AI Requires a New Operating Model

By Javan Ward & Kai Hoyt

When we first scoped an internal company brain for a late-stage digital health company, the business had already invested in AI. Gemini and Claude licenses had been distributed across departments, and employees were using them to make individual tasks faster. Yet the client success operation still worked much as it had before.

Account information was divided across the CRM, Google Docs, call notes, and the working knowledge of individual representatives. When leaders needed to understand churn risk, review the health of an account, or report across the book of business, someone had to find the relevant information and assemble it by hand. Consequential decisions were made from a picture reconstructed across systems and conversations rather than from an operating view the company could maintain.

The cost appeared throughout the operation. Leaders lost time assembling reports. Churn signals surfaced late. Decisions varied depending on who held the context, and adding headcount began to feel like the only way to keep up with a growing book of business.

The company needed a shared operating context that its existing AI licenses could use. The models could summarize a document or draft a report, but they had access to only a fraction of what the company knew. They could not reliably connect what customers were saying, what representatives were observing, what the CRM recorded, and what the organization had learned over the course of each relationship.

We traced how account information entered the company, where decisions stalled, which exceptions required judgment, and what the system failed to retain. The design would assemble approved CRM activity, call notes, account documents, and customer signals into a traceable account view while leaving the source systems authoritative. A new customer signal could then appear in the next churn review alongside the relevant account context and reach the person authorized to act.

The work remained at the design stage, so we cannot claim that the system worked in practice. It exposed the limit of license-led adoption: a company can buy AI for every employee without changing how the company operates.

Why faster work is not enough

Across industries, companies are approving point solutions and distributing AI tools across their teams. Emails are drafted faster. Meetings are summarized. Reports take less time to prepare. These gains matter, but many leaders still struggle to connect them to a meaningful change in company performance.

Most companies are introducing AI into an operating model built before AI could participate in the work. Information remains scattered across systems. Workflows depend on manual handoffs. Important decisions require someone to reconstruct the context. Each tool accelerates a piece of the process while the company continues to coordinate the whole process in the same way.

This is understandable. For decades, adopting technology has meant selecting software, buying licenses, training employees, and adding another application to the stack. AI exposes the limit of that approach because its usefulness depends on the context it can reach, the work it is allowed to perform, and the feedback it receives from what happens next.

Research on the Productivity J-Curve explains why returns from general-purpose technologies can lag investment. Companies absorb costs while building the processes, skills, organizational practices, and business models the technology requires.

AI can reduce the time required to find information, assemble context, monitor conditions, prepare decisions, route exceptions, document results, and carry out approved actions. Whether those task-level gains improve company performance depends on suitable work, reliable context, adoption, error controls, and measurement against a baseline.

The operating model determines whether faster tasks become better company performance.

Building around intelligence

An AI-native operating model begins with the operation. The company must make its knowledge accessible, establish which systems remain authoritative, define where people retain decision authority, and redesign workflows around a clear division of work between people and agents.

People continue to own purpose, relationships, exceptions, and consequential decisions. Agents can assemble context, prepare decisions, monitor changing conditions, perform approved actions within explicit limits, and escalate what requires human attention. Permissions, traceable evidence, evaluation thresholds, and named owners keep the work accountable.

When an employee corrects an output, resolves an exception, or makes a decision, the system should retain that evidence for review. Approved changes can then improve the context, rules, or evaluations used in the next cycle.

We call the architecture behind this model Cognitive Architecture: the structure, memory, authority, and feedback that allow people and agents to work from shared context and improve from operating evidence.

Early enterprise research points in the same direction. In a 2025 MIT CISR survey of 132 organizations, major workflow redesign was associated with stronger reported results from AI-based digital colleagues. About 22 percent of respondents reported redesign at that level. Redefined roles, performance measures, and higher levels of use were also associated with stronger results.

What leaders should see

For a CEO or COO, the test is whether the change becomes visible in the weekly rhythm of the selected operation. Leadership gains a current, traceable view of the accounts covered by the system, including emerging risks, unresolved exceptions, source conflicts, and the evidence behind each finding.

Decisions can move sooner because the relevant context arrives with them. Teams spend less time requesting information, tracking down approvals, or waiting for the person who knows how the process works. Meetings require less time to reconstruct status, while consequential exceptions still reach the named owner.

The first measures should follow the workflow. In client success, that could include account-review preparation time, portfolio coverage, time from a risk signal to owner action, unresolved exceptions, and accounts managed per representative without reducing service quality.

Those gains create room for the work people do best: building relationships, supporting colleagues, strengthening partnerships, exercising judgment, and finding new ways to create value.

Where to begin

Begin with one operation where missing context, repeated work, slow decisions, or dependence on a key person already constrains growth. Establish the baseline and define the acceptance criteria. Redesign the division of work between people and agents, then run one bounded capability beside the current operation.

Expand only when the evidence shows what improved, what failed, and what the company can safely carry forward.