Why ERP Implementation Takes Months (and What AI Changes)
Why ERP implementation takes months and runs over budget: what research says about the real causes of delays, and what an AI agent can and can't fix.
The CFO of a mid-sized manufacturer pulls a binder off the shelf labeled “ERP Rollout — Phase 1.” Inside: a 300-page requirements document, three change orders, and a go-live sign-off dated fourteen months after kickoff instead of the promised six. The system works. But half the reports are still built in spreadsheets, and the partner bills by the hour for every tweak.
If that sounds familiar, you’re in good company. Let’s look at why ERP implementation takes months and runs over budget — using research, not scare tactics — and then at which part of the pain AI actually removes, and which part nobody can remove for you.
First, about that “70% of projects fail” stat
You’ve seen it in a hundred blog posts. It’s worth retiring.
The number traces back to change-management writing from the 1990s. Researcher Mark Hughes examined five of the key sources behind it and found no reliable empirical evidence (Journal of Change Management, 2011). Gartner’s more recent framing is narrower: most ERP initiatives won’t fully meet their original business goals. “Didn’t get everything we wanted” is not the same as “failed.”
The real picture is calmer than the myth — and still uncomfortable.
How long ERP implementation really takes
| Metric | Figure | Source |
|---|---|---|
| Median ERP project duration | 9 months | Panorama Consulting, The 2026 ERP Report (170 organizations) |
| Went over budget | just over a quarter | same |
| Missed the schedule | nearly a quarter | same |
| SAP S/4HANA migration | about 1.5 years on average | ASUG, 2024 |
| S/4HANA projects over budget | 49% | ASUG, 2024 |
| S/4HANA projects finished on time | under 10% | Horváth, 2025 |
| Large IT projects: budget overrun | +45% on average, 56% less value than planned | McKinsey & University of Oxford, 2012 |
| “Black swan” projects with ~200% overrun | one in six IT projects | Flyvbjerg & Budzier, Harvard Business Review, 2011 |
The takeaway: multi-year timelines are real for large enterprise ERP in manufacturing or government. For mid-market companies, the median is around nine months. And smaller cloud systems already move in weeks: partners quote 2–16 weeks for Odoo, 2–8 weeks for a small Salesforce rollout, and NetSuite’s packaged approach targets about 100 days. Those are partner and vendor figures, so treat them as best case.
Famous disasters exist too — Lidl shut down its SAP program after spending roughly €500 million (CIO.com). But that’s a retail giant with massive customization. Using it to frighten a 50-person company isn’t honest. Smaller companies face different risks: the system never gets adopted, the budget goes to consultants, and the team drifts back to spreadsheets and chat apps.
Why implementations drag on: five real causes
1. People decisions, not configuration
Panorama’s 2026 report names organizational issues as the top cause of schedule overruns: unclear process ownership, approvals, resistance. Fewer than a quarter of companies gave change management serious attention.
In practice, the consultant waits a week for two directors to agree on who approves discounts. The system has been ready the whole time.
2. Too much scope at once
In Horváth’s 2025 study, 78% of respondents said too many topics were crammed into the project. Every “while we’re at it, let’s add inventory” adds a month.
3. Surprise extra systems
Panorama’s top cause of budget overruns: an unexpected need for additional technology. Halfway through, the team discovers the ERP doesn’t handle customer messaging, a knowledge base or HR documents — so three more tools get bought and integrated.
4. Rebuilding the old processes
Companies port everything over exactly as it was, workarounds included. Lidl’s program foundered on bending SAP to fit its own accounting model. The more customization, the more expensive every future upgrade.
5. Dependence on the implementation partner
Panorama lists weak vendor or integrator performance and unclear division of responsibility among the leading causes of delays. ASUG (2024) found consultant fees were the biggest unplanned expense.
Where the money goes
The license is only part of the bill. Salesforce partners use a rule of thumb that implementation costs one to three times the annual license. HubSpot charges a mandatory onboarding fee — around $1,500 for Sales Hub Professional. A mandatory paid onboarding is the vendor admitting the product doesn’t take off on its own.
Here’s where time and money typically go:
| Work | How it’s usually done | How long it takes |
|---|---|---|
| Discovery and requirements | Business analyst interviews, process mapping | weeks |
| Org chart and access matrix | Spreadsheet from HR, sign-off from managers | days of work, weeks of approvals |
| Job descriptions | HR or legal, often postponed | a few hours per role |
| Sales scripts | Sales lead or outside consultant | days per pipeline |
| Knowledge base | “Everyone chips in” | weeks to months, often never finished |
| Data migration | Exports, cleanup, import | depends on data quality |
| Training | Sessions, guides | ongoing |
There’s no independent study measuring the time spent on job descriptions, scripts and knowledge bases specifically — these are estimates based on typical effort. But anyone who’s lived through a rollout will recognize the list.
What an AI agent changes — and what it doesn’t
The honest answer: AI reduces the effort, not the complexity. Bain put it that way when reviewing SAP’s 2026 announcements, and it’s hard to argue with.
What the agent does faster than a person:
- builds the org chart and permissions from a staff list and managers’ answers;
- sends targeted questions to the right employees and follows up, instead of waiting for a workshop;
- configures pipelines, task lists and document templates from a plan you approve;
- turns chat and call history into draft scripts and knowledge articles;
- checks that a change in one service didn’t break another.
What the agent can’t do for you:
- decide who approves discounts;
- convince a salesperson who doesn’t want to use a new system;
- clean up data that’s been entered however-it-goes for ten years.
That’s why we don’t promise “everything in two weeks.” We promise a first working setup in 14 days: org chart, permissions, people, CRM pipelines, tasks, knowledge base and scripts. The decisions that were always yours stay yours. The day-by-day version is in AI implementation in 14 days.
We’re also not the only ones trying this. Salesforce offers conversational setup for admins; Doss builds an operations configuration from company data; a wave of AI-native CRMs assemble themselves from email and calls. Where we differ is the combination: the whole business system, org chart, permissions and a knowledge base built from chat history — in one conversation.
How to shorten any implementation, AI or not
If you’re choosing a system now, this checklist works with any vendor:
- Start narrow. Sales, tasks and the org chart first. Inventory and production in phase two.
- Appoint one owner who can decide, not just forward questions.
- Stay close to the standard product and defer customization until after go-live, when you know what you actually need.
- Ask what will be live on a specific date — as a list, not “phase one.”
- Count every tool you’ll need around the core system: messaging, documents, knowledge base. Surprise systems are the top source of overruns. More on that in Eight tools instead of one.
- Ask how to undo a change. If the answer is “through the partner,” plan for the delay.
Where PRANUS fits
In PRANUS, tasks, people, CRM, messaging, documents, finance and ERP are one system. So surprise systems don’t appear halfway through — messaging, the knowledge base and HR documents are already there. Permissions follow the org chart, so you don’t configure them separately in each tool.
An agent runs the implementation: every change is a plan, your approval, the change itself and a one-click undo. Manufacturing and costing in OPTIMUS are configured through the same conversation, as a separate phase — no “factory in two weeks” promises. Terms are on the implementation page.
The bottom line
Implementations take months not because the software is bad. They stretch because decisions don’t get made, scope keeps growing, and extra tools surface mid-project. An AI agent is good at removing configuration busywork. The decisions stay with you — and that’s how it should be.
Want to know what you could realistically run in two weeks? Book a demo. We’ll look at your setup and tell you plainly what fits in phase one and what doesn’t.