The future of AI is intelligent workflows

AI is delivering productivity gains for individuals, but organisations are still struggling to turn them into enterprise-wide results. McKinsey's 2026 State of AI survey found 80% of respondents say AI improved their own productivity, yet just 37% report any enterprise-level EBIT impact. In the UK, 75% say AI makes them more productive, but only 13% say it has significantly improved their organisation's performance.

This shows that models aren't the problem. The problem is where they sit: in a chat window, rather than in the flow of work. At OneAdvanced, we believe the next phase of enterprise AI will be shaped by intelligent workflows, not by the highest-ranked model. This article explains why models are becoming a commodity, what intelligent workflows are, how it differs from an AI agent, and how to move from scattered pilots to AI that is embedded, governed and measurable.

Why is the AI model becoming a commodity?

For the past two years, choosing an AI model has been treated as a strategic decision: GPT versus Claude versus Gemini. But as model capabilities converge, the question is becoming less about which model you choose and more about how you use it.

Stanford’s 2026 AI Index found that the gap between the best US and Chinese models had narrowed to just 2.7% by March 2026, down from double-digit gaps in 2023, with the lead changing hands repeatedly since early 2025. As UNU’s analysis notes, competition is shifting from raw capability towards cost and reliability.

Two things follow. First, model advantage is short-lived. Today’s standout capability can quickly become tomorrow’s baseline, available to competitors through the same API. Secondly, the model is becoming a swappable component.

We have seen this first-hand. OneAdvanced AI launched on a self-hosted, open-weight model inside a UK-sovereign environment. In a pilot with NVIDIA, a compact model trained on pseudonymised NHS online consultation requests outperformed leading frontier models in benchmark testing, at up to 150 times lower inference cost.

The lesson is not that small models beat large ones. It’s that context, data and workflow fit matter more than headline model size. Organisations that hard-wire themselves to one model can take on unnecessary risk without creating lasting advantage.

What is an intelligent workflow?

An intelligent workflow is a business process automation solution that connects people, processes, data and AI to optimise critical activities. Traditional automation follows fixed rules and stopes at the first exception. A chatbot answers questions but owns no process. But an intelligent workflow brings these capabilities together: applies rules where certainty is required, AI where judgement is needed, and people where accountability matters. That balance is what makes the workflow intelligent: flexible, controlled, and enterprise-ready.

What makes a workflow "intelligent"?

Not every automated process is intelligent. A genuinely intelligent workflow has four key characteristics:

Context-awareness

An intelligent workflow understands who is asking, which sector or policy applies, and what has happened earlier in the process, without requiring people to re-explain it at every step. This lets the same capability work appropriately across a housing association, law firm or NHS trust without separate builds.

Decision logic and guardrails

Rules and judgement should be explicit. Where decisions must be exact, such as payroll calculations, regulatory thresholds or eligibility checks, the workflow should be deterministic. Where reasoning is needed to interpret unstructured information, agentic AI can step in. Intelligent workflows make this distinction deliberately.

System integration

A workflow becomes more powerful when it can see and act across the systems that hold the data, from ERP software and case management to HR platforms. This reduces manual data entry and stops people from becoming the integration layer between systems.

Feedback loops

Finally, an intelligent workflow improves from its outcomes. Every exception handled and correctionsmade by a human reviewer should feed back into how the workflow performs next time. Without this loop, AI workflow orchestration is little more than automation using a language model.

Note: Governance runs through all four: permissions, audit trails and human oversight are designed in, not added after the pilot.

AI agents vs Intelligent workflows: What’s the difference?

AI agents work independently towards a goal, while traditional workflows follow predefined rules. Intelligent workflows bring the two together, using AI where judgement adds value, and rules and controls where consistency matters. Here’s how they are different:

Dimension	Standalone AI Agent	Intelligent Workflow (embedded)
How it works	Plans its own steps and chooses its own tools to reach a goal	Orchestrates AI, rules, data and people through governed paths
Autonomy	High: decides how to act	Calibrated: AI where it is safe, rules where required
Reliability	Can take different routes on the same input	Consistent outcomes with defined checkpoints
Governance and audit	Must be added around it	Built in: permissions, logs, approvals
Human oversight	Often after the fact	Designed in at set decision points
Best for	Open-ended, exploratory tasks	Repeatable, regulated, high-volume processes
Main risk	Unpredictable behaviour and cost	Needs upfront process design

 Explore our AI agents for business services and see the same approach in legal and government.

Why standalone AI pilots underdeliver

Enterprise AI pilots rarely fail because the model is weak. They fail because they focus on the wrong outcome, sit outside the wider workflow or lack clear ownership.

Pilots are scoped around tasks, not outcomes

A pilot that automates “drafting a reply” or “summarising a document” may be easy to demonstrate, but difficult to attribute to a business result. This is because the task is only one part of a wider process that the pilot doesn’t address. This scoping problem is a major reason why 80% of AI agent pilots never make it to production, according to RAND Corporation research.

Point solutions don't talk to each other

Many organisations now have multiple AI tools across disconnected processes, from a summarisation tool and a chatbot to agents embedded in CRM software. Without shared context or governance policies, these tools don’t add up to a connected system. And that’s one of the reasons why more than 40% of agentic AI projects would be cancelled by 2027, largely because of governance issues rather than technical limitations.

No ownership of results

Analysis of DSIT, ONS and industry benchmarks found that only around 7% of UK organisations are pursuing a genuinely enterprise-wide AI strategy. Everyone else is running pilots that sit under individual teams, with individual budgets, and no single owner accountable for the outcome. Embedding AI into an existing workflow creates a clearer path to value because someone already owns the process and its outcome.

Explore this approach in our earlier piece: AI in the enterprise: Build, buy or embed? A practical guide for business leaders

Where the real competitive advantage lives?

If the model is becoming a commodity, competitive advantage shifts to what surrounds it: workflow, governance and trust.

Workflow ownership creates proprietary context

Every AI system is only as good as the context it receives. Organisations that own their workflows control which data matters, which exceptions are common and what “good” looks like for a specific team or sector. That context cannot be replicated simply by licensing the same underlying model. It builds over time, workflow by workflow, making the system more valuable as it learns from real-world use.

Governance is a feature, not a constraint

This is where private, sovereign AI matters. Governance should not be a separate compliance layer added after deployment. It should be built into how the workflow operates. We built IQ, our intelligent system of work, around this principle, as the first system of work to apply ISO 42001-aligned policy directly alongside AI "in the flow of work”.

Trust compounds over Time

The final advantage is trust. A workflow that consistently produces accurate, explainable, on-policy outcomes can take on more autonomy over time. For business leaders the ask is simple: start with a high-value workflow, define the controls and outcomes, measure performance, and expand autonomy only when the evidence supports it. That is how AI moves from an experiment to a trusted part of the business.

How to move from isolated AI tools to embedded intelligent workflows?

Before you start, settle the sourcing question. Whether you build, buy or embed AI decides who owns the workflow, and therefore who captures the learning. Then work through this checklist:

• Start with an outcome, not a model: Choose one to three high-volume, high-friction processes, such as procure-to-pay, hire-to-retire or a patient or client pathway and define the result you want to improve.
• Map the workflow as it really runs: Document handoffs, duplicate data entry, workarounds and exceptions. Fix the process before you automate it.
• Connect the data and the systems: Give AI a shared data layer across ERP, HR, finance and sector systems so it works from context, not fragments.
• Separate the deterministic from the probabilistic: Use rules for compliance-critical steps and AI for judgement and unstructured content. Set confidence thresholds for human review.
• Build governance in from day one: Define permissions, audit trails, approval points and policy alignment (ISO 42001 is a useful reference) before go-live, not after.
• Embed AI where people already work: Adoption follows convenience. Put intelligence inside the tools teams use every day, not in a separate app.
• Measure outcomes and keep a learning loop: Track cycle time, error and exception rates and cost per transaction. Review monthly and be ready to swap models when a better or cheaper one arrives.

 Common mistakes to avoid

• Chasing benchmarks. A model that leads in the current month will be matched next quarter. Optimise for fit, cost and reliability inside your workflow.
• Automating without governance. Speed without control compounds errors at machine speed. IBM's 2026 Cost of a Data Breach Report found that 68% of breached organisations lacked policies to manage AI or detect shadow AI, and 92% of those that suffered an AI-related breach lacked adequate AI access controls.
• Automating a broken process. AI applied to a poor process delivers poor outcomes faster. Redesign first.
• Ignoring change management and measurement. Teams need to trust what the system decided, understand when and why it escalated to a human, and see evidence that oversight is real rather than theoretical. Skipping this step is the fastest way to turn a well-built workflow into one nobody actually uses.

How does OneAdvanced build intelligent workflows?

We built IQ, our intelligent system of work, on this premise. It brings processes, data, policies and AI together on a sovereign, UK-hosted platform. Its connected ethosunifies workflows, teams and data; Trusted secures them; Intelligent embeds AI-driven insight and automation directly in the flow of work.

Intelligent Workflows are one of IQ’s four elements, alongside Intelligent Experience, Intelligent Platform and Intelligent Services. It covers sector-focused workflows across finance, spend and governance and people and workforce management, with AI embedded throughout. Sector-specific agents for areas such as clinical coding, complaints handling, risk management and shift assignment work within these workflows rather than alongside them.

Because IQ is model-agnostic, the workflow and data layer persist even as models change, so you can adopt a better or cost-effective model without rebuilding the process. Sovereignty extends to the model itself: our approach to private, sovereign AI keeps data within UK jurisdiction and out of external model training.

Frequently Asked Questions

Why do many AI pilots fail to deliver ROI?

Most pilots are scoped around a single task rather than an outcome, run in isolation from other systems, and lack a single accountable owner. Forrester found that 88% of AI agent pilots never reach production, and Gartner expects over 40% of agentic AI projects to be at risk of cancellation by 2027.

How does workflow ownership create competitive advantage?

Owning the workflow means owning the proprietary context and feedback that flow through it, such as the specific exceptions, judgement calls and quality standards of a particular team or sector. That context can't be replicated by licensing the same underlying model.

What role does governance play in AI workflows?

Governance built into the workflow itself, rather than layered on afterwards, is what allows AI to be trusted with more autonomy over time. Without it, errors can compound at machine speed across every case a workflow touches.

How does OneAdvanced embed AI into existing business workflows?

IQ, our intelligent system of work, applies ISO 42001-aligned governance directly alongside AI in the flow of work, combining agentic capability with deterministic, auditable steps across finance, HR and sector-specific processes.

Can intelligent workflows work across finance, HR and operations?

Yes. The same principles apply across functions: deterministic accuracy where it's required, adaptive reasoning where it adds value, and shared governance throughout, whether the process sits in finance, people management or a sector-specific operation.