Organisational Change Through Experiments

First up, a huge thanks to Mike Pollard for the inspiration on this one. This all started with a meeting invite from Mike to set up some experiments in organisational change. We all know that organisational change is hard. Organisations tend to resist change so doing any sort of substantial change is a lot of work, and also prone to failure as organisations slip quietly back into their old way of doing things. Since real agile success relies somewhat on changing some pretty fundamental things in the organisation, this has always been a pretty major limiting factor in agile adoptions - success relies on change and is limited by how much change we can introduce. Change is hard which limits the amount of success we can have.

Mike's idea was quite simple - rather than try to change the whole organisation, why not set up some small experiments instead? That gives the organisation a low risk way to see what works and what doesn't. Once we have some successful experiments we should have some good, hard data to back us up when we push for a wider rollout.

The problem with organisational change is that we usually look at it in the context of the whole organisation. It's a big thing. Changing a large organisation is a huge task and carries some pretty serious risk as well. Particularly when we start messing with things like the way organisations fund projects, or how they hire people. These changes are big, and generally irreversible. Once you change, it's really hard to change back if things go wrong. Big change is hard. Especially big, risky change. No exec is going to sign up for that unless you can really sell them on it. That sort of change for them is a career defining move - either it works and they are a hero, or it doesn't and they are unemployed.

The real problem here is that most of the time we don't have a good way to sell the change we need. We have no evidence. We can point to case studies, but case studies are almost always based on somewhere else. The big fear in an exec's mind when they read a case study is that here is different to there in some way that will cause whatever worked there to fail here. Being able to point at lots of case studies help - it's worked in lots of places so it has a better chance of working here, but what execs really want is proof that it will work here. Unfortunately, the way to see if it works here has usually been to jump in and try it, which is a high risk scenario.

Experimentation gives us another way. Mike's approach was to treat a big organisational change as a lean startup. We have a vision for where we want to go. That comes with a series of assumptions. Form some hypotheses around those assumptions so that we can start to run some experiments to test them. If our experiment is a success, great, we run another to prove the next assumption. If it doesn't, we pivot and try something else. We have to be careful to frame these as actual experiments. They should be answering a specific question and we have to have some good way to measure success. The output of an experiment should be data, not anecdotes.

The key here is that we keep the experiments small. If we want to see the effect of a new team structure, don't try to change all 500 teams in the enterprise, change one or two. Let them run for an few months and see what happens. If we want to experiment with funding models, don't change the whole funding model, fund a single project or program under the new model and see what happens.

This takes the risk right out of the change. These are small, reversible changes. If they don't work, just reset that tiny part of the organisation back to the default. If they do, then fantastic. Either way it's great because we now have some hard data that we can apply to the organisation and use it to guide our next set of experiments.

Even better, because the experiments are small we can run lots of them and learn quickly. Need to change the team structure but not sure which new structure will work? Set up a dozen trial teams to see what effect different structures have. Need to change funding model but not sure what will work? Run a bunch of trial projects, each funded under a different model.

At the end of each experiment, analyse the data you collected (you did collect data, didn't you?). What does it tell you? Do we persevere and roll out the change wider? Or do we pivot and try something different? Either way we have learned something and we have some hard data to back us up. It's not just opinion any more. We aren't asking execs to take what we say on faith. We have real data about this organisation that tells us this will work (provided we set up our experiments correctly of course).

Although this sounds simple, there are some complexities. For a start, large organisations usually have one process for doing something so having a bunch of strange, experimental things happening will cause mass panic in those parts of the organisation dedicated to maintaining order. These experiments need to come with a level of executive support and a willingness to shield the experiments from the process police. That should be a much easier sell than a whole-scale transformation though.

Like any large piece of work, the best way to do a large organisational change is to split it into small chunks, each of which adds some value, then work on them in priority order. Organisational change, agile style.