nightmare and surely sends shudders down the spine of any program manager. However, for the Rio+20 agenda to be successful, development agencies must start learning how to fail. Doing so will help them to learn rapidly and innovate.
The priority areas of the Rio+20 agenda are complex. Few easy, cookie-cutter solutions exist. What’s the best way to create more jobs? How do we encourage improved agricultural practices that don’t put our ecosystem at risk? These are difficult questions with no easy answer. A process of experimentation—rapid iterations of trial, error, and redesign—will be fundamental to achieving important development goals while not compromising the needs of future generations.
Unfortunately, development agencies are failing precisely because they don’t know how to fail. Development agencies are not considered hotbeds of innovation—precisely because they are unwilling to take the necessary risk (and concomitant failure) that breakthrough innovation requires. Moreover a culture fearful of failure means that failures that could have been caught early snowball into large headline-grabbing failures.
But, not every development agency is failing to know how to fail. Engineers without Borders, for example, produces an annual Failure Report. FAILFaire holds conferences and hosts a blog on failures in development. Space for acknowledging failure helps development agencies to learn and to maintain an agile, high-performing portfolio of projects in organizations otherwise sluggish to recognize and move-on from failing programs.
So, how can development agencies better learn to fail?
1. Differentiating good failure from bad failure: Creating the political space for failure by managing expectations.
Failure fails when it is used as an excuse to shirk responsibility or diffuse accountability. The first step to managing failure in development agencies is differentiating between good and bad failure. Amy Edmondson, a professor at Harvard Business School, describes three types of failure:
- Preventable failures in predictable operations – Like a bus driver failing to stop at a red light, this failure is unambiguously bad. Better processes and training to prevent this type of failure (and address it when it does occur) are important for maintaining accountable organizations.
- Unavoidable failures in complex systems – This type of failure may be inevitable, but can be managed in order to lower the costs of failure.
- Intelligent failures at the frontier – This is the best type of failure. While this failure can still be made cheaper and earlier, failure associated with “trial and error” must be actively encouraged, not punished.
Being clear up front about what types of failure are okay should help to create more political space in the organization for “good” failure. After-the-fact “failure audits” can help get to the root cause of why failures occurred and incorporate learning back into the organization.
2. Greater use of data combined with both structured and unstructured experimentation
A process of experimentation is fundamental to allowing the best kinds of failure (those at the frontier of knowledge) to occur in real conditions. Incorporating data in management systems is just as important in order to more rapidly identify failed experiments, understand what caused them and adapt.
The experiments on the cutting-edge of development are randomized evaluations. These experiments are able to carefully tell development agencies the impact of a given program, isolating out all the other possible factors that may have influenced program outcomes. The strategic use of randomized evaluations, combined with a careful understanding of field conditions can help development agencies weed out poorly performing programs and scale-up high-impact, cost-effective programs.
But, experimentation need not be so structured. Encouraging greater autonomy at a local level can be just as effective at inducing a culture of experimentation that enables high-impact local innovations to be scaled. For example, the Charlotte School District, through a policy of “Freedom and Flexibility with Accountability,” allowed high performing principals to circumvent established policy and procedures as long as they continued to deliver high results. Some principles tried out single-gender classrooms; others put multiple teachers in one classroom; still others used new technology in the classroom. Complementing these local innovations was the systematic use of data which allowed Charlotte School District to better understand what was working best.
3. Fail cheaper and earlier in the design cycle
This brings us to the third key to enabling better failure in development agencies: Failure needs to happen cheaper and earlier. The randomized evaluations discussed above are great at cleanly identifying causal impact—but they often cost several hundred of thousands of dollars and take multiple years before the results come out. An over-reliance on randomized evaluations means that failure is caught far too late in the design cycle and after far too much investment.
Cheaper and earlier failure in development agencies would encourage “good-enough” experimentation. Good enough experimentation attempts to isolate as many confounding variables as possible and prioritizes rapid prototyping and iteration over perfectly constructed counterfactuals in order to learn and innovate faster.
The question is not will development agencies fail, but how they fail and will they learn. As governments, civil society and other stakeholders convene in Brazil to discuss the sustainable development agenda, they must recognize that today’s challenges require an improved framework more conducive toward learning and risk-taking. More open, acceptable failure is therefore needed in order to drive innovation and keep agencies agile.
Michael Eddy is a Class of 2012 Master in Public Administration-International Development candidate at the Harvard Kennedy School.