AI platform consolidation starts with shared operating rules

AI platform consolidation starts with operating rules

Buying a shared AI platform leaves a management question open: how should teams develop and govern AI? A shared AI standard, the organization’s agreed rules for development, deployment and oversight, gives that purchase a purpose. For chief information officers (CIOs) concerned that a growing collection of AI platforms is weakening investment returns, the first question is what the business needs everyone to agree on.

Zoldi recommends setting that standard before putting it into practice through a unified platform, backed by coordinated leadership and disciplined spending. FICO has a commercial interest in this recommendation because it sells AI platform technology. The management question is how a platform will carry out the organization’s decisions, since choosing technology cannot settle disagreements about how AI should be used or controlled.

Reported AI returns leave room to improve operating costs

The FICO report gives leaders a starting point for that spending discussion: most of the reported outcomes cited here met or exceeded expectations. Only 5% of organizations surveyed said AI significantly exceeded their return on investment, or ROI, expectations. More than 37% said AI met their expectations, while roughly 43% said it somewhat exceeded them. So improving an already useful investment is a relevant management goal.

Useful results can still come with avoidable costs, because an organization may meet its business goals while paying for activity that contributes little to them. Cutting that spending improves the economics of AI work even when leadership is satisfied with the overall return. Zoldi’s recommendation addresses this question: how can shared rules and coordinated deployment control activity that raises costs without adding business value?

Define the standard before choosing the platform

To control operating costs through a shared platform, the organization first needs to define the work it will support. Zoldi says the standard should specify which algorithms, the computational methods used by AI systems, the organization plans to use. Deciding this before platform selection gives teams an agreed development approach against which to assess the available technology.

That approach also needs shared practices for explainability and monitoring, both of which Zoldi includes in the standard. Explainability means explaining an AI system’s behavior or outputs; monitoring means observing its operation over time. Agreed practices give teams a common basis for explaining results and tracking system behavior after deployment.

These practices must also account for regulatory requirements. Zoldi says businesses should describe how they will meet those requirements within the standard, linking compliance planning to development and deployment. Once that approach is defined, leadership can assess whether a proposed platform supports it.

A unified platform then becomes the means of putting the agreement into practice. “That’s when we see the ROI developing,” Zoldi said. He ties returns to implementing a shared standard, so agreement on how AI should operate underpins his platform recommendation.

Leadership must turn the standard into purchasing decisions

Putting the standard into practice requires leadership support, so Zoldi advises CIOs to promote it and build agreement among leaders. CIOs need to explain how shared decisions about development, oversight and compliance serve the business. With that agreement, technology selection has a clear purpose, and teams have a common basis for using the chosen platform.

Leadership agreement also supports purchasing discipline. Zoldi calls for CIOs to be frugal and responsible when buying technology, since additional purchases can contribute to platform sprawl, the growth of separate AI platforms across an organization. For each proposed addition, leaders can ask which part of the agreed approach it supports and what business value it contributes.

Purchasing is part of a broader responsibility for infrastructure efficiency. Zoldi argues that CIOs should explain how coordination and efficient infrastructure contribute to company value, helping leaders assess spending beyond the appeal of individual tools.

Spending control continues after deployment

The responsibility continues after deployment, because costs depend on what teams keep running and consuming. “Very often, that sprawl means shadow projects that are being done, things that are left on that raise your costs along with hyperscaler token use that goes to no business value whatsoever,” Zoldi said. His examples cover projects outside shared oversight, resources left active and usage-based charges. Each needs attention in everyday operation.

Shadow projects, work undertaken outside the organization’s shared oversight, make coordination harder because leaders may lack a common view of AI activity. Bringing them into agreed development and governance practices gives the organization a basis for assessing their purpose and costs. Resources left running create a related problem: charges can continue even when keeping them available contributes little value.

Usage-based charges give leaders another reason to examine ongoing activity. Hyperscalers are large cloud service providers, and tokens are the units of text that language models process. Zoldi identifies hyperscaler token consumption that produces no business value, so his concern extends to whether paid usage contributes to useful outcomes. A tidy list of purchased platforms can still leave these operating costs unresolved.

Agent governance adds security and data requirements

Continuing oversight becomes broader with agentic AI, meaning systems that carry out tasks through sequences of actions. Around 40% of organizations in the FICO report cited security concerns as a barrier to achieving returns on agentic AI. Roughly 38% cited data integration and quality challenges as the biggest barrier. These concerns add security and data requirements to leaders’ operating decisions.

Vendors competing for enterprise AI spending are building products around those governance needs. Salesforce and Databricks launched AI agent governance features “earlier this year,” while AWS introduced Agent Registry, a platform intended to support building and governing agents. These companies have a commercial interest in making governance part of the technology purchase, just as FICO benefits from demand for shared AI platforms.

For buyers, the task is to assess how each offering supports the organization’s agreed governance requirements. Security requires decisions about what agents may access and do, while data integration and quality affect the information available for their work. A shared platform can support these responsibilities. But leadership and technical teams still need to define the requirements against which they will assess its capabilities.

Key highlights

• Set operating rules first: CIOs need leadership agreement on AI development, deployment and oversight before consolidating platforms. That agreement gives technology selection a clear business purpose.
• Review costs alongside returns: Most surveyed organizations reported AI returns that met or exceeded expectations. Finance and technology leaders can improve those economics by identifying spending that contributes little business value.
• Define platform requirements: AI development and governance teams need agreed algorithms, explainability practices, monitoring and regulatory requirements. Use those standards to assess whether a proposed platform supports the organization’s needs.
• Tie purchases to shared priorities: CIOs need to connect each proposed technology purchase to the agreed AI standard and its business value. Leadership support gives purchasing teams a common basis for controlling platform sprawl.
• Track spending after deployment: Infrastructure and AI teams need ongoing reviews of shadow projects, resources left running and token consumption. Assess whether each activity justifies its continuing cost.
• Specify agent governance requirements: Security concerns and data integration and quality challenges are reported barriers to agentic AI returns. Security and data teams need to define agent access, permitted actions and data requirements before buyers assess governance products.