Share this post
AI doesn't have a speed problem, AI has a quality problem. Without Quality, we'll never see the value it can bring.
New tools, models and use cases are appearing at a pace few organisations could have imagined even a few years ago. The opportunity is enormous, from improving productivity and customer experiences to helping organisations make better, faster decisions.
But being able to move quickly isn't quite the same as being ready to move quickly.
Because as AI becomes embedded in more important decisions, processes and services, another question becomes increasingly difficult to ignore:
How confident are we in what it's doing?
That's the challenge at the heart of a new report from the Tony Blair Institute for Global Change, AI-Assurance Ecosystems: Building the Infrastructure to Enable Confident Adoption of Safe AI.
It's also a conversation 2i has been pleased to contribute to. Adam Pettman, VP of Artificial Intelligence at 2i provided input and expertise during the report's research and is acknowledged in the published paper.
As Adam explains:
“The UK has an opportunity to lead the field in AI assurance. AI will only deliver value if people trust it. Without clear assurance, AI adoption will stall exactly where the technology becomes most useful.
2i has been working with organisations for the past two years to build confidence in their AI solutions, helping to turn AI spend into business value.”
And while the report is primarily focused on what governments need to do to create effective AI-assurance ecosystems, the underlying challenge is just as relevant for organisations adopting AI today:
Speed creates opportunity. But confidence is what allows you to act on it.
"It works" is only the beginning
An AI system can work exactly as expected in a demonstration.
Good start.
But once it begins interacting with real data, real customers, real employees and real-world decisions, the questions get harder.
Can you explain the outcome?
Is the data appropriate?
Could bias influence the result?
Will behaviour remain consistent as conditions change?
What happens when the system encounters something nobody anticipated?
And what evidence would you want in front of you if a customer, regulator or board member asked tomorrow whether you could stand behind its decisions?
These aren't reasons to slow AI adoption down.
They're part of what allows organisations to accelerate it with confidence.
Trust needs evidence
The TBI report describes AI assurance as the process of testing, evaluating and monitoring AI systems and communicating the resulting evidence to provide justified confidence in their trustworthiness.
That emphasis on evidence matters.
Trust in AI can't simply come from the fact that a pilot worked, a system has been deployed successfully or people are using it.
Those things demonstrate adoption.
They don't necessarily demonstrate quality.
And without understanding the quality of the outputs, the risks involved and how the technology behaves in the real world, it's difficult to understand the value being created.
The report points to the commercial impact of that confidence gap.
TBI polling found that 38% of UK adults identify lack of trust as the single biggest barrier to using AI, while 61% of organisations report delaying or reducing planned AI investment because of concerns about AI-reliability risks.
Confidence isn't something to think about once adoption has happened.
It can determine whether adoption happens at all.
Assurance and acceleration aren't opposites
There's an understandable concern that more governance, testing and assurance inevitably means moving more slowly.
But that's the wrong trade-off.
The report makes the case that governments shouldn't have to choose between innovation and governance. Instead, effective assurance can create the infrastructure that allows both to thrive.
The same principle applies inside organisations.
Good assurance isn't there to create another hurdle before AI can move forward.
It's there to provide the evidence needed to make better decisions.
Where can we move quickly?
Where is the risk acceptable?
Where do we need stronger controls?
What needs further investigation?
And where do we have enough evidence to proceed with confidence?
The objective isn't to remove every possible risk.
It's to understand risk well enough to make an informed decision.
Assurance can't be a final checkpoint
One of the most important points in the report is that assurance should operate throughout the AI lifecycle.
Testing and measurement generate evidence. Evaluation and audit help determine whether systems – and the people and processes around them – meet the required expectations. Monitoring then checks whether that remains true as the system and its operating environment change.
That evidence can inform decisions from design and development through procurement and deployment to ongoing operation.
In other words, assurance shouldn't be something that appears at the end with a red or green light.
By then, some of the most important decisions have already been made.
Bringing assurance in earlier means evidence can shape those decisions while there is still time to act on it.
Not every AI system carries the same risk
Assurance also needs to be proportionate.
A low-risk internal productivity tool and an AI system influencing a customer's financial outcome clearly don't carry the same consequences if something goes wrong.
They shouldn't necessarily face the same level of scrutiny either.
Effective AI assurance starts by understanding the context in which the technology is being used, the potential impact of failure and the evidence decision-makers actually need.
That's a much more useful conversation than simply asking:
Has the AI passed the test?
A better question is:
Do we have enough evidence to make this decision with confidence?
Independent assurance has a role to play
Independence becomes particularly important as AI moves into more consequential areas.
The teams designing, implementing and championing an AI solution are understandably invested in making it successful.
But self-assessment alone has limits.
The TBI report highlights the role of independent scrutiny and third-party assurance, particularly for AI deployed in higher-risk contexts.
Independent assurance provides another perspective.
Not to find reasons why AI shouldn't be used.
Not to slow innovation down.
But to challenge assumptions, identify where the greatest risks sit and provide objective evidence about how technology performs in its intended environment.
That can give leaders a clearer view of where they can move quickly, where additional controls are required and where the risk warrants further investigation.
From AI investment to AI value
And ultimately, this is where the conversation needs to go.
For the last few years, much of the AI discussion has focused on capability.
Then came adoption.
How many licences do we have? How many people are using the tools? How many processes can we automate?
But usage isn't the same as value.
If an organisation can't establish whether the outputs from its AI are accurate, appropriate, reliable and delivering the intended outcome, it becomes very difficult to say what that AI is actually worth.
That's the missing middle between investment and value:
confidence in the quality of the outcome.
The organisations that realise the greatest value from AI won't necessarily be those that simply deploy it fastest.
They'll be the ones that know where AI is working, understand where it isn't, and have enough evidence to make confident decisions about what happens next.
Speed gets you moving. Confidence gets you further.
The TBI report makes a compelling case for building assurance into the infrastructure surrounding AI at a national level.
There's an equally important lesson for organisations deploying it.
AI will continue to move quickly.
The answer isn't to try to remove every uncertainty before moving with it.
It's to build the capability to understand those uncertainties, test what matters, continuously generate evidence and make informed decisions about risk.
Because the question is no longer simply:
Can we use AI?
It's:
Can we confidently stand behind how we're using it – and the value it's creating?
AI delivers speed. Confidence unlocks value.
Read the report: AI-Assurance Ecosystems: Building the Infrastructure to Enable Confident Adoption of Safe AI, Tony Blair Institute for Global Change.
Find out more: Explore how 2i's independent AI Assurance services can help organisations understand risk, build confidence and realise more value from AI.