How to scale AI responsibly

Fewer than 1% of organisations have fully operationalised responsible AI. Governance is no longer an abstract ethics issue. The organisations that build responsible AI into how they work will scale faster, earn trust more consistently, and avoid preventable failures.

The SRAIS programme (Scaling Responsible AI Solutions), managed by CEIMIA under the GPAI/OECD framework across 44 countries, states it directly:

“In the race for AI adoption, those who prioritize responsible scaling and working collaboratively can gain a distinct competitive advantage.”

That is the operating premise of a 44-country programme. And the data from the past twelve months backs it up.

What “responsible scaling” actually means

Let’s be honest: the phrase gets used loosely. So here is what it means in practice, inside a marketing function. Responsible scaling is how your team actually uses AI as it grows. Does someone review the output before it ships? Does the team know which use cases need a human in the loop and which do not? Is there a shared understanding of what AI can and cannot do reliably? Are people trained, or are they figuring it out alone?

The structural side matters too: who owns accountability, how you assess quality, where the documentation lives. But the behaviour is what makes or breaks it. A governance policy that nobody follows is decoration. A team that checks, tests, and learns as it scales is responsible use in action.

The cost of scaling without it

The EY Global Responsible AI Pulse Survey (October 2025, 975 C-suite leaders, 21 countries) found that 99% of organisations have already suffered financial losses from AI-related risks. Nearly two-thirds of those losses exceeded $1 million. The conservative average: $4.4 million per organisation.

McKinsey’s November 2025 State of AI report (1,993 respondents, approximately 105 countries) adds the pattern: 51% of organisations using AI have experienced at least one negative consequence. Inaccuracy is the most commonly cited risk. And across six years of tracking, McKinsey notes that few AI risks are mitigated by most organisations.

Six years. The tools got better. The way organisations use them did not keep pace. Meanwhile, according to Gartner’s February 2026 analysis, only 27% of executives have a comprehensive AI strategy and just 20% believe their workforce is ready to execute it. The workforce-readiness number is the one that should keep marketing leaders up at night. Your team is already using AI. The question is whether they are using it well.

Responsible use delivers measurable results

A Gartner survey of 360 organisations (November 2025) broke this down into specific behaviours. Organisations performing regular AI system assessments are over three times more likely to achieve high GenAI business value. Those offering persona and role-based AI guidance are twice as likely to report higher value. Those providing GenAI ethics training, 1.7 times more likely. And organisations that expand AI rollouts beyond low-risk users are 3.3 times more likely to see returns.

Look at what is actually driving those numbers. Assessment. Training. Role-specific guidance. Confident expansion beyond the pilot group. These are not compliance activities. These are capability-building activities. From the same EY data: organisations with real-time AI monitoring are 34% more likely to see revenue growth improvements and 65% more likely to see cost savings.

What Mastercard built (and why it became a revenue asset)

Mastercard created an AI Governance Council in 2022 and built bias-testing frameworks and model documentation templates that now serve as customer-facing trust assets. When regulated banking clients ask whether a fraud detection model is fair, Mastercard hands over an auditable development trail.

The part worth noting: Mastercard’s banking customers drove the commercial value of governance. The documentation trail became proof of quality. The responsible use practices became a reason clients chose them. In March 2026, Cornell Tech announced a formal partnership with Mastercard to advance AI governance standards, citing the company’s model as a reference for the sector.

A four-step plan for marketing leaders

• Step 1 — Map the real usage. Not the approved tools list. The actual tools, workflows, and AI-assisted processes your team is using today — including the ones nobody asked permission for. Most risk lives in the gap between what is sanctioned and what is happening.
• Step 2 — Build role-specific guidance. Gartner’s data shows persona and role-based guidance doubles the likelihood of value. A content strategist, a paid media buyer, and a CRM manager need different guidance for different use cases. One generic AI policy does not cover it.
• Step 3 — Train, then expand. GenAI ethics training makes teams 1.7 times more likely to report value; regular assessments triple it. Train the first group, assess the results, then widen the circle. The organisations stuck in perpetual pilot mode are the ones that skipped the training step.
• Step 4 — Make the quality trail visible. If your AI-generated content, personalisation engine, or attribution model touches a client, a partner, or a regulated audience, the trail of how you tested, reviewed, and approved it is not overhead. It is the asset. Mastercard proved this. Your marketing function can do the same at a smaller scale.

Who wins the next two years

The EY data says the cost of irresponsible scaling is $4.4 million on average. The Gartner data says responsible use triples the return. The McKinsey data says six years of warnings have not changed most organisations’ behaviour. The next two years belong to the marketing leaders who treat responsible AI use as a capability to build. Train your team. Assess your outputs. Expand deliberately.

The responsible scaling question was never “if”. It was always “how soon and how well”.

← All insights