When scaling AI becomes the real challenge
Most organizations know how to launch an AI pilot. Far fewer know how to transition from controlled experimentation to enterprise-wide adoption.
That transition introduces a new set of questions. How should access be managed? What guardrails are required? How do you balance innovation with governance? How do you maintain visibility as adoption expands across teams and business functions? And how do you ensure that cost, security, and compliance considerations evolve alongside usage rather than becoming obstacles later?
These were the same questions Fractal faced as interest in Anthropic Claude expanded across the organization. The challenge quickly became larger than providing access to a new AI tool. The focus shifted to establishing the governance, security controls, cost management practices, and operational visibility needed to support AI adoption at enterprise scale.
What emerged from that journey was not simply a Claude rollout, but a practical operating model for managing how the organization requests, provisions, governs, monitors, and adopts AI. The experience provided valuable lessons on balancing accessibility with accountability and helped shape Fractal’s approach to operationalizing AI in a way that could grow with business demand
The demand for Claude was already there. The challenge was scaling access without losing sight of who owned each license or what it was costing. We put the right data controls in place, started with conservative consumption limits, and adjusted those limits as we saw how teams were using Claude.
Deepak Sharma,
Principal Architect, Fractal
Demand shaped the move to Claude Enterprise
Before moving to Claude Enterprise, the IT team reached out across business units to understand the nature and scale of demand. The assessment distinguished between employees who required Claude for everyday knowledge work and those who expected heavier usage for development and product-building activities. This bottom-up exercise informed the initial Enterprise allocation of 400 licenses and established business ownership before access was provisioned.
The expansion from Claude Teams to Claude Enterprise changed the nature of the rollout. The task was no longer limited to making a new tool available. The IT team needed an operating model that could support a growing user base while keeping access, security, consumption, and accountability connected to legitimate business requirements.
Governed access, with ownership built in
A Claude Enterprise request began with an IT helpdesk ticket. The employee provided a project code, approval from the project owner, and the expected period of use. After validation, the IT team provisioned access centrally and linked the allocation to its business owner. This made each license traceable to a requirement, approver, cost center, and defined usage period, rather than allowing access to remain open-ended.
Security operated at three distinct levels. Claude was blocked by default and made accessible only to approved users through enterprise access controls. Licenses were then assigned centrally to validated users. Endpoint-level Data Loss Prevention policies monitored uploads and could block content that did not meet compliance requirements. Separating access, entitlement, and data protection gave the rollout more structure than a single permission or login decision.
Cost controls that evolved with actual usage
Token-based consumption introduced a different management challenge because the IT team did not yet know how usage would vary across roles. Fractal therefore began with a standard monthly consumption limit of $30 per user. This was a ceiling rather than a prepaid allocation, so an employee who consumed $9 generated a cost of $9, not $30.
After observing usage for one or two months, the team found that developers and other heavy users sometimes required higher limits for project work. Those users could request an increase, with project-owner approval, allowing more demanding use cases to continue without removing financial accountability. The same principle was applied at the organization level, where the overall monthly ceiling was raised progressively as license allocation and actual consumption increased.
This created a practical feedback loop: begin with a defensible baseline, observe real demand, and make exceptions visible and approved. The approach avoided assuming that every user would consume the same amount while retaining a clear mechanism for users whose work required more capacity.
Enablement was part of provisioning
New users received practical guidance on signing in, monitoring usage, selecting an appropriate model, understanding spending limits, and adopting credit-efficient habits. They were also given clear dos and don’ts for responsible use. This meant that access was accompanied by guidance on how to use Claude effectively within the operating model.
From product analytics to an operating view
Native administration analytics answered important questions about product usage, but Fractal needed additional business context. The team wanted to understand who owned each allocation, which project funded it, how consumption was distributed across the organization, and whether actual spend was aligned with budget.
To address that need, the IT team developed AI Hub, an internal platform built with support from Claude Code. AI Hub combined Claude administration data with internal systems and selected manually maintained inputs, bringing together consumption, license type, project mapping, approval history, business-unit allocation, and spend-versus-budget views. Some information was available more quickly, while other fields refreshed daily, reflecting an integration that was still being refined at the time of the discussion.
The platform also helped the team identify highly engaged users and understand which Claude surfaces, connectors, and activities were gaining traction. Those signals created a basis for conversations with power users and for identifying working practices that could potentially be reused more broadly.
Note: The dashboard data has been anonymized and is shown for illustrative purposes.
The next step: governed organizational skills
As AI adoption matures, the focus naturally shifts from individual productivity to sharing successful practices across the organization. One area Fractal is exploring is how useful Claude Skills developed by practitioners can be made available more broadly, helping teams benefit from proven workflows and expertise.
To support this, Fractal is establishing an approach for evaluating and governing shared Claude Skills. Practitioner-created skills can be reviewed by relevant subject matter experts, assessed for security and intellectual property considerations, and then made available for wider use where appropriate. The goal is to scale successful workflows while maintaining quality, governance, and consistency across the organization.
This next phase of the journey is focused on turning individual knowledge and experimentation into reusable organizational capabilities that can accelerate adoption and create greater value at scale.
Practical lessons for enterprise AI leaders
Let business demand shape the first scaling decision. Fractal’s initial Enterprise allocation followed a direct assessment of standard and developer needs across the business.
Separate access, entitlement, and data protection. Distinct controls provide more clarity than relying on a single permission layer.
Treat consumption limits as adjustable guardrails. A baseline can provide control, while approved exceptions support heavier project requirements.
Connect product usage with business context. AI Hub brought ownership, projects, approvals, consumption, and budgets into one operating view.
Make enablement part of provisioning. Practical guidance should accompany access rather than arrive later as a separate adoption effort.
From internal rollout to practical client experience
Fractal’s Claude journey shows that enterprise AI adoption requires coordinated decisions across governance, security, cost management, operational visibility, and user enablement. It also shows why the operating model needs to evolve as demand and usage patterns become clearer.
Organizations facing similar questions can draw on Fractal’s firsthand experience of moving Claude from a limited deployment to a governed enterprise rollout. Fractal can help bring together the technology, operating model, governance, and change considerations needed to move enterprise AI from pilot to practice.
