Enterprise AI strategy is often framed as a software question: which model, which copilot, which vendor, and whether to build or buy. That matters, but it is not the hard part.
The harder work is organizational. Companies already run on a dense patchwork of systems of record, vertical SaaS products, scripts, spreadsheets, shared drives, emails and manual workarounds. AI can make it cheaper to create a tool—or in some cases perform a task directly—but it does not automatically reveal which process deserves redesign, what the new process should be, or how to make hundreds of people use it consistently.
Software exists because work is messy
A large company may have hundreds or thousands of applications not simply because it buys too much software, but because its work has accumulated exceptions, controls, handoffs and specialized needs. Some workflows belong in institutional systems such as SAP, Workday or a dedicated industry application. They require consistent data, permissions, auditability, maintenance and clear ownership.
Other work lives in more improvised spaces: spreadsheets, email, slide decks, CSV files and shared folders. Those tools persist because they handle edge cases and one-off questions that formal systems cannot accommodate easily.
The key transition occurs when an improvised workaround becomes recurring, important and widely shared. At that point, an organization has to “pave” the workflow: define it, secure it, govern it and assign accountability. That is one reason the enterprise software stack keeps expanding rather than collapsing.
AI changes both sides of this equation. It adds capability to existing applications and free-form tools, while creating a new general-purpose workspace in the chatbot. A small company may be able to keep a recruiting process in Google Sheets longer if AI makes the process more manageable. A team inside a large company may use an AI assistant to work around the inflexibility of a formal HCM system. Eventually, either pattern may become structured enough to warrant a dedicated product again.
The adoption gap is not a prompt-writing problem
Giving employees access to ChatGPT, Copilot or Claude is useful, but it should not be confused with transformation. In practice, usage will be uneven: a small group of employees may use these tools intensively, another group may use them occasionally, and many may struggle to connect a general-purpose assistant to the work in front of them.
That outcome resembles earlier technology transitions. Equipping every employee with a PC or a web browser created broad potential, but it did not itself redesign invoice processing, supply-chain management or e-commerce operations. Those changes required process owners, software choices, integration work and deliberate operating-model decisions.
For leaders, the practical implication is to separate broad access from targeted redesign. Broad access can build familiarity and surface grassroots use cases. It is not a substitute for changing high-value workflows that span teams, systems and regulatory requirements.
A more useful AI agenda
AI deployment should be managed across three distinct questions:
1. How should the company acquire and deploy the capability? Evaluate bundled options from existing platform vendors alongside specialist products and internal builds. 2. Which operations should change? Identify workflows with measurable pain, repeatable volume, clear owners and a realistic path to integrating data and controls. 3. What changes in the market? Consider whether AI alters unit economics, customer expectations, competitive differentiation or the viability of the current business model.
Pilots remain valuable, particularly for testing whether a workflow can be automated safely and whether employees will adopt the new process. But a portfolio of isolated pilots is not a transformation plan. Leaders need a mechanism for turning successful experiments into standardized, supported operations—and for deciding when experimentation should remain local.
What to watch next
The important signal will not be the number of AI licenses purchased or pilots announced. It will be whether companies can repeatedly discover hidden workflow bottlenecks, redesign them with the people who do the work, and institutionalize the changes when scale and risk demand it.
AI lowers the cost of trying new tools. It does not lower the cost of organizational clarity to zero. That makes workflow discovery, change management, governance and process ownership more—not less—central to the enterprise technology agenda.
