For Scott Atkinson, CEO of TribeTech, experimenting with AI isn’t about chasing the latest shiny technology, it’s about finding practical ways to make businesses faster, more productive and more responsive – and understanding when AI isn’t the right answer.
“If we don’t, our clients will, and we don’t want the cart pulling the horse”
TribeTech’s experience has increasingly become a co-learning exercise, with the MSP helping clients understand what’s possible while customers contribute their detailed knowledge of their own business processes.
When DIY AI needs an MSP
As AI tools become more accessible, TribeTech is increasingly seeing customers build their own solutions, only to discover they are inefficient, expensive or introduce governance and data risks they hadn’t considered.
In one case, an automation was consuming $10,000–$15,000 a month in token and service costs. For MSPs, this creates an opportunity to step in after the initial experimentation to optimise workflows, control costs and ensure appropriate security and governance are in place. And then potentially act as consultant for the next project, now the ROI has been proven.
Start with an obvious business problem
Scott and the TribeTech team look for client processes consuming resources or creating slow response times, working with around a dozen clients so far on business process automation.
Successful applications include scheduling and workflow automation. Scott also points to insurance renewals as a good example of where AI can become one component within a broader automated process.
A process that previously took hours can potentially be reduced to minutes – with humans retained at the points where judgement or verification matters.
“If you’re doing, say, insurance renewals, you’d have someone following these 10 steps. Well, if you can sort of work out eight of those with AI, that makes things faster and then just insert people where the sanity check makes sense.”
Specialisation uses fewer tokens
Another lesson from TribeTech’s experimentation is that general-purpose AI models aren’t always the best tool for the job. For an industry-specific or tightly defined business process, a smaller language model trained or focused on a narrower body of knowledge can deliver the required answer with far fewer computing resources.
TribeTech has explored specialised models small enough to run locally, potentially reducing token costs while also providing greater control over where data is processed. Scott compares it to searching an encyclopedia – if you already know the answer will be found within a couple of relevant pages, there’s little point searching the entire collection:
“All your answers are going to be in these couple of pages. Therefore… you can do it a lot more simply.”
Take digital governance to the board
Scott sees an opportunity for MSPs to engage business leaders (those approving expenditure) in terms they understand. Board members are accustomed to financial governance and risk management, but digital governance can be less well understood – particularly as employees begin adopting AI tools and connecting business data to new services.
Rather than leading with AI features and technical terminology, MSPs can frame the conversation around familiar board-level questions: What information is being used? Where is it going? Who has access to it? What controls are in place, and who is accountable if something goes wrong? This turns AI governance from a technology discussion into a business risk discussion, creating an opportunity to help establish the policies, security controls and oversight needed for responsible adoption.
Turning AI into revenue
The commercial model around AI is still evolving. While customers understand paying for a professional services project, converting that work into ongoing MRR is harder. “We’ve been trying for a couple of years now to really put MRR around it… It hasn’t been overly successful,” Scott says.
Customers will approve a block of hours or a $10,000 project because they can clearly see what they are buying. The challenge comes once the workflow or automation has been built: much of the visible effort has already occurred, making the value of a monthly fee harder to demonstrate.
For smaller MSPs, that suggests AI may initially be easier to monetise as consulting and project work, while the longer-term opportunity lies in building an ongoing advisory relationship as customers continue their AI journey.
Scoping is another challenge and experimental projects can easily expand. The upside is that each project builds reusable knowledge, but MSPs can’t assume they will recover that investment immediately if at all. Things change quickly and not every learning experience is going to pay dividends down the road… that’s the thing about experimentation.
“The new norm is not keeping up 😊”
Ultimately, TribeTech’s experience suggests MSPs don’t need to have AI completely figured out before getting started. In fact, nobody does.
The important thing is to keep learning, choose worthwhile problems and share your learnings and experience with others in the industry. It’s big, it’s new, and it’s moving fast. None of us can stay on top of it alone.
As Scott puts it, AI adoption is “a journey, it’s not a race with an end. It’s an ongoing journey.”