ERP projects in 2026: The risks, opportunities and warnings
15/09/2026
By Richard Serpell, ERP Coach and Umbrella Club member
I’ve spent more than two decades working inside ERP projects, across some of the world’s largest and most complex implementations through to mid-market platforms used by growing Australian businesses. In that time, ERP software has changed almost beyond recognition. The fundamentals of why these projects succeed or fail have not changed at all.
That gap between what has changed and what hasn’t is exactly where I think business leaders need to focus their attention in 2026. AI has added a genuinely useful new layer to ERP systems, and it has also introduced a new set of risks that most businesses aren’t watching closely enough. Underneath all of it, the same old problems are still quietly wrecking projects the way they always have.
The real cause of failure hasn’t moved
I’ve seen ERP implementations fail so badly that a business had to roll the entire system back and revert to what they were using before, causing a second wave of disruption on top of the first. In cases like that, the reason the business ground to a halt is almost never the software itself. It’s the way it was implemented.
That distinction matters, and it hasn’t changed in the twenty-plus years I’ve been doing this work. Businesses still walk into ERP projects without having worked out why they want the software in the first place, or what they actually intend to do differently once they have it. Most companies’ data is also still a mess by the time they start. Vendor masters, customer masters, charts of accounts, years of accumulated reports, dashboards, and workbooks scattered across spreadsheets that were never meant to be permanent systems.
The three questions almost every failed ERP project skipped at the start:
| Question | Why it matters |
|---|---|
| Why do we actually want this software? | Without a clear answer, requirements end up vague and vendors can’t scope the work properly |
| What will we do differently once we have it? | Software alone doesn’t fix broken processes, it just makes them faster to run badly |
| Is our data clean enough to migrate? | Duplicate, outdated, or inconsistent data quietly undermines the whole system from day one |
Smaller businesses are structurally set up to struggle
Larger organisations at least have the infrastructure to absorb an ERP project. Smaller and medium businesses usually don’t have that cushion, and I see the same pattern play out again and again.
A business without a dedicated IT department will typically find whoever happens to be available and hand them the project. That person becomes the point of contact, running discussions with the vendor, building the project plan, and pulling in the rest of the internal team. There might be a project sponsor watching from a distance, but they’re rarely doing the actual work.
The person left holding the project usually gets overwhelmed. None of it is familiar territory for them, they’re getting pressure from their manager to move faster, and the vendor can only carry about a third of the workload. The other two thirds sits with the client, whether the client is ready for that or not.
I think of my own role a bit like a buyer’s advocate in real estate. I’m helping the business choose the right system for the right reasons, protecting their interests through the process, and making sure they aren’t left to figure it out alone. Given that an ERP system is typically in place for around twenty years, getting the early decisions right is worth far more than it costs.
Where AI is genuinely useful
I don’t think businesses should be cautious about AI for the sake of being cautious. Used well, it removes a lot of the repetitive administrative load that used to eat up people’s time inside an ERP system.
A simple example: you can set AI to find every open purchase order, and for any order where more than 95 per cent of the goods have already been delivered, automatically close it out. That’s not AI doing anything a person couldn’t do themselves. It’s just doing it faster, and it frees that person up for planning, strategy, and the kind of thinking that actually needs a human in the room.
Where AI genuinely helps vs where it doesn’t, inside an ERP system:
| Task type | AI’s role |
|---|---|
| Repetitive admin (closing fulfilled orders, flagging anomalies) | Strong fit, saves real time with low risk |
| Reporting and insight generation | Useful, but the output still needs a human check on accuracy |
| Requirements gathering and vendor selection | Limited value, this needs judgement and business context AI doesn’t have |
| Governance and oversight of AI tools themselves | Not something AI can manage on its own, this needs a clear owner |
That’s the upside, and it’s real. The problem is what tends to happen around it.
The governance gap nobody is watching closely enough
The governance of AI development inside businesses is critical, and in my experience it’s often simply not there. I’ve watched this happen directly with clients: staff independently signing up for paid AI tools with no central oversight, no one coordinating what’s being built, and six different people quietly building the same thing without knowing anyone else is doing it too.
That’s not a hypothetical risk. It’s a licensing cost, a security exposure, and a waste of time all at once, and it’s happening inside businesses that would never let six people independently negotiate the same vendor contract without anyone noticing.
Around 12 per cent of Australian businesses reported using AI in their workplace in 2024-25, and adoption is climbing fastest inside larger and more innovation-active organisations, which tells me this governance gap is only going to widen before it narrows. The businesses moving fastest on AI adoption are often the same ones without the internal structure to manage it properly. Scalesuite
I’d also add a second, quieter risk: treating AI as a toy rather than a business tool. I’ve seen this with technically capable people who find AI genuinely fun to experiment with at work, which is understandable, but it isn’t why they’re there. Left unchecked, that novelty factor wastes real time and money, and produces output that’s only as good as the prompt behind it.
Vendor selection mistakes that have nothing to do with features
When businesses do get to the vendor selection stage, the mistakes I see are rarely about functionality. They’re about the details that get skipped because they don’t feel urgent at the time.
- Support model and geography. A vendor needs to be supported properly in the region where the business actually operates, in the way that business wants to be supported. Some vendors only offer email support and never give you the option to get someone on the phone. Whether that’s acceptable depends entirely on the business, but it’s worth asking before signing, not after.
- Cultural fit. We’re careful about matching culture when we hire someone into a business. The same thinking should apply to the vendor and the company supporting that software, because a mismatch there causes exactly the same kind of friction a bad hire would.
- Genuine market coverage. Businesses often canvas one or two vendors they’ve already heard of rather than properly assessing what’s actually suitable for their size, industry, and geography.
Timing determines how expensive a problem becomes
An ERP project can go wrong at any point along the way, and some problems are far easier to recover from than others. The earlier something surfaces, the cheaper and simpler it is to fix.
The expensive version looks like this: a business is well into implementation before anyone mentions that part of their operation involves equipment rental, not just equipment sales, because nobody thought to raise it earlier. Now there’s no functionality built for that at all. Sometimes that’s an easy retrofit. Sometimes it has implications running much deeper into the system than anyone expected.
Internal politics is still underrated as a risk
If I had to name the most common reason ERP projects actually go wrong, it isn’t technical. It’s contention between departments that can’t reach a consensus, egos that get in the way of a decision, and requirements that were never clearly agreed on in the first place. Add in a lack of clarity about what happens after go-live and who’s responsible for ongoing support, and you have most of the real risk sitting right there before a single line of code gets touched.
None of this is new. AI hasn’t changed it, and I don’t expect it to. What AI has done is add a new layer of opportunity on top of these same old fundamentals, along with a new set of risks that deserve just as much attention as the ones we’ve always had to manage.
Getting it right
Businesses that get ERP right in 2026 are the ones treating it the same way they always should have: understanding what they actually need before they go to market, cleaning up their data, being honest about internal capability gaps, and choosing a partner who is genuinely on their side of the table rather than the vendor’s.
That’s the role I try to play as an ERP coach. Not building the system, not running the project day to day, but working alongside a business’s own project lead to make sure the early decisions get made properly, when they’re still cheap to get right. It’s strategic advisory, and vendor-agnostic by design, because my only job is to look after the business’s interests, not any software vendor’s.
If your business is heading toward an ERP selection or implementation and wants an experienced, independent second opinion in the room, contact Umbrella Club to arrange a conversation with me.
Richard Serpell is an ERP coach with over two decades of experience across enterprise and mid-market ERP platforms. He runs an advisory practice helping small and medium businesses select and implement ERP software, working alongside internal project leads rather than in their place. He is vendor-agnostic by design, focused exclusively on ERP, and is a member of Umbrella Club.
