How to lead an underfitting team

A SDM told me he was hired into a new organization, leading a team of new engineers, to own a new charter of the org.

"That sounds like a recipe for disaster!" I said.

"I know. But life is not always ideal...", the SDM said, "I am in this mess already, how do I make the best out of it?"

Using a term from Machine Learning (ML), the team is underfitting. The model of the team, its structure and collective skills etc., is unable to capture the relationship between the input (stakeholders' requirements) and output (value deliveries) accurately, causing missing deadlines, low quality product or projects dead on arrival.

Since we are using an analogy from machine learning systems, here are some ideas to fix an underfitting team.

1. The Need for More Data A new manager with an underfitting team lacks the insights to make precise decisions. A ML model needs data to train on and refine its understanding. Similarly, a manager needs input from team members, stakeholders, and the market to get a holistic view of the challenges and opportunities that lie ahead. It's a misconception that only high-quality data can lead to good results. What's more crucial is the quantity of data. Even if some of it is noisy or less relevant, as long as it aids in quick learning and decision-making, it's worth considering. A new manager cannot expect to get everything right in the first go. The strategy should be to implement, gather feedback, learn, and iterate. Every iteration helps the manager improve, much like every additional round of training strengthens the predictive power of a model.

2. The Power of Fast Feedback Feedback, especially if it's timely, can be instrumental. For machine learning models, quick feedback in the form of backpropagation helps the model move in the right direction. For managers, fast feedbacks from sprint demos or Minimum Viable Product (MVP) aid in real-time course correction.

3. Recognizing When to Stop or Pivot Just as in machine learning, where we might introduce an early stop to try a different model altogether if the current one isn't working, a new manager too must know when to halt a particular strategy or pivot to a new one.

One of the more subtle pitfalls in both domains is getting caught up in 'abstract learning.' In machine learning, this might look like spending excessive time in theoretical model design or feature engineering without practical data. For managers, this translates to getting bogged down by extensive requirement documents, premature estimates, or over-designing without actionable insights.

The core message? While planning and theory are useful, pay attention to "The Unreasonable Effectiveness of Data"