MSPs: The better part of customer AI adoption is caution and strategy

The artificial intelligence (AI) revolution is truly underway and has captured the attention of public and private organisations across industries. However, a conveyer belt of AI-branded products has created a trust gap between channel partners and customers.

In fact, the majority of everyday Australians (54 per cent) and New Zealanders (53 per cent) have substantially lower comfort with AI than many other markets around the world (70 per cent), according to the Ernst and Young AI Sentiment Index Analysis New Zealand and Australia 2025.

Only 37 per cent of Australians and 29 per cent of New Zealanders surveyed believe the benefits of AI outweigh the negatives, compared to 59 per cent of all other markets.

What this shows managed service providers (MSP), resellers and valued-added resellers is that they need to be a guide to businesses, especially the smaller ones, when it comes to AI.

In a previous article by ARN, it was shown that better collaboration and understanding between all parties in the channel was essential to improving service delivery and client retention.

Fool’s gold

MSPs need to have the right packages pricing, and strategy to go out to customers with AI and agentic AI products.

With this in mind, the market is seeing a surge of products from vendors, platform providers and others offering consumer grade products that on a surface level are easy to use, without providing information on the complex infrastructure behind them.

The problem with that is it gives customers the perception that consumer-grade AI tools are plug-and-play solutions that can be used at a business grade level in smaller businesses.

This is far from the reality, because in a “gold rush mentality, everybody’s doing everything really, really quickly, which is actually insanely risky”, said IT Architect as a Service (ITAaaS) managing partner Pat Devlin.

He noted that there’s a lot of things happening in the space and that is problematic.

Continuing the theme of “gold rush”, Devlin said the reason it’s “exciting is because there is gold” and “nobody gives up their comfy life and moves to the hills to dig holes in the ground because they like the lifestyle”.

“If you find the gold, it’s totally worth it,” he said. “AI isn’t the gold, it’s the dynamite and [it] blows open a well of opportunity: competitive advantage, speed, productivity [and] cost savings.”

Devlin explained if AI is used recklessly, it can be explosive and right now there are “people lighting fuses without a map, without training and without guardrails … then acting surprised when things go wrong”.

“The problem is there’s many people doing many things and it’s really hard to tell what’s useful or not.” he said.

The number of AI projects undertaken by companies has risen, but the number of projects that have failed have also risen, which Devlin considered to be “insane”.

“That’s what’s nuts. There’s huge potential, but it’s all moving so fast and there’s so much noise that most teams are struggling to get it right,” he said.

“The real problem is the hype. There’s so much noise, it’s genuinely hard to tell who’s doing something real and who’s full of crap.”

An example of this was the recent failure of AI startup Builder.ai, noted Devlin.

The AI-powered app development startup was once hailed as a game-changer but it proved that even AI vendors aren’t immune to failure, he wrote in a blog post on ITAaaS’ website.

“Despite raising millions and promising the world, Builder.ai entered insolvency in May 2025. But they’re far from alone … [as] 966 startups shut down in 2024, up 25.6 per cent from 2023 [according to equity management company Carta],” he said.

“We’re watching a gold rush where most miners are dying of dysentery before they even reach the mountain. The mortality rate for AI startups is approaching pandemic levels.”

Devlin’s other main concern is that consumer-based AI products are built on top of foundational models and infrastructure developed by a handful of vendors.

“Everyone’s claiming they’ve built something unique with AI,” he said. “In reality, most of these so-called ‘proprietary’ tools are just slick front ends built on top of foundational models like ChatGPT.

“Even when they’re not using the public version, they’re usually just wrapping a model that someone else built. At the end of the day, a lot of these companies are more interested in harvesting data than delivering real value and they’re doing it with minimal guardrails in place.”

This leads to two separate issues, one around privacy and the other is called the “black box” system. The latter where machine learning or deep learning model can see the information that comes in and then provides a decision based on the data provided, Devlin explained.

“Recently, a non-profit organisation wanted an AI to make decisions on claims that involved vulnerable people,” he said. “These systems are often black boxes; where data goes in, a decision comes out, but there’s no justification or chain of reasoning provided.”

He said if it’s not known “how those decisions are being made, what data the model was trained on, or what biases are baked in, you’re opening yourself up to massive risk”.

“The danger isn’t just that you don’t understand the rules, it’s that no one does,” cautioned Devlin. “The AI isn’t applying transparent logic; its generating outcomes based on patterns, not principles.

“If you’re going to deny a vulnerable person access to a critical service, you better know why.”

Despite these issues, Devlin sees huge opportunity for AI in the channel. The reason why this current stage of AI is called the “gold rush” is because the people who will succeed and do well aren’t waiting for everyone else to figure it out, he said.

“The thing that kills me every time we do something like this is it’s a knowable, repeatable pattern that we’ve done in the industry a squillion times and yet we seem to forget the lessons,” Devlin said.

It’s not easy for an MSP

This rush to AI everything has created a situation where there are so many tools that do so many different things, but Notitia managing director Alex Avery said they do pretty much the same thing within a specific niche.

“The challenge now for client is deciding what best of breed tooling looks like,” he said. “[Putting] it all together in a way that actually works for them is something we’re doing all the time.

“I think there is a real challenge now for clients where they’re overloaded with options. As a partner, we have an obligation to have lots of partnerships but also to try out these, these tools and work out what goes together.

For Avery, the idea that AI agents will handle and replace everything is a fool’s errand. Data gets transformed mainly because system structures differ and because of quality issues.

“You need a human in the loop to articulate what good or bad looks like and define that,” he said. “In the case of something like data transformation with AI agents, you want a deterministic approach.

“You want someone to say this and that needs to be not a black box.”

When it comes to data, that’s a real challenge because the number one problem that organisations have is trusted information, Avery explained.

“If you’re saying, ‘Oh well, the way that we transform this information from my CRM into my sales system is something I use this black box [for] … but I can’t really tell [you] how it works, nor can I repeat it exactly, accurately every time’, then there is no trust,” he said.

“That’s going to rear its head more and more.”

To bring it back to the MSP, Avery acknowledged that they end up being the “coalface” of it going wrong and “they cop it, like the people on the help desk are going to cop it”.

“That’s a rubbish position to be in as an MSP,” he said. “You didn’t sign up for that.”

Avery said often data analytics and digital transformation consultants like Notitia will get called into a project and become involved to effectively act as a “business middleman in projects where we communicate from the customer back to their MSP”.

“We can speak a common language across both groups [MSP and business],” he said. “We see that the MSP gets thrown out off the back of a bad project.

“The MSP, either willingly or unwillingly, [is] being pushed to flog a new product that they’re not fully comfortable with and they end up wearing the implementation risk.”

These organisations aren’t set up to do that work, noted Avery. These are companies that maintain help desks and are meant to be built on long, multi-year relationships where people don’t churn through.

“The expectation gap on the MSP is huge,” he said. “Where it’s coming from is the large vendors have made massive AI investments, so they need to see a return on those investments, and that’s through pushing products.

“Then the MSPs are incentivised to push a product and it’s not done well.”

Avery believes the days of people pushing product is going away and that’s good to be a good thing.

“Hopefully [it’ll] be a world where product is now easy,” he said. “It’ll be about articulating what the solution is and articulating what and where the organisation needs to be to get there.

“That is the more human challenge, which is awesome.”

What it takes to deliver an AI project

Devlin said AI projects require time, data, advice and guardrails to be in place.

“Don’t just do it because, ‘Oh my god, I must do it’,” he said. “Take that extra month or two to prove that it works. Do a proof of value instead of a proof of concept.

“Get security assessed – all that other stuff you do for everything else – which gets forgotten because we’re rushing with AI.”

Data and AI consultancy V2 AI’s chief technology officer, Pete Stanski, told ARN that more mature organisations lead with governance.

V2 AI talks to customers about the important issues related to an organisations’ usage of AI. For example, the risk of making the CEO’s salary information available when the AI is asked about everyone’s salary balance.

“[They] don’t want that and there have been cases which I think we’ve all read about [where] that’s transpired,” he said. “What it comes down to is having a very mature approach to these things.”

This requires having the right foundations in place check to make the project isn’t being rushed and don’t almost run with scissors type of scenario, noted Stanski.

“You’re going to get hurt, or you’re going to hurt the customers and when you think about the customer sentiment Australia hasn’t had the best track record when it comes to … data breaches,” he said.

“That has eroded customer trust and with AI and agentic you are opening to potential future disasters. Therefore, you really want to make sure that you have a certain number of guardrails in place.”

V2 has a structure called Velocity, which essentially in about five stages goes from the idea of what the AI will do for the business, then methodically goes through building a prototype that has gone through security and governance, Stanski said.

“AI is a team sport. IT historically has very often built IT solutions in isolation,” he noted. “You build it and the customer will come, or the businesses will get what they get.

“The difference with AI is you have to build your solutions differently. You need to work from who is the user, what are the use cases, what are the scenarios, what are the workflows … you’re trying to make agentic or empower them to be more autonomous.”

“That is a secret sauce … they need to be more personalised.”

Stanski’s advice for MSPs working on AI projects is to build it internally first.

“Don’t forget improving employee experience,” he said. “Test it with your people in a safe, secure environment before you test it with your customers and do the customer experience improvements through your AI solution.

“You still have to think about [the] AI operating model to make sure you don’t introduce vulnerabilities when you make changes once and after you’ve launched it.”

To do this, IT service providers will need to continue to tick all the boxes for risk and compliance and ensure there are no data breaches.

“That is a path to a successful AI deployment in an organisation,” he added.