Templafy 报告发现：95% 的知识工作者在 AI 生成文档达到商用标准前仍需手动编辑

Templafy对美国和英国2,000名知识工作者的调查显示，90%每周至少使用一次AI创建商业文档，但95%仍需编辑，平均每周花近4小时审阅、修正和验证输出；48%表示因此几乎没有净节省时间。准确性（40%）、泛化或套话内容（34%）及格式问题（31%）是主要缺陷，凸显企业AI生产力仍受缺乏内部知识、经批准内容和工作流上下文整合的限制。其平台数据则显示，在启用功能的用户中，AI智能体采用率由2025年12月的31%升至2026年8月的66%，并将文档创建时间中位数从约2小时降至8分钟。

Analysis

The investable implication is a shift in enterprise AI budgets from model access toward governance, retrieval, workflow orchestration and approved-content layers. MSFT is best positioned near term because Copilot sits inside the document workflow, but weak first-pass output raises the risk that seat adoption does not translate into sustained premium usage or measurable ROI. This favors vendors that can reduce review liability rather than merely add generation features; the relevant competitive set extends to ServiceNow (NOW), Adobe (ADBE), OpenText (OTEX) and privately held enterprise-content platforms.

For GOOG and CRM, the issue is less AI capability than enterprise data permissioning and implementation friction. Gemini and Salesforce Agentforce can benefit where customer-specific context is available, but deployments requiring fragmented content repositories, legal approval chains and bespoke templates will lengthen sales cycles and push value capture toward systems integrators and data-governance providers. Over the next 1-3 months, scrutinize AI attach-rate commentary versus reported net revenue retention and services costs: rising adoption without lower support or implementation burden would undermine margin-expansion narratives.

Consensus likely overstates the immediacy of broad knowledge-worker productivity gains and understates the compliance-driven consolidation opportunity. In the next 6-18 months, regulated verticals may standardize on fewer, deeply integrated platforms, benefiting MSFT and NOW while pressuring stand-alone copilots lacking trusted enterprise content access. The key falsifier is independently disclosed evidence that AI users reduce document-cycle time and external-review costs without incremental human oversight; vendor-sponsored survey metrics alone are insufficient.

Key Decisions for Investors

• Maintain a selective long bias in MSFT versus GOOG over a 3-6 month horizon: MSFT has the stronger workflow distribution advantage, but size modestly until Copilot renewal and paid-seat utilization demonstrate ROI beyond trial deployment. Reassess if commercial remaining-performance-obligation growth or M365 AI attach commentary weakens materially.
• Watch-list long NOW on any AI-related pullback over the next 1-3 months. Its workflow, governance and enterprise-record positioning should capture a larger share of remediation spend than generic document-generation vendors; require evidence of stable subscription margin and accelerating large-deal activity before entry.
• Avoid adding to CRM solely on Agentforce adoption headlines. Upgrade only if AI bookings convert into improving data-cloud consumption, retention and services-margin leverage; otherwise implementation intensity can dilute the expected multiple benefit over the next 2-4 quarters.
• Use ADBE and OTEX as second-order monitoring names rather than immediate trades: increased demand for governed brand assets and content control is plausible, but financial sensitivity depends on enterprise AI packaging and cross-sell disclosure not provided here.