Scale AI Hired Google Cloud’s COO: Is Enterprise AI Consulting Becoming the Main Battleground?

Scale AI’s appointment of former Google Cloud COO Francis deSouza suggests enterprise AI competition is moving beyond models and infrastructure into implementation. For Sydney and NSW organisations, the important battleground is increasingly who can connect AI to real workflows, data, security, approvals and measurable business outcomes. Consulting is becoming less about strategy decks and more about taking operational responsibility for getting AI into production.

Scale AI’s choice of chief executive looks, at first, like another high-profile movement of technology leadership. It is more significant than that.

Francis deSouza became Scale AI’s chief executive on 10 August 2026 after serving as Chief Operating Officer and President of Security Products at Google Cloud. Scale says the appointment comes as it expands both its data and applications businesses, with enterprise customers including BP and Mayo Clinic alongside substantial government work.

The appointment matters because Scale was built around one of the foundational inputs of modern artificial intelligence: the data, human evaluation and infrastructure required to improve advanced models. Its current positioning is much broader. Scale now describes itself as helping enterprises and governments build, deploy and oversee AI applications, while its customer-facing engineering roles explicitly involve integrating production AI into existing enterprise systems.

That shift points towards a larger competitive question. As increasingly capable models become accessible from several providers, does the most valuable commercial territory move away from the model itself and towards the difficult work of making AI function inside an organisation?

Increasingly, the answer appears to be yes.

The New Contest Is Not Simply About Who Has The Best Model

Enterprise buyers have spent much of the generative AI cycle comparing model quality, token costs, context windows, reasoning capability and cloud infrastructure.

Those differences remain important. They are no longer the whole procurement decision.

A capable model does not automatically know how a Sydney construction business approves variations, how a property operator manages access instructions, how a professional-services firm separates privileged material, or which employee is authorised to release a particular document.

The underlying AI may be sophisticated while the deployment around it remains commercially useless.

This creates an implementation gap between what models can theoretically do and what organisations can safely allow them to do.

Closing that gap requires work that looks increasingly like a combination of consulting, systems integration, software engineering, data architecture, security, governance and operational redesign.

Elyment has previously examined why AI consulting advice should not be confused with an implementation plan and how enterprise AI agent platforms are entering operational workflows.

The emerging issue is different. The question is now who owns the space between strategic ambition and production delivery.

Enterprise AI Consulting Is Starting To Look Like Productised Engineering

Traditional management consulting often begins with diagnosis. Consultants map the organisation, identify opportunities, recommend a future operating model and produce a transformation roadmap.

Enterprise AI increasingly requires the adviser to go further.

Someone has to connect the model to the document repository. Someone has to determine which CRM records can be retrieved. Someone has to build the approval logic, restrict credentials, evaluate output quality, monitor costs and define what happens when the AI is uncertain.

Scale’s own recruitment provides a useful indication of this delivery model. Its forward-deployed engineering roles involve working directly with enterprise customers to architect, integrate, deploy and operate production AI systems across cloud platforms, data warehouses, internal APIs and business applications.

That is not conventional software licensing.

It is also not conventional consulting.

It is a hybrid model in which the supplier sells technology while embedding technical personnel deeply enough into the customer environment to make the technology operational.

Why The Consulting Layer Could Become More Valuable As Models Improve

Better AI models can paradoxically make implementation expertise more important rather than less important.

When a model can perform only narrow tasks, the organisation has relatively few places where it can be deployed. As capability increases, the possible use cases multiply.

A business might move from experimenting with document summaries to considering:

• Customer enquiry triage
• Quote preparation
• Contract and specification review
• Project scheduling support
• Supplier communication
• Document classification
• Compliance evidence preparation
• CRM updates
• Financial exception detection
• Internal knowledge retrieval
• AI agents capable of taking defined actions across several systems

Every additional use case introduces operational dependencies.

The model may be the intelligence layer, but value depends on the infrastructure surrounding it.

The Real Enterprise AI Stack Is Larger Than The Software Diagram

• Workstream: Use-case selection
• What the business sees: “Where should we use AI?”
• What usually has to be solved: Process economics, risk, workflow volume, exception rates and measurable outcomes
• Workstream: Data
• What the business sees: “Connect our information”
• What usually has to be solved: Permissions, data quality, duplication, retention, classification and retrieval design
• Workstream: Integration
• What the business sees: “Connect the CRM and project systems”
• What usually has to be solved: APIs, authentication, field mapping, system ownership, retries and failure handling
• Workstream: AI behaviour
• What the business sees: “Make the agent reliable”
• What usually has to be solved: Instructions, context, evaluations, tool restrictions, escalation logic and testing
• Workstream: Security
• What the business sees: “Keep the data safe”
• What usually has to be solved: Identity, access controls, credential management, logging and third-party risk
• Workstream: Human authority
• What the business sees: “Automate the workflow”
• What usually has to be solved: Approval thresholds, accountability, exception management and intervention rights
• Workstream: Adoption
• What the business sees: “Roll it out to staff”
• What usually has to be solved: Training, workflow change, user trust, ownership and operating procedures
• Workstream: Commercial control
• What the business sees: “Reduce our costs”
• What usually has to be solved: Usage monitoring, model routing, support costs and cost per completed outcome

This is why the competitive advantage of an enterprise AI supplier may increasingly depend on how much of this stack it can own.

Four Different Industries Are Converging On The Same Enterprise Budget

Enterprise AI creates an unusual competitive market because several categories of supplier can plausibly claim ownership of the same programme.

Cloud Providers

Cloud companies already control infrastructure, identity systems, databases, security products and significant enterprise relationships. They can extend from infrastructure into AI applications and implementation.

AI Model And Platform Companies

AI companies begin closer to the underlying intelligence. Their opportunity is to move further into applications, evaluations, orchestration and customer deployment.

Global Consulting And Systems-Integration Firms

Traditional consultancies understand large transformation programmes, procurement, change management and executive decision-making. Their challenge is maintaining sufficient technical depth as AI architectures change quickly.

Specialist AI Operators

Smaller specialists can compete through speed, domain knowledge and close implementation support, particularly where a customer needs a working operating system rather than a multi-year transformation programme.

Scale’s strategic opportunity sits at the intersection. It already has technical infrastructure, model-evaluation capability and increasingly customer-facing engineering. Hiring an executive with enterprise cloud, security and large-scale commercial experience strengthens the part of the organisation that must sell and deliver into complex institutional environments.

What This Means For Sydney Businesses

Sydney organisations do not need to mirror the procurement structures of Fortune 100 companies for the same trend to matter.

The implementation problem appears at a smaller scale whenever an organisation attempts to move AI beyond an employee using a standalone chatbot.

Consider a property or renovation operator that wants an AI workflow to review new enquiries.

The prototype might appear simple. Read the customer's message, identify the requested work and prepare the next response.

Production delivery is more complicated.

1. Identify the property and work type.
2. The system may need to distinguish carpet removal, concrete grinding, floor levelling, demolition, painting or another service.
3. Determine what information is missing.
4. Floor area, access, strata conditions, existing substrate, waste movements and preferred timing may all affect delivery.
5. Retrieve the correct operating rules.
6. The workflow needs current service boundaries, escalation conditions and scheduling requirements.
7. Separate preparation from commercial authority.
8. An AI system may prepare information without being authorised to commit the business to a price, scope or completion date.
9. Write back to business systems.
10. CRM updates must occur in the correct record without creating duplicates or overwriting verified information.
11. Escalate exceptions.
12. Complex strata conditions, hazardous-material concerns or unclear project requirements may require human review.
13. Preserve accountability.
14. Staff still need to know what the AI did, which information it relied upon and who approved the final action.

The model is only one component of that operating sequence.

This is where a structured AI readiness assessment for Sydney businesses can be more useful than beginning with a vendor demonstration. It allows teams to identify the workflow, information sources, approval points and operational risks before deciding what technology should sit underneath them.

Governance Is Becoming Part Of The Implementation Product

Enterprise AI consulting can no longer treat governance as a legal appendix added after a system has been designed.

NSW Government agencies are required to apply the NSW AI Assessment Framework where applicable. The framework addresses AI across its lifecycle and places emphasis on accountability, privacy, security, transparency, fairness and risk assessment.

Private NSW businesses are not automatically subject to the NSW Government framework simply because they use AI. The principles are nevertheless useful indicators of the questions increasingly being asked during serious AI procurement.

At the Commonwealth level, the Office of the Australian Information Commissioner’s guidance on commercially available AI products makes clear that existing privacy obligations continue to apply where personal information is handled through AI systems.

The Australian Government’s Guidance for AI Adoption also places emphasis on human oversight, operational accountability and mechanisms for intervention throughout the AI lifecycle.

Cyber security adds another layer. Australian Signals Directorate guidance on agentic AI highlights the risks created when AI systems gain access to external tools, data sources, memory and action pathways.

For an enterprise buyer, this changes what a credible consulting engagement should contain.

The supplier should not only explain what AI can do. It should be able to explain:

• What information the system can access
• Where that information travels
• Which actions are permitted
• What credentials are exposed
• How outputs are evaluated
• When human review is required
• How failures are detected
• How changes are approved
• Who remains accountable after deployment

The Procurement Mistake Is Buying A Demonstration Instead Of An Operating Model

Enterprise AI demonstrations can compress a complicated future into a persuasive five-minute experience.

A document arrives. The AI reads it. A dashboard updates. An agent produces a recommendation. The workflow appears complete.

Production environments are less tidy.

Documents are incomplete. Customer names differ between systems. Permissions have accumulated over years. Employees use undocumented workarounds. APIs fail. Policies change. Managers make exceptions. Old records conflict with new ones.

The implementation partner is valuable precisely because it has to design for this operational mess rather than the demonstration.

Sydney organisations comparing vendors should therefore treat the decision to build, buy or combine AI systems as an operating-model decision rather than a feature comparison.

Seven Questions To Ask An Enterprise AI Partner

1. What business outcome will be measured?
2. Avoid projects where success is defined only as launching an AI tool.
3. Who owns workflow discovery?
4. Someone needs to understand how the work is actually completed before automating it.
5. Who owns integration after the prototype?
6. Determine whether the supplier stops at strategy, stops at software configuration or remains accountable through production deployment.
7. How will model performance be evaluated?
8. Production AI requires representative tests, not a handful of impressive examples.
9. What remains portable?
10. Understand whether business rules, data pipelines, evaluation sets and integrations can survive a future change of model or platform.
11. How are permissions and human approvals enforced?
12. Important controls should exist in system architecture, not merely in a prompt asking the model to behave appropriately.
13. What happens after launch?
14. AI systems require monitoring, evaluation, cost review, security updates and change management as underlying models and business processes evolve.

The Most Valuable AI Partner May Be The One Willing To Own The Last Mile

The phrase “AI consulting” risks understating what the market is becoming.

The winning supplier may not be the company capable of producing the best strategy presentation. It may be the organisation capable of moving from strategy through data, integration, controls, evaluation, deployment and operational support without losing accountability at each handover.

That is particularly important for businesses whose workflows cross physical operations, professional services, compliance requirements and customer communication.

Technology has to fit the organisation that actually exists.

AI IMPLEMENTATION & OPERATIONAL REVIEW

Review The Operating Model Before Choosing The AI Platform

Review workflows, integrations, data access, approval controls, privacy, security, implementation costs and delivery ownership before moving an enterprise AI project into production.

Request A Project Review

Is Enterprise AI Consulting Becoming The Main Battleground?

Scale AI’s leadership change does not prove that software platforms or frontier models have become commodities. Technical capability will continue to matter enormously.

It does, however, reinforce a change in where enterprise value is being contested.

Access to powerful AI is becoming only the beginning of the commercial relationship. The larger opportunity is helping organisations convert that capability into secure, integrated and economically defensible operations.

That means understanding the customer, selecting the right workflow, connecting fragmented information, building production systems, governing access, training staff, testing performance and accepting some responsibility for whether the project actually works.

Scale’s move towards enterprise applications and forward-deployed delivery is one example of that convergence.

For Sydney and NSW businesses, the implication is practical. The next enterprise AI procurement should not begin with the question, “Which model should we buy?”

It should begin with a harder question: “Who can take responsibility for turning this workflow into a reliable operating system?”

Increasingly, that is where the real competition in enterprise AI is moving.

Sources and References

• Elyment: Why AI Consulting Advice Should Not Be Confused With an Implementation Plan
• Elyment: Enterprise AI Agent Platforms in Sydney Business Operations
• Elyment: AI Readiness Assessment for Sydney Businesses
• NSW Government: NSW AI Assessment Framework
• Office of the Australian Information Commissioner: Privacy and Commercially Available AI Products
• Australian Government: Guidance for AI Adoption
• Elyment: Build vs Buy AI Sydney