When to kill a software project

2026-05-28 · Source: LeadDev · Field: Technology & Digital — Software Development & Engineering, Software Project Management · Depth: Intermediate, medium

Summary

NinjaOne's SVP of data and AI, Joel Carusone, details a structured approach to terminating software projects, noting that only 31% succeed and 19% are canceled before production, per the CHAOS Report. Carusone advocates for setting predefined "kill criteria" and timeboxes for new initiatives, often measured in engineering hours or weeks. For example, a vulnerability tracking feature at NinjaOne required 90% confidence in data matching; failure to meet this within the timebox led to a pivot to a hybrid coding and large language model (LLM) strategy. Projects undergo regular reviews for cost, financial viability, and market competitiveness, involving product managers. A "single dissenter kill switch," a senior technical contributor, holds authority to terminate projects, asking if they would be started today with current knowledge. This strategy prevents sunk cost fallacy, maintains morale, and focuses resources, as NinjaOne invests in roughly one in ten explored ideas.

Key takeaway

For Directors of AI/ML or Product Managers overseeing new software initiatives, you should implement clear, upfront "kill criteria" and timeboxes for all projects. Designate a senior technical contributor as a "single dissenter" empowered to terminate projects if they fail to meet predefined goals or market viability checks. This approach prevents the sunk cost fallacy, preserves team morale by stopping efforts early, and ensures your valuable engineering resources are consistently focused on the most promising and strategically aligned endeavors.

Key insights

Establishing upfront kill criteria and empowering a single decision-maker optimizes software project lifecycles.

Principles

• Define project exit conditions proactively.
• Designate one senior leader for termination decisions.
• Continuously assess project viability against market.

Method

Implement predefined timeboxes and kill criteria. Conduct regular reviews tracking cost and market fit. Empower a "single dissenter" to make final termination calls based on current knowledge.

In practice

• Timebox initial proofs of concept to weeks.
• Set confidence thresholds (e.g., 90%) for key unknowns.
• Involve product managers in early financial reviews.

Topics

• Software Project Management
• Project Termination
• Kill Criteria
• Timeboxing
• Sunk Cost Fallacy
• Engineering Leadership
• LLM Project Costs

Best for: Director of AI/ML, Product Manager, Software Engineer

Related on AIssential

• Cognitive Offloading in Agile Teams: How Artificial Intelligence Reshapes Risk Assessment and Planning Quality — Hybrid human-AI sprint planning enhances risk capture and adaptability, outperforming AI-only or human-only approaches.
• Stop Tuning Hyperparameters. Start Tuning Your Problem. — The article asserts that over 80% of Artificial Intelligence (AI) projects fail, not due to model or data deficiencies, but from "misunderstanding (or miscommunicating) what problem needs to be solved…
• What Oracle’s Layoffs Really Signal For B2B Marketing, Sales, And Revenue Operations — Organizations are shifting to fund outcomes over effort, accelerating a re-evaluation of roles based on decision integrity.
• Your AI Transformation Will Collapse by Month 8. Here is the 5-Stage Framework That Prevents It — Rushed, top-down AI transformations often fail due to lack of strategic planning and stakeholder buy-in.
• Does every feature we build with AI need a token budget? — AI features require comprehensive token budgeting across build, run, and maintenance, guided by hypotheses.
• The solution might be cancelling my AI subscription — AI tools can amplify both project creation and abandonment, but also enhance focus for some with ADHD.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by LeadDev.