Why AI still needs human intervention to succeed

Artificial Intelligence (AI) is no longer a futuristic concept – it’s already here, in your systems and being used by your teams.

Last year a global study found, 78% of businesses were using AI in at least one business function – a 55% increase on the previous year.

There’s no denying the rapid rise of AI but despite its increased prevalence, AI models often aren’t ready to roll out independently and still need human intervention to reduce risk and build trust.

There are many reasons for this, so we sat down with Salocin Group’s Alex Keep, Director of Business Operations and Alex Holt, Strategic Consulting Director to discuss the current state of AI models and why humans must play a vital role in their adaption and adoption.

No one-size-fits-all AI

There’s more to using AI successfully than simply ‘plug-and-play’. Each system has its own architecture, guardrails, and strengths – and as a result, the experience of working with them feels different:

Claude – often comes across as measured and thoughtful. Its large context window means it’s strong at digesting long documents or sustaining extended conversations, which can feel like talking to a calm, attentive colleague
ChatGPT (OpenAI) – tends to be more dynamic and creative, good at rapid back-and-forth idea generation or answering a wide range of questions. It feels versatile and generalist, like someone who always has an answer at hand
Microsoft Copilot – feels more formal and predictable. That’s deliberate it’s not a model, but a service that layers Microsoft’s enterprise compliance, security, and integration on top of OpenAI models (usually GPT-4/4o and more recently GPT-5). The result is a more constrained, “business-safe” style of interaction, deeply embedded in Microsoft 365 and Dynamics

“Even AI models designed on similar principles behave very differently. Take Microsoft Copilot and OpenAI’s ChatGPT, for example. They both rely on OpenAI’s foundation models, but Microsoft adds its own enterprise wrapper. That means if you enter the same prompt in both, the tone and focus of the response will differ,” says Alex Holt.

This is why choosing the right model matters. As Alex Keep notes:

“This highlights the importance of matching the model to the ask and task – or prompt – and that it still requires human insight and control. The choice of GenAI model should be a strategic decision, not necessarily an automatic one.”

And it’s not just about matching to the task – governance must run alongside it:

“A lack of AI governance can leave the business and their customers open to serious risks – including biased outcomes and privacy breaches. Strong AI governance is important for organisations; it allows them to assign accountability for those times AI decides to go awry,” adds Keep.

The extra layers Microsoft has implemented create a smoother, safer experience for Copilot users – boosting productivity by enabling collaboration across the existing Microsoft tech stack.

Role of humans in building trust in AI

The scary thing about AI is that it can sometimes confidently present incorrect information back to you – it hallucinates.

48% of survey respondents claimed their GenAI models displayed signs of hallucinations – generating incorrect outputs despite it appearing plausible and confident in its response. This is a concerning risk, particularly when data used could include customer information.

This goes beyond a technical issue – it’s about trust, or lack of. This is why it’s so important for organisations to have an AI governance framework in place – especially as AI evolves faster than guardrails can be implemented.

“Human oversight of AI is absolutely critical. This isn’t a ‘set and leave’ rigid automation, it’s experimental and can be unpredictable. We need to embrace its potential without losing sight of its limitations and weaknesses,” says Alex Keep.

Treat AI like the new hire

Giving AI a personality is one of the most effective ways to collaborate with it, although it’s important to remember, it’s data and tech you’re communicating with – not a person.

“We’re now able to create AI agents – they’re like incredibly smart new junior hires that can still lack some skills that can only be learned on-the-job. So, they need your support and guidance to successfully complete tasks and achieve goals,” says Alex Holt.

“AI can learn quickly and produce great work but if it’s not provided with context, structure and feedback, it can swiftly go off track.”

This is where treating AI like you would a new hire often works best:

Onboard it properly – provide context about your business, brand and goals (in a safe environment – eg a test workspace)
Break down tasks into simple steps – ask for focused outputs, instead of overwhelming, open-ended tasks
Assign the right ‘person’ (or agent) to the right job – use different models for different stages of a process or aims of a task

Just as you would support a new employee, when prompting AI, you should split it into bitesize, easily understandable chunks.

Creating an AI centre of excellence

You might find yourself not knowing where to start when it comes to introducing AI governance into your organisation – establishing a working group or taskforce is a recommended first step.

Many businesses – including Salocin Group – are now creating an AI centre of excellence. This is a team in charge of ensuring responsible use of AI across a company.

Alex Holt describes: “You absolutely should create an AI centre of excellence, not just to control the use and scale of AI organisation-wide, but to also embrace its full potential. And that centre of excellence isn’t just about data and technology – you also need well-executed processes, governed by experts.”

Implementing an AI centre of excellence would combine technical expertise with creative insight – allowing you to centralise and govern your AI use and set the benchmark for how your teams best utilise AI across your organisation.

It’ll also help grow trust in AI – with a study by KPMG stating that 42% of the UK are willing to trust AI. Any opportunity to build on this and create further trust is important, AI isn’t going anywhere – businesses need to embrace it by implementing best practice to build trust amongst users, and that trust fuels productivity instead of mistakes.AI still needs a human at the wheel

The most successful companies aren’t those replacing people with AI, they’re combining human insight with AI’s growth capabilities – shaping AI agents like a new hires and instilling AI excellence and best practice organisation-wide.

“AI can radically speed up workflows and uncover insights, but it’s people who keep it accurate, relevant, and aligned to the organisation” says Alex Keep.

We’ve completed more in-depth research comparing the benefits and limitations of GenAI models and put together a guide on how to get started using AI, read more here.