BayesInsights: Modelling Software Delivery and Developer Experience with Bayesian Networks at Bloomberg Conference: 2026 ACM International Conference on the Foundations of Software Engineering (FSE); July 5–9, 2026; Montreal, Canada CCS: Computing methodologies Bayesian network models CCS: Social and professional topics Software management CCS: Software and its engineering Software development process management

Serkan Kirbas, Federica Sarro, David Williams · 2026

Read BayesInsights: Modelling Software Delivery and Developer Experience with Bayesian Networks at Bloomberg Conference: 2026 ACM International Conference on the Foundations of Software Engineering (FSE); July 5–9, 2026; Montreal, Canada CCS: Computing methodologies Bayesian network models CCS: Social and professional topics Software management CCS: Software and its engineering Software development process management by Serkan Kirbas, Federica Sarro, David Williams (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-29929.txt.

Introduction

Kirbas, Sarro, and Williams report on BayesInsights, a Bloomberg tool for modeling software delivery. Engineering data piles up, but root causes stay hard to spot. The tool maps causal links across the development process with Bayesian networks. Its structure mixes published research, expert input, and learned graphs. A survey of 24 senior staff found 95.8 percent called it useful. It now runs in preview for seven teams. A wider rollout is planned.

Catalog record

CREATORSerkan Kirbas, Federica Sarro, David Williams
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30529
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.29929

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

BayesInsights: Modelling Software Delivery and Developer Experience with Bayesian Networks at Bloomberg Conference: 2026 ACM International Conference on the Foundations of Software Engineering (FSE); July 5–9, 2026; Montreal, Canada CCS: Computing methodologies Bayesian network models CCS: Social and professional topics Software management CCS: Software and its engineering Software development process management Serkan Kirbas, Federica Sarro, David Williams Abstract . As software in industry grows in size and complexity, so does the volume of engineering data that companies generate and use. Ideally, this data could be used for many purposes, including informing decisions on engineering priorities. However, without a structured representation of the links between different aspects of software development, companies can struggle to identify the root causes of deficiencies or anticipate the effects of changes. In this paper, we report on our experience at Bloomberg in developing a novel tool, dubbed BayesInsights , which provides an interactive interface for visualising causal dependencies across various aspects of the software engineering (SE) process using Bayesian Networks (BNs). We describe our journey from defining network structures using a combination of established literature, expert insight, and structure learning algorithms, to integrating BayesInsights into existing data analytics solutions, and conclude with a mixed-methods evaluation of performance benchmarking and survey responses from 24 senior practitioners at Bloomberg. Our results revealed 95.8% of participants found the tool useful for identifying software delivery challenges at the team and organisational levels, cementing its value as a proof of concept for modelling software delivery and developer experience. BayesInsights is currently in preview, with access granted to seven engineering teams and a wider deployment roadmap in place for the future. 1. Introduction At Bloomberg, the DevX team distributes quarterly surveys to software engineers regarding the state of the firm’s software engineering (SE) practice. The survey comprises Likert-scale questions capturing engineers’ perceptions of their work environment, including satisfaction and engagement, and covers topics such as how meaningful they find their work, how much time is lost due to obstacles, the quality of peer feedback, as well as aspects of the SPACE framework ( 14 ) and the DevOps Research and Assessment (DORA) metrics ( 13 ; 10 ) . The results of these surveys, along with contribution metrics, are visualised via dashboards and reporting tools. While such dashboards provide visibility into delivery performance, they do not explain why certain outcomes occur or how changes in one area might affect others. For example, a change in one metric, such as reduced deployment frequency, may reflect improvements in related factors like CI/CD or automated testing, making interpretation ambiguous ( 13 ) . As a result, teams can struggle to estimate trade-offs or predict the impact of engineering changes. In response to this, we investigated the use of Bayesian Networks (BNs) for causal analysis in SE practices. We propose an interactive tool, dubbed BayesInsights , to model causal and probabilistic dependencies among key SE factors at Bloomberg. We use data from one internal engineering survey (20 questions, >2,000 responses) as the source for our BNs, in which each question is represented as a graph node (with the node states corresponding to the question options). We derive the structure of links between these nodes using a hybrid and iterative modelling approach consisting of ① reviewing established literature (§ 2.1 ), ② refinement through expert opinion (§ 2.2 ), and ③ validation using structure learning algorithms (§ 2.3 ). Figure 1. The DevEx Bayesian Network in BayesInsights showing causal links among factors influencing developer happiness. Users can explore “what-if” scenarios by selecting a bar within a node, fixing that outcome and updating the network to reflect its impact on related factors. Mocked probability distributions are shown to protect sensitive data; values are illustrative only. The DevEx Bayesian Network in \toolName{} showing causal links among factors influencing developer happiness. Users can explore “what-if” scenarios by selecting a bar within a node, fixing that outcome and updating the network to reflect its impact on related factors. Mocked probability distributions are shown to protect sensitive data; values are illustrative only. Figure 1 illustrates an example of a BN in BayesInsights (with mocked probability distributions to protect sensitive data). Here, we can see that factors such as focus without distraction , environment performance , and code understanding influence time lost to obstacles , while both time lost to obstacles and meaningful work contribute directly to developer happiness . With BayesInsights , users can conduct impact analyses through “what-if” scenarios by selecting evidence. For instance, users can estimate the extent to which minimising time lost to obstacles would improve developer happiness, or how maximising development environment performance cascades through the system to influence outcomes. We conducted a mixed-methods evaluation of BayesInsights that combines quantitative performance benchmarking with qualitative feedback collected through a post-session survey following focus group sessions with 28 senior Bloomberg practitioners (24 of whom responded). BayesInsights demonstrated both practical responsiveness and strong perceived usefulness: 95.8% reported that BayesInsights helps identify delivery challenges at the team or organisational level, and 75% found the outputs easy to interpret. Moreover, 79.2% would use or recommend the tool, and many suggested avenues for improvement, motivating further development. BayesInsights is currently deployed in early access to seven engineering teams, with a development roadmap in place for wider deployment in future. This paper provides an in-depth description of our BN creation methodology and insights into the value of causal modelling for SE practice at a large-scale software company. 2. Bayesian Network Structure Definition For the initial implementation of BayesInsights , we developed two BNs based on internal engineer survey data. The first BN models software delivery performance , represented by two dimensions: throughput , which captures the speed and frequency of delivering changes, and stability , which reflects the reliability of changes. Throughput is modelled using change lead time , deployment frequency , and failed deployment recovery time , while stability is represented by change failure rate . These metrics form the core model and are influenced by other upstream engineering practices. The second BN modelling DevEx (Developer Experience) 1 1 1 In this context, DevEx refers to the satisfaction and engagement of engineers within their work environment. and the factors that shape it (Figure 1 ) is derived from other aspects covered in the internal engineering survey, such as perceptions of meaningful work , time lost to obstacles , and focus without distraction . Constructing Directed Acyclic Graphs (DAGs) for BNs in SE contexts presents significant challenges ( 35 ) . Software delivery metrics exhibit complex interdependencies that are rarely observed in isolation. Hidden confounders, including organisational culture, toolchain maturity, and team composition, can create spurious associations between variables. This challenge is compounded by the inherent limitations of SE data. Surveys are prone to interpretation and recall bias, and the temporal nature of software development processes further complicates causal inference, as relationships between variables may differ across teams and evolve with project or organisational changes. These factors collectively render naive, data-only structure learning approaches unreliable and highlight the necessity of a multi-source methodology to construct credible DAG structures for causal inference ( 35 ; 6 ) . To devise our network structures, we adopted a hybrid modelling approach combining theoretical insights from the literature ,

Read the complete text