Workplace AI Usage & Security Pulse Survey

Published: 13 November 2026| Version 1 | DOI: 10.17632/khx7dkyhm7.1
Contributor:
Anurag Banerjee

Description

Generative Artificial Intelligence (GenAI) is rapidly penetrating enterprise knowledge workflows, yet organizations face a critical tension between operational productivity gains and data governance risks. This dataset provides a structured cross-sectional benchmark of 1,782 organizational responses evaluating generative AI tool adoption, functional use cases, perceived bottlenecks, and data security awareness across key business units (Engineering/Product, Marketing & Communications, HR/Legal, Finance & Accounting, Sales & Account Management, and Customer Support & Operations).Structure & Variable Codebook The dataset contains 1,782 records structured across 9 core survey dimensions: Department / Team: Primary business division of the respondent. AI Usage Frequency: Longitudinal engagement frequency ranging from daily integration to non-use. Tool Diversity & Category: AI coding assistants, enterprise-licensed platforms, public LLMs, visual generative tools, and transcription assistants. Primary Workflow Tasks: Code generation/review, brainstorming, communication drafting, synthesis/summarization, and quantitative spreadsheet analysis. Primary Bottlenecks: Licensing deficits, training gaps, accuracy/hallucination verification overhead, and unclear corporate guidelines. AI Security Policy Familiarity: Self-reported familiarity with corporate AI governance. Data Security Check: Objective assessment testing employee knowledge of permitted vs. prohibited data types in public tools (e.g., PII, internal roadmaps vs. generic code/anonymized text). Quarterly Resource Prioritization: High-impact interventions requested by staff (enterprise licenses, specialized prompt engineering training, internal custom GPTs, cheatsheets). Qualitative Feedback / Tool Requests: Open-text input highlighting specific tool dependencies, licensing approvals, and prompt engineering workshops. 5. Methodological & Data Integrity Note: Synthetic Dataset Design Privacy-Preserving Synthetic Benchmark Disclosure To ensure absolute compliance with enterprise data privacy, non-disclosure obligations, and proprietary confidentiality standards, this dataset is composed of high-fidelity synthetic data. Rather than sharing sensitive real-world enterprise employee responses or potentially identifiable corporate telemetry, the records were generated using structured probabilistic modeling and scenario simulation. The dataset replicates realistic organizational response distributions, intra-departmental behavioral variance, and authentic enterprise operational challenges. This approach provides an open, accessible, and ethical benchmark for researchers, educators, and policy analysts studying workplace AI adoption without compromising institutional or individual privacy.

Files

Steps to reproduce

To evaluate organizational generative AI adoption while safeguarding corporate privacy and proprietary telemetry, this dataset was developed using an in silico synthetic generation pipeline. The methodology combines survey instrument design with probabilistic multivariate simulation, producing 1,782 realistic employee response records across six core departments. 1. Instrument Design: The questionnaire was designed using extensions of the Technology Acceptance Model (TAM) and organizational governance frameworks. It captures nine core dimensions: departmental role, tool engagement frequency, active software categories, primary workflow tasks, adoption bottlenecks, corporate AI security policy familiarity, objective data classification compliance, desired quarterly enablement resources, and qualitative tool requests. 2. Generative Modeling Protocol: To avoid publishing sensitive internal employee data or violating non-disclosure constraints, an automated sampling architecture was executed in Python using standard statistical modeling libraries (NumPy, SciPy, and Pandas). A discrete joint distribution was defined across enterprise functional units (Engineering, Marketing, HR/Legal, Finance, Support, and Sales). Conditional probability tables were parameterized to model realistic real-world covariance: engineering profiles exhibited higher probabilities of using coding assistants and encountering licensing bottlenecks, while communications and HR profiles were weighted toward transcription utilities, multimodal drafting, and security ambiguity. 3. Objective Security Gap Synthesis: To reflect authentic organizational blind spots, a controlled discrepancy rate was modeled between subjective policy familiarity and objective security compliance. This simulated the well-documented corporate phenomenon where individuals report high policy awareness yet incorrectly identify internal roadmaps or personal accounts as acceptable for proprietary inputs. 4. Qualitative Field Generation: Open-text responses were generated using conditioned natural language prompting aligned with the categorical attributes of each simulated persona, followed by a stochastic dropout to simulate realistic non-response rates. 5. Validation and Reproduction: The resulting records were validated for tabular consistency, schema conformity, and distribution integrity in Google Sheets and Python before export. Researchers can reproduce this dataset by sampling across the described conditional probability trees and mapping the outputs to the nine standardized schema columns.

Institutions

Categories

Computer Science, Information Security, Business Management

Licence