The Path to Engineering Maturity 🗺️
A deep industry report to figure out what practices allow teams to perform best
Hi everyone! In October we launched a wide survey to understand how engineering teams work and how they use data to improve.
Since then, we have collected ~350 answers from engineers and managers, and spent two months working through them to gather insights.
Today we are finally presenting our findings. It’s a long, deep work, to kick off the new year and hopefully inspire some new resolutions! ✨
Here is the agenda:
🔍 Scope — topics, audience size, and goals of this work.
📖 Methodology — how we worked through quantitative and qualitative data.
🏆 What is a successful engineering team? — working backwards from people’s answers.
🔧 Successful Practices — teams that do these things consistently do better than others.
📊 The Role of Metrics — how teams use data to improve.
🔺 The Pyramid of Engineering Maturity — weaving our findings into a simple top-down framework.
Let’s dive in!
🔍 Scope
We got 334 full responses, from people with a variety of roles in tech:
1) Role 👑
After removing “other” roles that are outside of software engineering (hello to the four Marketing Managers who answered this survey!), we get almost perfect parity between IC and Manager roles (187 vs 184). That’s by counting Tech Leads in both camps, as the role varies significantly depending on the company.
2) Company size 👥
Company size distribution is similar to that of other surveys we have run in the past. About one third of participants work at companies with less than 25 engineers:
3) Work Setup 🏢
The vast majority of respondents work from home in some capacity, with only 11.7% working full-time in an office.
However, the office population almost doubled from last year (11.6% vs 6.3%), where we asked the same question in a similar survey. The full remote share, instead, stayed the same, with most of the migration happening from hybrid to office setups.
This suggests—as is to be expected—that full remote is significantly harder to revert than hybrid.
4) Questions 🗳️
Then, we asked the following questions:
It’s a lot of stuff! So here is how we worked through the answers 👇
📖 Methodology
Finding good, reliable insights is hard, for many reasons:
Correlation ≠ causation — even if some things are clearly correlated, it’s hard to figure out what’s upstream and what’s downstream.
Statistical significance — given the (relatively) small numbers, you can only trust big deviations.
Human bias — we are humans so we come to the party with our own set of opinions and biases. If I expect some result, I may more or less intentionally look for it in the data.
So, to find good patterns, we put all the data on a giant Airtable, where we exploded results by several dimensions.
Then, for each candidate insight, we looked for confirmation from multiple angles, including similar questions, related topics, and open answers where respondents open up a bit more about their team’s practices.
About open answers, we went through all of them manually to extract individual stories and ideas, we tagged them into various categories, plus used AI to uncover gems we might have missed.
The result is a work that, while still subject to our own interpretation, aims to be a profound, reliable analysis that blends thousands of quantitative and qualitative data points.
🏆 What does a successful team look like?
To figure out what practices lead to successful engineering teams, first we need to understand what a successful team looks like.
Out of all the questions we asked, there are three we are particularly fond of:
Engineering is well-regarded among the leadership / non-tech stakeholders
Engineers are happy about dev practices
Projects ship on time
These are the elements that correlate the strongest with the highest number of positive behaviors. That is: the gradient of answers for these questions is most likely to be matched by a similar gradient in other behaviors.
Some examples (see the picture for more):
In teams where devs are very happy about dev practices, vs those where they are neutral, focus time is 30% more satisfactory.
In teams where engineering is not well regarded by non-technical stakeholders, vs those where it is, people need to wait for others 29% more time.
In teams that ship 75% of projects on time, vs those who ship only 30%, people spend 28% more time as it was planned.
So, the first two questions are strictly qualitative and collect the opinions of both technical and non-technical stakeholders. We believe these are extremely important: in fact, most quantitative practices (e.g. how often you ship, how much time is spent on KTLO, …) depend on the team’s context and do not decisively sort good teams from bad ones. Conversely, we would be hard-pressed to find low performing teams where all stakeholders—engineers, managers, and leadership—are happy.
The last question is about predictability. We found that shipping on time correlates strongly with most good behaviors we polled for. By most measures, predictable shipping is even more important than frequent shipping.
So, what kind of traits are exhibited by teams that ship on time and where stakeholders are happy? As you can see from the pictures, three traits are the most common:
🔬 Engineers have enough focus time
⏱️ Engineers spend time as it was planned
⏳ Engineers don’t need to wait for others
This is not surprising, but it is always nice to see it backed by numbers.
Engineering thrives when developers they can cultivate autonomy and mastery. Conversely, it suffers when developers are not in control of their time, need to constantly wait for others, and spend time in meetings and putting out fires instead of doing creative work.
But what enables focus time, no waiting, and predictable work? Let’s get into the core of the report, looking into specific practices 👇
🔧 Successful practices
A good chunk of the survey was made up of questions about team or engineering practices. Once we identified what successful teams look like, we tried to figure out what practices were the most correlated with success indicators.
We selected the six where such a correlation was the strongest: