The ERP Reality Check · Part 2 of 4
The series so far: Part 1 argued that ERP programs usually fall short because old processes get copied into the new system, and that payback is decided early, in a business case with baselined KPIs and owners in the business.
The steering committee is an hour in. The program update went fine, more or less. Then someone asks the question that has been hanging around for months: should finance start using AI now, or wait for the new ERP?
Everyone in the room knows the safe answer: no AI before ERP upgrade work is done. Finish the migration, clean the data, put AI on a stable core later. Nobody gets called into the CFO's office for waiting.
We'd push back on that default. Waiting usually means designing finance processes twice, once for the ERP and again when AI shows up. If you run finance mid-migration, the question is probably on your desk too.
AI before ERP upgrade: two ways to play it
Any finance team in the middle of a migration has two broad options. Neither is reckless. They just lead to different places.
Wait for go-live
The appeal is obvious: one big change at a time. The catch is timing. The ERP design gets locked in the early waves, and nobody in those workshops is designing with AI in mind, because AI isn't in scope yet. When it finally arrives, it has to fit around processes built without it. So the redesign happens twice, once for the ERP and then again to make room for AI.
The value gets pushed back too. Go-live is rarely the finish line. There's stabilization, then a backlog of fixes, and only then does anyone pick AI back up. A year or more after go-live is a reasonable expectation for when AI value starts to show up on that path.
Start now on a few processes
The other option is to pick the work that's slowing finance down today and put AI on it while the ERP program runs. For a manufacturer or distributor, that's usually supplier invoice matching, customer deductions and intercompany across plants.
You learn two things quickly: where the value sits, and where the data breaks. Both are exactly what the ERP design team needs while the template can still change. The process gets designed once, with AI in it from the start.
We'd take the second path. It runs on the ERP you already have, and it turns the AI work into input for the migration instead of a project queued up behind it.
Should finance wait for the new ERP before using AI?
The instinct to wait is reasonable. The program will run for years, the data is messy and the SI is already stretched. Adding AI feels like adding a second moving part to a machine that barely holds together.
The ERP program doesn't stand still while you wait, though, and neither does the close. Every month of pause is another month of the team matching invoices by hand, chasing deductions and reconciling intercompany in spreadsheets. That work doesn't get cheaper because a new system is coming.
AI could change the ERP program itself
There is a second reason not to wait, and it has to do with the program rather than finance operations.
McKinsey research published in May, The end of ERP as we know it?, argues that AI agents could cut implementation effort by at least half and halve program duration. If that holds, the expensive part of the program stops being configuration and testing. Change management becomes the main constraint.
Keep the scale in mind. The same analysis puts migration costs at $100 million to $1 billion for most large enterprises, with payback often taking four to five years for companies that use the migration to transform the business. Those figures describe big companies. A mid-market migration will be smaller. The direction still applies.
For a CFO, that is a budgeting question. If half the build effort could disappear, the money and attention should move to the people side: training, process ownership, controls. A program planned in 2024 probably did not assume any of this.
AI that runs the whole process works on the ERP you have today
There is a ceiling worth knowing about, and it isn't the age of the ERP. What hits it is technology that covers one step: a tool that suggests the next action, or a script that clicks through the same screens a person would. That step gets quicker. The process around it stays the same, along with its workarounds and manual handoffs.
Work that runs the whole loop is different. Take a supplier invoice with a price variance. The AI reads the invoice, checks it against the purchase order and goods receipt, clears the routine matches, posts in the ERP, and sends the exceptions to an AP lead with the context already pulled together. People still approve what matters. None of that needs a new system. It runs on the ERP you already have.
It also feeds the migration. Every exception routed to a person shows where the process breaks, which is exactly what the ERP design team needs to know.
Start with the data you have
The biggest stated reason for waiting is data quality. "We'll do AI once the data is clean" is a line we hear a lot in finance transformations.
Recent research on CFOs and AI points the other way. Its advice is to start with the data you have, imperfect as it is. In our experience, teams that wait for perfect data stall, and the ones making progress improve the data as they go.
Two things make that work in practice.
First, finance should own data governance for finance data, rather than handing it to IT or the SI. Finance knows which fields drive the close and the forecast, so finance should set the standard.
Second, give people early wins. AI assistants that answer policy questions, draft variance commentary or pull reports are low risk and visible. They build support for the harder work and they show you quickly where the data breaks.
The ERP data cleanup still has to happen. Running AI early just tells you which parts of the cleanup matter most.
A 90-day sequence for a CFO mid-migration
The steps below are our own practical guidance, not findings from the research above. They assume you are already somewhere inside an ERP modernization roadmap.
Days 1–30: pick the processes
List the five to ten finance processes the ERP program will touch in its next two waves. For each, write down a measurable target for faster, cheaper or better. Use AI tools on the data you have to find where time and errors cluster. Close, reconciliations and forecasting are usual suspects. For a manufacturer or distributor, add supplier invoice matching and customer deductions.
Days 31–60: test two changes
Pick the two processes with the clearest value and redesign them, not just the tasks inside them. Name the owner, the decision rights, the handoffs and the controls. Run a small test with AI in the loop, then measure against the target. Put one AI assistant in front of the wider team as a quick win.
Days 61–90: feed the ERP program
Take what you learned into the ERP design authority. Where the test showed a better process, ask for it in the template. Where it showed bad data, put that field on the cleanup list with a finance owner. Re-check program budget assumptions against what AI could do to build and test effort.
The CFO owns the spend decision
Throughout, the CFO should be the visible sponsor and the person who decides which AI spend continues. That includes compute and token costs, which rarely show up in the original business case.
After 90 days you should know where the value sits and what the ERP template has to support. Teams that start now have that answer before the template is locked. Teams that wait get it after go-live, when changing the design costs far more.
For the wider picture of what the research says about ERP programs, start with Your ERP Will Go Live. The Payback Is Decided Before It Starts.
Mid-migration? See what can run on your current ERP now, without waiting for go-live. Book a 30-minute session.
FAQ
Should finance wait for the new ERP before adopting AI?
Usually not. Waiting locks ERP designs without AI in mind, so the processes get redesigned again when AI arrives, and the value often slips to a year or more after go-live. Starting now on a few processes, such as supplier invoice matching or customer deductions, shows where the value and data gaps are while the ERP design can still change.
Can AI speed up an ERP implementation?
Possibly by a lot. Recent research argues AI agents could cut implementation effort by at least half and halve program duration. If it holds, change management becomes the main constraint, so budget and leadership attention should shift toward training, process ownership and controls rather than build effort.
Do you need clean data before using AI in finance?
No. Recent research advises finance teams to start with the data they have, imperfect as it is. In our view, the practical version is: start with what you have, put finance in charge of governing finance data, and fix the fields that your first AI use cases expose. Early AI work tells you which parts of the cleanup actually matter.
Take the next step
Ready to Move From Exploration to Execution?
Leading manufacturers are closing the AI execution gap right now. Book a demo and see your operation transformed in 90 days.
