Short answer: AI is replacing consulting tasks, not consulting judgment. Configuration, data migration scripts, documentation, and test cases are getting dramatically faster. Deciding what your business actually needs, saying no to bad requirements, and owning the result when something breaks are not. We say this as a firm that builds AI products, so we are arguing against our own hype here.
That was true a year ago on instinct. It is now testable, because the vendors moved and the analysts published numbers.
What actually changed in 2026
The threat to ERP consultants was never a chatbot. It is ERP vendors shipping the delivery layer themselves.
SAP made Joule agents generally available across Ariba and Fieldglass in June 2026, and now sells Joule Studio, the tooling partners used to build agents with. Oracle shipped an official MCP service for NetSuite. Microsoft’s 2026 release wave 1 covers AI agents in Business Central and further work on Dynamics 365 MCP servers. Odoo 19.4, released in July, turns Odoo itself into an MCP server. Every one of those is a piece of work a partner used to be paid for.
Then in May 2026, McKinsey put a number on it: AI agents could reduce the effort needed to implement an ERP by at least half, and cut programme duration by half again. They also said the part that should interest you most as a buyer, which is that integrators will be under pressure to pass those gains along.
What AI already does to ERP consulting work
Inside our own delivery team, AI now drafts module configurations, writes and reviews migration scripts, generates test scenarios, and produces first-pass documentation. Work that took a consultant a day takes an afternoon. Anyone in this industry telling you otherwise is either not using the tools or not being straight with you.
The compression is uneven, and the pattern is consistent enough to plan around. Testing and documentation collapse the hardest, which matches McKinsey’s estimate of roughly 80% less testing effort and 90% less training-material preparation. Configuration and migration scripting move a long way. Data decisions, meaning what is worth migrating and what should be left behind, barely move at all, because that is a business question wearing a technical costume. Discovery does not move either.
The honest consequence: the billable hour is shrinking, and pricing models built on it are under pressure. If a firm still quotes open-ended hourly work for tasks AI has compressed, you are paying for their margin, not their effort. It is one reason we quote fixed fees and publish our support pricing.
What AI cannot do on an ERP project
Requirements archaeology. The real requirements of a business are not written anywhere. They live in how the warehouse actually receives goods, in the workaround accounting invented three years ago, in what the owner checks every morning. Extracting that takes interviews, observation, and the experience to know which answers are wrong. An AI cannot notice that two departments gave contradicting descriptions of the same process, then sit both managers down.
Saying no. The most valuable sentence in consulting is “you do not need that”. AI tools are agreeable by design. A consultant who has watched a customization rot across three version upgrades will push back; a model prompted by an enthusiastic manager will generate the customization.
Accountability. When go-live wobbles, someone has to own the fix, at 2 a.m. if needed, and carry the relationship afterward. A subscription does not do that. Ask anyone who has tried to escalate a support ticket to a chatbot.
Trade-offs with consequences. Choosing between platform editions, deciding what data is worth migrating, sequencing a phased rollout so the business keeps shipping: these are judgment calls with your money on the line. AI can inform them; it should not make them.
Political cover. Sometimes the finding is that a department’s process is the problem, and the person who has to hear it reports to the person paying for the project. Delivering that takes a named outsider who can absorb the reaction. No software subscription will stand in that room.
Why AI does not rescue a bad project
Here is the part that undercuts the panic. McKinsey’s own figures put 65 to 80 percent of large technology programmes over budget or over schedule, and only 25 to 35 percent hitting the financial impact they were approved for. Those projects did not fail because configuration was slow.
Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, on costs, unclear value, and inadequate risk controls. It also found that out of thousands of vendors describing themselves as agentic, only around 130 genuinely were. So the tooling that is supposed to replace consultants is itself going through a cull.
AI compresses the build. The build was never the thing that killed ERP projects.
What changes for you as a buyer
Three things are worth demanding from any ERP partner in 2026:
AI-compressed delivery should show up in your price. If the work got faster, fixed quotes should reflect it. Hourly billing quietly pockets the difference. Worth knowing when you push on this: McKinsey reports that even at its own scale only about a quarter of fees are outcome-linked, so a firm resisting the conversation is following the industry, not defending a principle.
Your consultant should ship AI, not just talk about it. Ask what AI the firm has actually built and where it runs. Our answer: agents inside client Odoo systems and a connector that lets Claude or ChatGPT query Odoo directly, listed on the Odoo App Store, with the plain-English version on our ERP AI chatbot page. Whatever firm you talk to, make them show you something running.
The advice layer matters more, not less. As implementation hours compress, the differentiator shifts to diagnosis: understanding your workflow before touching software. That is the standard we hold on our ERP consulting side, and it is the part AI moves last.
Then put four things in the contract, because almost nobody does yet. Whether AI was used in delivery and where. Who owns the code, agents and prompts produced during the project. Whether AI-written customizations are supportable through version upgrades, which is the one that bites eighteen months later. And what happens to a deployed agent when the engagement ends.
What this looks like on a mid-market project
Every analyst figure quoted above describes migrations costing between $100 million and $1 billion. Yours does not. On a project in the tens of thousands, “half the implementation effort” does not halve the invoice, because the phases that compress are a smaller share of the work to begin with. Discovery, data decisions and getting your team to actually use the thing are close to fixed costs at any size. Expect the savings to land in build and testing, expect them to be real, and expect anyone promising you a 50% discount to be quoting a headline written about somebody else’s project.
The version of this that should worry consultants
Some consulting work deserves to disappear. Firms that survived on configuration hours, copy-paste customizations, and documentation nobody reads are in real trouble, and clients inherit the wreckage when those projects stall. We see the aftermath often enough that we built a service around it; a fair share of the projects we take over start with “our previous partner billed a lot of hours and we cannot tell what for”. AI will thin that end of the market, and it should.
The thinning has already started at the bottom of the pyramid. The Big Four cut UK graduate intake sharply in the last cycle, KPMG by around 29%, and graduate job postings across the sector fell roughly 44% year on year. Those are the roles that used to do the configuration hours.
Where this lands
The ERP consultant of 2026 looks less like a configurator and more like a diagnostician with power tools: faster delivery underneath, human judgment on top, and accountability that does not expire when the chat window closes. AI will keep moving the line between those layers, and we wrote about where the technology itself is heading in AI in ERP: what’s actually real in 2026.
Frequently asked questions
Will AI replace ERP consultants?
It is replacing their tasks, not their judgment. Configuration, scripting, and documentation are compressing fast. Requirements discovery, honest pushback, and accountability for outcomes remain human work, and they are the parts that decide whether a project succeeds.
Will AI replace SAP consultants?
The pressure is sharper there, because SAP now sells the agent tooling itself. Joule agents went generally available in June 2026 and Joule Studio is sold as a product. The configuration-heavy end of SAP consulting is most exposed; the process and governance end is least.
Can AI implement an ERP on its own?
No. An implementation is mostly decisions about the business, not software steps: what to migrate, what to customize, what to refuse. AI accelerates the execution of those decisions once a person with context makes them.
How much faster is an AI-assisted implementation?
McKinsey estimates at least half the implementation effort and about half the duration, based on very large programmes. On a mid-market project the saving is real but smaller in percentage terms, because discovery, data decisions and user adoption are close to fixed costs at any size.
Should AI make ERP projects cheaper?
Yes, and you should see it in the quote. AI compresses delivery hours, so fixed-fee pricing should trend down or include more scope. Be skeptical of hourly models that stay flat while the underlying work gets faster.
If our ERP has AI built in, do we still need a partner?
For a small, standard deployment, often no, and a good partner will tell you so. You need one when the decisions are hard rather than the configuration: multi-entity structures, a migration off a system nobody documented, processes that do not match how the software expects you to work.
What should we ask an ERP consultant about AI?
Ask what they have shipped, not what they predict. A firm advising you on AI should be able to show its own AI running in production, explain where it fails, and tell you which of your problems AI will not fix. Then ask who owns the code and agents they build for you.
Is ERP consulting still a good career in 2026?
It is a harder entry and a better ceiling. The junior configuration work that used to train people is the work compressing fastest, which is why graduate intake across the big firms fell in the last cycle. The people doing well are the ones who moved toward process, data and client relationships early.
If you are weighing a quote right now and want a second opinion on whether the scope and the price still make sense, book a call. We will tell you if you do not need us.