Why Bad Requirements Remain the Software Industry’s Biggest Toxic Problem

Demand problems are described as a long‑standing toxic issue in software development. The 2024 Standish Group Chaos Report shows that 73% of global project failures stem from "requirement information distortion"; domestic surveys report that rework due to poor requirements averages 42% of total project investment, and 83% of online incidents trace back to gaps in the requirement phase.

Six major sources of requirement distortion are identified:

Imagination‑driven assumptions : a maternity‑app’s "night feeding reminder" was built on personal guesswork, achieving <1% usage until real user data revealed a need for an automatic timer.

Semantic ambiguity : vague terms like "optimize registration process" led developers to reduce form fields, yet users complained about opaque audit progress. Each ambiguous word raises rework probability by 15%.

Pseudo‑agile changes : an online‑education live‑class system underwent 57 requirement changes in six months, expanding capacity from 1,000 to 100,000 users without clear planning.

Late discovery : a car‑infotainment system required wireless CarPlay only during acceptance testing, forcing a costly Bluetooth module redesign; such post‑stage omissions occur in 68% of domestic projects.

Ineffective documentation : a 200‑page ERP requirement spec failed to define inventory‑alert thresholds; over 40% of documents contain internal contradictions, e.g., differing import limits in user stories versus business rules.

Technical gap : a logistics app initially rejected an "automatic exception dispatch" feature citing data volume, but after business data showed a 5 万元 daily loss, the scope was narrowed to three key fields and the solution was delivered.

Business value of precise demand management is illustrated with three cases:

Bank: a "requirement freeze" 15 days before major promotions cut change requests from 32 to 9, eliminated online faults, saved >8 M RMB in rework, and raised user satisfaction by 15%.

Medical software: adopting Acceptance‑Test‑Driven Development (ATDD) increased requirement clarification efficiency by 60%; AI‑generated test‑case drafts reduced clarification meetings from 5 days to 2, cutting rework caused by vague requirements by 70%.

E‑commerce platform: applying MoSCoW prioritization reduced delivery cycle from 45 to 27 days and overtime by 50%.

Constructing a modern demand‑quality governance system involves:

Standardized specifications: replace vague phrases with structured "user story + acceptance criteria" (Gherkin format). Large models can act as validation assistants, checking completeness and flagging missing scenarios.

Visual change‑cost modeling: build a calculator that estimates impact scope, rework cost, and schedule effect; establish a Change Control Board for multi‑discipline review.

Integrating LLMs into ATDD: automatically generate multi‑scenario acceptance tests, compare requirement and design docs to spot gaps, and produce compliant test data.

End‑to‑end traceability: assign a unique ID to each requirement and link it to feedback, design, code commits, tests, and production metrics, enabling rapid root‑cause tracing and periodic quality reports.

In conclusion, while the industry often focuses on AI‑generated code or intelligent operations, the true leverage lies in using AI to amplify requirement certainty; the ultimate value of software is determined by the precision of its initial demand, not by the elegance of the code.