There is a divide in the way Objectives and Key Results (OKRs) are practiced. A big divide, a divide between the way some of the original authors describe OKRs and the way successful agile teams implement them. If you haven’t spotted it yet it might explain some of your problems, if you have spotted it you might be feeling guilty.
The first school of thought believes OKRs should be set by a central figure. Be it the CEO, division leadership or central planning department, the OKRs are set and then cascaded, waterfall style, out to departments and teams.
Some go as far as to say “the key results of one level are the objectives of the lower levels.” So a team receiving an OKR from on high take peels of the key results, promotes each to Objective status. Next they add some new key results to each objective and hand the newly formed OKR to a subordinate team. The game of pass the parcel stops when OKRs reach the lowest tier and there is no-one to subordinate.
The second school of thought, the one this author aligns with, notes that cascading OKRs in this fashion goes again agile principle: “The best architectures, requirements, and designs
emerge from self-organizing teams.” In fact, this approach might also reduce motivation and entrench the “business v. engineer” divide.
Even more worryingly, cascading OKRs down could reduces business agility, and eschew the ability to use feedback as a source of competitive advantage and feedback.
Cascading OKRs
We can imagine an organization as a network with nodes and connecting edges. In the cascading model information is passed from the edge nodes to the centre. The centre may also be privy to privileged information not known to the edge teams. Once the information has been collected the centre can issue communicate OKRs back out to the nodes.
One of the arguments given for this approach is that central planning allows co-ordination and alignment because the centre is privy to the maximum amount of information.
A company using this model is making a number of implicit assumptions and polices:
- Staff at the centre have both the skills to collect and assimilate information.
- That information is received, decisions made and plans issued back in a timely fashion. Cost of delay is negligible.
However, in a more volatile environment each of these assumptions falls. Rapidly changing information may only be known to the node simply because the time it takes to codify the information — write it down or give a presentation — may mean the information is out of date before it is communicated. In fact the nodes may not even know they know something that should be communicated. Much knowledge is tacit knowledge and is difficult to capture, codify and communicate. Consequently it is excluded from formal decision making processes.
The loss of local knowledge represents a loss of business agility as it restricts team’s ability to act on changing circumstances. Inevitably there will be delays both gathering information and issuing out OKRs. As an organization scales these delays will only grow as more information must be gathered, interpreted and decisions transmitted out. Connecting the dots becomes more difficult when there are more dots, and exponentially more connection, to connect.
This approach devalues local knowledge, including capacity and ambition. Teams which have no say in their own OKRs lack the ability to say “Too much”, they goals are set based upon what other people think — or want to think — they are capable of.
Similarly, the idea of ambition, present in much OKR thinking, moves from being “I want to strive for something difficult” to “I want you to try doing this difficult thing.” Let me suggest, people are more motivated by difficult goals that they have set themselves more than difficult goals which are given to them.
Finally, the teams receiving the centrally planned OKRs are likely to experience some degree of disempowerment. Rather than being included and trusted in the decision making process team members are reduced to mere executers. Teams members may experience goal displacement and satisficing. Hence, this is unlikely to lead either to high performing teams or consciences, responsible employees.
Any failure in this mode can be attributed to the planners who failed to anticipate the response of employees, customers or competitors. Of course this means that the planners need more information, but then, any self-respecting planner will have factored their own lack of information into the plan.
Distributed OKRs
In the alternative model, distributed OKRs, teams to set their own OKRs and feed these into any central node and to leaders. This allows teams to factor in local knowledge, explicit and tacit, set OKRs in a timely fashion and determine their own capacity and ambitions.
One example of using local knowledge is how teams managing their own work load, for example balancing business as usual (or DevOps) work with new product development. As technology has become more common fewer teams are able to focus purely on new product development and leave others to maintain existing systems.
Now those who advocate cascading OKRs will say: “How can teams be co-ordinated and aligned if they do not have a common planning node?” But having a common planner is not the only way of achieving alignment.
In this model teams have a duty to co-ordinate with both teams they supply and teams which supply them. For example, a team building a digital dashboard would need to work with teams responsible for incoming data feeds and those administering the display systems. Consequently, teams do no need to information from every node in the organization — as a central planning group would — but rather only those nodes which they expect to interact with.
This responsibility extends further, beyond peer teams. Teams need to ensure that their OKRs align with other stakeholders in the organization, specifically senior managers. In the same way that teams will show draft OKRs to peer teams they should show managers what they plan to work on, and they should be open to feedback. That does not mean a manager can dictate an OKR to a team but it does mean they can ask, “You prioritising the French market in this OKRs, our company strategy is to prioritising Australia. Is there a reason?”
A common planner is but one means of co-ordination, there are other mechanisms. Allowing teams the freedom to set OKRs means trusting them to gather and interpret all relevant information. When teams create OKRs which do not align it is an opportunity not a failure.
When two teams have OKRs which contradict, or when team OKRs do not align with executive expectations there is a conversation to be had. Did one side know something the other did not? Was a communication misinterpreted? Maybe communication failed?
Viewed like this OKRs are a strategy debugger. Alignment is not mandated but rather emerged over time. In effect alignment is achieved through continual improvement.
These factors — local knowledge and decision making, direct interaction with a limited number of other nodes and continual improvement — are the basis for local agility.
Pull don’t push
Those of you versed in the benefits of pull systems over push systems might like toes this argument in pull-push terms. In the top down approach each manager, node, pushes OKRs to the nodes below them. As with push manufacturing the receivers have little say in what comes their way, they do their bit and push to the next in lucky recipient in the chain.
In the distributed models teams pull their OKRs from their stakeholders. Teams ask stakeholders what they want from the team and they agree only enough OKRs to do in the coming cycle.
This may well mean that some stakeholders don’t get what they wanted. Teams only have so much capacity and the more OKRs they accept the fewer they will achieve. Saying No is a strategic necessity, it is also an opportunity to explore different options.