This webinar was set up to unpack what’s happening inside MSPs and cloud businesses right now, how AI is shaping them, and where the opportunity sits.
Terry, RoboShadow CEO, and James Edwards, 3Gi CEO, have both come at this from practical backgrounds. Terry built 3Gi into a serious MSP with James, later creating Cloud Combinator as an AI-focused business working deeply within AWS.
What follows is based on lived experience across MSP delivery, cloud transformation, and now AI.
AI on its own is not currently the biggest revenue generator in an MSP. The professional services fees aren’t enormous, and the margins on pilots aren’t life-changing, so if you’re expecting a dramatic new profit centre overnight, you’ll be disappointed.
However, AI really does change the commercial dynamic at different levels.
"There's a big demand for businesses to have an answer, not just at board and investor level for what their AI strategy is, but also to their users that want access to the best tools and capability.” - James Edwards, CEO, 3Gi
Boards need an AI story, investors are asking about it, staff are using tools already, and customers assume it’s being adopted; there’s pressure from every direction.
If you can help a board articulate a coherent AI strategy, even before anything substantial is built, you immediately move up the value chain and become the partner that's helping them navigate risk and opportunity.
That shift in positioning has a “sticky” effect, and part of that stickiness comes from operating at a higher level in the organisation.
In 3Gi’s case, over time they found themselves effectively replacing CTO and CIO functions in some businesses, roles that are notoriously hard to recruit for; most boards don’t really know how to assess them without external help.
But if you can step into that advisory space credibly, you could see your base fees shift accordingly, and AI has been one of the key enablers of that move.
James’ AI journey accelerated through AWS.
“We ended up doing a lot of advanced work within AWS for specific use cases for ISVs… that allowed us to go through a raft of use cases in ISVs over the last year and a half and build the basis of that company - specifically delivering and building highly developed complex tooling within their AWS environments.” - James Edwards, CEO, 3Gi
After becoming an advanced AWS partner, 3Gi developed strong relationships with AWS sales teams. Around two years ago, demand started growing significantly within AWS’ ISV customer base. Independent Software Vendors (software companies building products for scale) were exploring early-stage machine learning, data pipelines and analytics using services like:
SageMaker unified ML pipelines, Bedrock enabled foundation model access and Retrieval-Augmented Generation (RAG), and AgentCore now packages agentic systems at scale.
The early demand wasn’t from traditional SMEs, it came from tech-heavy ISVs - developers, product teams, DevOps engineers. These organisations needed help building serious AI capabilities inside AWS environments, and that work became the foundation of Cloud Combinator.
Over the past year and a half, they’ve built highly complex tooling in AWS for specific ISV use cases. In doing so, they learned two important things:
“We learned a lot of where the actual hard parts in building AI solutions are, but also the fabric of building around AI solutions from when Amazon brought out Sage Maker, all the way to Bedrock, and now to their latest release of Agent Core.” - James Edwards, CEO, 3Gi
Interestingly, investment has largely started in ISVs because their tools scale. Non-tech SMEs tend to build for internal use only, so their user base caps out quickly.
Cloud Combinator has deliberately stayed in the ISV space, but the growth in low-code and no-code AI demand among SMEs is undoubtedly getting much harder to ignore.
There’s a myth that building AI capability requires hiring £80k–£100k specialists immediately, and that wasn’t our experience.
When Cloud Combinator began, there was effectively no one with a year’s experience building LLM-based solutions in AWS, not to mention that the talent pool didn’t really exist.
“We picked some individuals with really good educational backgrounds, often some sort of data or science, ideally data science capability… Also individuals who really wanted to learn and learn quickly, because we’re never using the same tools that we were using six months ago.” - James Edwards, CEO, 3Gi
Grads who started the business with them became senior architects quickly. The technology is well documented and vendors like AWS provide plenty of support, so the key trait was intellectual curiosity and the willingness to stretch.
At RoboShadow, we see the same thing - there are broadly two types of technologists:
Both contribute to the integrity of your company in different ways, but the AI business seems to favour the second option a bit more.
There’s a widely quoted statistic that most AI projects fail (about 95%).
In Terry and James’ experience, failure rarely comes from model capability. It comes from skipping the fundamentals.
“We’ve had a lot of success in that area just by following ancient practices of the Software Development Life Cycle, starting with the business requirements and analysis stages... Then you get to a point where you can start to associate solutions to those use cases.” - James Edwards, CEO, 3Gi
The framework that has worked repeatedly for non-tech businesses looks very familiar, because it is familiar. It follows the traditional Software Development Life Cycle (SDLC):
The discovery stage is unglamorous; it involves spreadsheets, mapping processes, analysing datasets, understanding existing CRM and ERP limitations. Many AI failures look suspiciously similar to failed CRM or ERP projects, with poor requirements, unclear ROI, and no structured testing.
Here's the key: once processes are mapped, you select high-ROI use cases and design around them. In SMEs, that often means buying software or using low-code/no-code tools rather than building from scratch. Note that most businesses with 250 users don’t want ten developers on payroll; they want leverage, not a full tech department.
Another point is that pilots should be small and fast. If you can fail quickly and cheaply, you avoid expensive production mistakes. Ultimately, a pilot that proves a concept is different from a pilot that is intended to explore and discard.
In short, the “unsexy” part of AI (cleaning data, connecting APIs, transforming datasets) is still very much required, and if you avoid it, you’ll most likely end up paying for it later on.
“There are many reasons why AI solutions fail, and many are a case of having a set of poorly designed and implemented CRM systems, or ERP systems. They follow the same trail of not having clear requirements, not understanding ROI and not having good test plans, not following the SDLC.” - James Edwards, CEO, 3Gi
Compared to building traditional software from scratch, building a useful GPT-powered workflow today is far less complex than people assume. Much of the value now sits in:
A lot of this work can be done by relatively junior engineers. There is DevOps involved, but much of it is composable. Lego-block architecture isn’t a terrible analogy for this.
The bigger shift is happening in no-code and low-code environments...
Unfortunately, Copilot initially suffered from being perceived as “the thing that augments Word and Excel.” This limited understanding stuck, but Copilot Studio, combined with Dataverse and the broader Power Platform, is a different proposition entirely.
“What we can build in Copilot Studio and Dataverse now is actually quite remarkable; 12 months ago we used to have to build that in Bedrock with a complex RAG structure and different data transformation tools to use different data solutions within AWS, and now you just connect things up with Dataverse, and you already have your whole knowledge base structure attached to an LLM.” - James Edwards, CEO, 3Gi
You can now:
For MSPs with strong Microsoft 365 foundations, this is an obvious growth path. If you understand your client’s business processes, you can create serious value through agentic workflows without writing large volumes of code.
This is likely where the bulk of SME AI growth will sit, not in custom foundation model builds, but in intelligent orchestration of existing SaaS stacks.
“I think that's where most of the growth in AI, outside of ISV, will be headed, and where I think the most valuable elements for clients is.” - James Edwards, CEO, 3Gi
Both AWS and Microsoft provide AI enablement funds, assessment funds and build funds. These are competitive areas for them; they want cloud footprint growth.
If you hold the right partnership tiers, such as Advanced AWS Partner, Microsoft Gold, or solution designations, you can access meaningful funding support for:
This funding de-risks early-stage work for clients.
In tenders, being able to say that AWS or Azure will co-fund the proof of concept materially improves your position. It helps you win the relationship, which is where that lifetime value is.
When customers worry about data being sent to AI models, they just need some clarity.
An LLM doesn’t “hold” your data in the way a database does. It processes input and generates output.
“There was a great demonstration by one of the senior AWS AI compliance people… he said that a language model doesn't actually hold any data at all, it's effectively a nothingness of data, it's a processing shop.” - James Edwards, CEO, 3Gi
The compliance question is the same as any SaaS integration:
Personally identifiable information (PII) remains regulated whether AI is involved or not. If you understand SharePoint permissions and SaaS integrations, you already understand most AI compliance concerns. Knowing this and being able to articulate it calmly will position you well in board conversations.
During the UK downturn, many customers asked to reduce bills. Interestingly, the ones engaged in AI projects with us were far stickier.
“It definitely makes competition on MSPs way easier if you can cover that data digital RPA piece for sure.” - James Edwards, CEO, 3Gi
If you’re embedded in data strategy and digital transformation, clients are less inclined to replace you. And if they do explore alternatives, they tend to have transparent conversations about it.
There’s a growing debate about whether agentic AI will eventually erode traditional SaaS platforms.
The theory goes like this:
Most ERP and CRM systems don’t perfectly meet business requirements without customisation. Agentic AI can sit on top of them, connecting APIs, transforming data, bridging gaps between payroll, accounting and operational systems.
Over time, those intelligent overlays could become more valuable than the core SaaS application itself. We’re already seeing cases where it’s easier to connect an API, transform data, and push it back into a platform than to negotiate feature changes with the SaaS vendor. Whether that gradually consumes parts of SaaS functionality remains to be seen, but it’s a trend worth watching.
“Where I see, at least initially, agentic workload starting to threaten the SaaS functionality is when businesses decide that actually it's easier to build agentic bolt-ons, integrations and automation processes, that link up all of my SaaS applications, and bridge the gap that my accounting platform didn't, properly to my payroll system.” - James Edwards, CEO, 3Gi
RoboShadow, as a SaaS product, isn’t threatened by this shift; we’re building into it. The move towards being a Managed Intelligence Provider aligns with where the market is heading: more data, more automation, more AI-first thinking.
In many ways, it already has. Venture capital and private equity over-invested, expecting rapid returns. When those returns don’t materialise on schedule, capital tightens and it affects the wider economy. But the bursting of a funding bubble doesn’t mean the technology disappears. For example, the dot-com era proved that.
The tooling will remain, the use cases will mature, and the hype will subside. For MSPs, the real risk isn’t that AI vanishes; it’s more that someone else integrates it into their offering more effectively than you do.
As Terry says, “you’re not going to get replaced by AI, but you might get replaced by someone using AI”, and at the moment, that theory seems to be coming to life faster than most people think.
Thank you for your ongoing support and feedback, and if you have any questions, please don’t hesitate to reach out to us at hello@roboshadow.com
Source:
The insights above are based on a conversation between Terry Lewis, CEO of RoboShadow, and James Edwards, CEO of 3Gi.