---
id: "framework-four-steps-knowledge-decay"
type: "framework"
source_title: "Don't Let AI Slop Muck Up Your Company's Processes"
source_url: "https://hbr.org/2026/06/dont-let-ai-slop-muck-up-your-companys-processes"
source_timestamps: ["§ How to Deal with Knowledge Decay?"]
tags: ["strategy", "implementation"]
related: ["action-track-provenance", "action-restrict-unstructured-inputs", "action-use-proprietary-slms", "action-redesign-interorganizational-processes"]
speakers: ["Matthias Holweg", "Thomas H. Davenport"]
steps: ["Keep track of the provenance of unstructured data to separate ground truth from generated content.", "\\\"Restrict the use of generative AI by requiring structured data inputs (e.g.", "questionnaires instead of free-form CVs) to prevent AI optimization arms races.\\\"", "Define what value is being added by shifting focus from public LLMs generating generic prose to proprietary SLMs generating insights based on proprietary data.", "\\\"Understand the implications for the entire process by redesigning end-to-end workflows", "especially interorganizational ones", "to preserve content integrity.\\\""]
sources: ["execution"]
sourceVaultSlug: "hbr-seg-execution"
originDay: 8
articleStem: "hbr-sig-54-ai-slop-processes"
sourceUrl: "https://hbr.org/2026/06/dont-let-ai-slop-muck-up-your-companys-processes"
sourceTitle: "Don’t Let AI Slop Muck Up Your Company’s Processes"
---
# Four Steps to Deal with Knowledge Decay

A strategic framework for leaders to manage the process and knowledge implications of generative AI. It starts from the premise that outright policing of the technology is impossible ([[claim-policing-ai-impossible]]), so structural design — not prohibition — is the lever.

1. **Track provenance of unstructured data** → [[action-track-provenance]], grounded in [[concept-unstructured-data-provenance]]. Separate human ground truth from generated content.
2. **Restrict generative AI via structured inputs** → [[action-restrict-unstructured-inputs]]. Replace free-form CVs/cover letters with specific questionnaires to defuse the 'AI optimization' arms race.
3. **Define the value being added** → [[action-use-proprietary-slms]], grounded in [[claim-public-llms-low-value]]. Use proprietary SLMs on proprietary data for insight; relegate public LLMs to formatting.
4. **Understand implications for the entire process** → [[action-redesign-interorganizational-processes]], grounded in [[claim-process-redesign-required]] and [[concept-productivity-paradox]]. Redesign end-to-end and cross-boundary workflows to preserve integrity.

The enrichment overlay affirms this maps well onto governance guidance (NIST AI RMF; PwC/HITRUST/Wolters Kluwer), while cautioning that steps 1–2 depend on detection capabilities the authors themselves admit are weak (see [[question-detecting-ai-content]]).