AI can’t fix a business system with broken processes

The article argues that AI cannot fix businesses with fragmented, inefficient underlying processes, despite its ability to accelerate operations, automate repetitive tasks and identify problems quickly. The core investment implication is that AI spending may produce weak returns unless companies first streamline operational workflows.

Analysis

This is not an AI-demand signal; it is a warning that enterprise AI spend will migrate from broadly marketed copilots toward vendors embedded in systems-of-record and process redesign. The near-term revenue risk sits with application software names carrying high AI monetization expectations but limited evidence of workflow-level ROI; seat-based add-ons are vulnerable to budget scrutiny if customers have not standardized data, permissions, and operating processes. ServiceNow (NOW), Salesforce (CRM), SAP (SAP), and Microsoft (MSFT) are relatively better positioned because implementation friction creates demand for orchestration, governance, integration, and consulting rather than merely model access.

Over the next 1-3 quarters, the key transmission mechanism is a longer sales cycle and lower attach rate for discretionary generative-AI modules, not an abrupt cut in total IT budgets. Accenture (ACN), Cognizant (CTSH), and IBM (IBM) can capture remediation and implementation work, but margin upside is uncertain: labor-intensive process mapping can dilute margins before reusable AI-enabled delivery offsets it. The more durable 6-18 month implication is consolidation around platforms with proprietary workflow data and credible measurement of cycle-time, error-rate, and headcount outcomes; firms selling horizontal AI features without integration moats face multiple compression.

Consensus remains too focused on model quality and inference cost. The binding constraint for many enterprises is change management and data/process ownership, which favors incumbents but also means AI revenue recognition may lag investor expectations. Falsification would be broad evidence in upcoming earnings calls of accelerating paid AI-module adoption accompanied by measurable retention or margin uplift, particularly at CRM and NOW; absent that evidence, premium AI software multiples remain exposed.

Key Decisions for Investors

• Maintain a quality bias within enterprise software: long NOW versus short a basket of higher-multiple, weaker-workflow AI beneficiaries through the next two earnings cycles. Use a 10-15% relative adverse-move stop; the thesis works if NOW sustains subscription growth while peers fail to show AI attach-rate acceleration.
• Watch CRM and SAP for post-earnings entry rather than chase AI narratives pre-results. Initiate only if management quantifies paid AI adoption, net-revenue-retention uplift, or implementation backlog; without those disclosures, treat AI commentary as insufficient to underwrite multiple expansion.
• Consider a 3-6 month tactical long ACN or IBM only on evidence that AI transformation bookings are converting to revenue faster than traditional consulting displacement. The upside is backlog-driven estimates revisions; principal risk is utilization pressure and lower-margin implementation mix.
• Avoid broad long exposure to AI application software ETFs as a proxy for enterprise AI monetization over the next 1-3 months. Prefer MSFT as the liquid large-cap beneficiary of governance and integration demand, but reassess if Azure AI growth decelerates without offsetting Copilot monetization.