Un informe de Templafy revela que el 95% de los trabajadores del conocimiento edita los documentos generados por IA antes de que estén listos para uso empresarial

Nearly half of workers in the U.S. and UK say reviewing AI-generated business documents can leave little of the anticipated time savings. The finding highlights an adoption and productivity risk for enterprise AI tools, as human quality-control requirements may offset automation benefits.

Analysis

The relevant read-through is not a broad AI-demand reversal but a shift in enterprise buying criteria from seats and pilot adoption toward measurable workflow-level ROI. Vendors whose monetization depends on high-volume drafting features face greater renewal and pricing scrutiny if human verification remains embedded in the process; this is most exposed in productivity software and legal/document-automation use cases, including MSFT, CRM, NOW, ADBE and DOCU. In contrast, vendors selling AI into bounded, auditable workflows—coding assistance, cybersecurity triage, contact-center routing and data extraction—should retain stronger willingness-to-pay because errors are easier to measure and constrain.

Over the next 1-3 months, the risk is incremental pressure on management commentary around Copilot/GenAI attach rates, usage persistence and gross-margin dilution from inference costs. The market has generally capitalized AI revenue before proof of labor displacement; a weak ROI narrative can therefore compress application-software multiples even if top-line AI bookings continue growing. The 6-18 month implication is favorable for the infrastructure layer—NVDA, AVGO, ORCL and hyperscalers—only if enterprises move from open-ended content generation to production deployments, which may require more integration spending and longer sales cycles rather than fewer compute workloads.

Contrarian view: review requirements are often a feature of early deployment, not evidence of zero value. Human-in-the-loop processes can still raise throughput, improve first-draft quality and create compliance audit trails; the missing metric is total cycle time and error-adjusted output, rather than whether a reviewer remains necessary. The tradeable distinction is between vendors reporting paid production usage and those emphasizing pilot counts or anecdotal productivity claims.

A near-term broad software short is not justified by this survey alone. Treat it as a diligence trigger ahead of earnings: a material thesis change requires evidence of falling weekly active usage, delayed expansion contracts, higher support costs, or explicit reductions in enterprise AI budgets.

Key Decisions for Investors

• Maintain a quality-biased software basket: favor NOW and PANW over DOCU and ADBE for the next 1-3 months, reflecting more defensible workflow integration and clearer error-containment economics. Reassess if NOW/PANW disclose slowing AI-related expansion or if DOCU/ADBE demonstrate sustained paid usage and net-retention acceleration.
• Ahead of the next earnings cycle, use any AI-driven multiple expansion in document-centric SaaS as an opportunity to hedge with a small short DOCU versus long MSFT or NOW. Target a 10-15% relative move over 3-6 months; stop out if DOCU reports AI products lifting net retention or billings guidance by more than 2 percentage points.
• Do not add directional NVDA exposure on this signal. Instead, monitor hyperscaler capex guidance and enterprise production-deployment commentary over the next two quarters; a shift from pilots to governed workflows would support infrastructure demand, while broad capex moderation would falsify that constructive read-through.
• Add an earnings-call watch item for AI ROI disclosure: paid active users, inference cost per customer, renewal uplift, hours saved measured at workflow level, and professional-services requirements. Companies unable to quantify at least one of these metrics warrant lower position sizing despite favorable AI narratives.