Skip to main content
A Leader’s Decision Framework for AI and ERP: Where to Start and What to Do in 90 Days
Home / Blog / Digital Architecture & ERP
Digital Architecture & ERP

A Leader’s Decision Framework for AI and ERP: Where to Start and What to Do in 90 Days

Before buying another AI tool, ask what business outcome you are trying to change. That one question separates companies that get a return from AI from companies that collect subscriptions.

This is the final part of the series on why AI will not replace proper ERP. Part 6 gave the architecture. This article converts the whole series into a decision framework and a 90-day plan that a managing director, finance head or operations lead can use on Monday.

How do I decide whether to start with AI or with ERP?

Start from the situation, not the technology. If the problem is a person’s speed at drafting or searching, start with AI. If the problem is data entered twice, slow approvals, invisible bottlenecks or control risk, start with ERP. If the problem involves money, stock or compliance, put controls first and AI second.

Situation Best starting point Why
An employee spends too long drafting or searching AI copilot The bottleneck is individual effort and the risk is low
The same data is entered in multiple systems ERP integration and master-data cleanup The root cause is fragmentation, which AI would only speed up
Approvals are slow or inconsistent ERP workflow redesign Rules, delegation and escalation fix the queue
Teams cannot see bottlenecks ERP dashboards and process KPIs You cannot improve what you cannot measure
High-volume document handling AI extraction plus ERP validation AI reads, the ERP checks and posts
Recurring decisions need judgement AI recommendation with human approval Judgement helps, accountability stays with a person
Financial, inventory or compliance risk ERP controls first, AI second Deterministic control must exist before probabilistic help

What should I test before committing to any AI project?

Use the inversion test. First define what total failure would look like, then design to avoid it. For an AI and ERP project, total failure usually means one of three things.

  1. The AI works, but nobody can trust the data it uses. Fix: audit master data first. See Part 3 and The Resource Master Audit.
  2. The AI works, but the process cannot absorb the speed. Fix: redesign the workflow and its approvals. See Part 4.
  3. The AI acts, and nobody can control or explain what it did. Fix: guardrails, identity and logs. See Part 5 and the pre-launch checklist.

Gartner’s prediction that more than 40 percent of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls, maps almost exactly onto these three failures. Each is avoidable with decisions made in the first month.

What is a realistic 90-day roadmap?

Days 1 to 30: map processes and data gaps

  • Choose one end-to-end process, such as order to cash or procure to pay.
  • Trace twenty real transactions and record every handoff, wait and re-keying.
  • Count where each master record, such as customer or product, lives.
  • Name an owner for each step and each master record.
  • Agree three process KPIs, for example cycle time, error rate and cost per transaction.

Days 31 to 60: standardise workflows in the ERP

  • Clean the master data for the chosen process.
  • Configure the standard path, approval limits and exception routes in the ERP.
  • Remove duplicate entry points and retire the spreadsheets that shadow the system.
  • Train the team on the standard path and publish a one-page procedure.
  • Measure the baseline KPIs on the new workflow.

Days 61 to 90: pilot AI in one controlled workflow

  • Pick one high-volume, low-risk step, such as document extraction or ticket triage.
  • Keep posting, matching and approval inside the ERP, and let a person review exceptions.
  • Give any agent its own identity, limited permissions and a full log.
  • Compare KPIs to the baseline. Keep the pilot only if the numbers improve.
  • Decide, with evidence, whether to widen the autonomy or the scope.

A scenario: the managing director who wanted an AI assistant

The managing director of a growing wholesaler asks for an AI assistant because staff are overloaded. The 30-day review shows that the real overload comes from the same order being typed into the sales system, a spreadsheet and the accounting package. Approvals wait for her signature, and she travels. By day 60 the order flows once through the ERP, with delegated approval limits and exception alerts. By day 90 a small AI pilot reads incoming purchase orders and drafts the sales orders for review. Staff overtime falls because the duplicate entry is gone, and the AI pilot adds a further saving on top. The assistant she first asked for would have helped with neither root cause.

Are you ready? A ten-question scorecard

Answer yes or no to each question for the one process you plan to improve. Do it with the people who run the process, not alone.

  1. Can you name the owner of this process from start to finish?
  2. Is there one written standard path, and do people follow it?
  3. Does each customer, product and supplier exist once in your records?
  4. Do you know today’s cycle time for this process, measured rather than guessed?
  5. Do approvals have defined limits and delegates?
  6. Are the key documents created in the ERP, not re-typed into it later?
  7. Do you know which exceptions occur most often and who handles each?
  8. Can you show an auditor who approved a recent transaction and when?
  9. Do you know every AI tool currently used by staff on business data?
  10. Has someone been named to be accountable for the AI pilot’s results?

Eight or more yes answers: you are ready to pilot AI in this process. Five to seven: run the ERP and data work in parallel with a very small AI trial. Four or fewer: fix the foundations first, and use the AI budget to pay for it. The score is a conversation starter, not a verdict, but it is remarkably good at ending arguments about where to begin.

Who does what during the 90 days?

Role Responsibility
Managing director or sponsor Chooses the business outcome, removes obstacles, decides whether to scale
Process owner Accountable for the process design, the KPIs and adoption by the team
Finance lead Owns the chart of accounts, the controls and the business measures
ERP consultant or administrator Configures workflows, permissions, data loading and integrations
Data owners Approve changes to customers, products and suppliers
IT or security lead Manages access, logging and vendor checks for AI services
Frontline users Test the workflow, report friction, and review AI suggestions

In a small company one person may hold several of these roles. That is fine, provided the roles are named and the conflicts are understood. If the person who approves a payment is also the person who configures the approval rule, someone else should review the configuration.

What objections will you hear, and how should you answer them?

We are too small for ERP. Size is not the test. Complexity is. If you have multiple people handling orders, stock and payments, you already have a process that spreadsheets and chat cannot hold together. Modular systems let you start with the two or three apps you need.

Our current ERP is old and people avoid it. Then the first project is adoption, not replacement. Find out why people avoid it: missing data, slow screens, awkward workflows or lack of training. Often a focused clean-up and a few workflow changes deliver more than a new product, and AI assistance can later make the existing screens easier to use.

Vendors promise that AI will do all of this automatically. Ask for evidence on a process like yours: the error rate, the human review step and the controls. Gartner has warned about agent washing, where ordinary assistants are relabelled as agents, and it expects many agentic projects to be cancelled for lack of value or risk controls. A good vendor welcomes hard questions. Our checklist of questions to ask before buying an AI agent tool is a useful starting point.

Our staff fear that AI will replace them. Be honest about what will change. In most SMEs the first effect is that people stop re-typing and start reviewing exceptions, which is more interesting work but still a change. Involve them early, show them where the time goes, and let them help choose the pilot.

We cannot afford a long project. That is an argument for the 90-day structure, not against it. One process, one pilot, one review date, and a decision based on numbers.

What does success look like at day 90?

Success is not that the AI demo impressed the board. It is a short list of measurable facts. One process now runs on a single standard path in the ERP. The master data behind it is clean and has named owners. Cycle time and error rate are measured and have improved against the baseline. One AI pilot is running under defined limits, with logging and a review routine. The team can explain what the AI does and what it must never do. And the leadership group has evidence on which to decide the next process, instead of opinions. That is a better position than most companies reach after a year of buying tools.

What breaks without ERP in this plan

  • There is no baseline to measure the AI against, so success is a feeling.
  • The pilot depends on one person’s spreadsheet and ends when that person leaves.
  • Nobody can prove what the AI changed.
  • Scaling the pilot means repeating the cleanup that should have been done once.

What AI cannot own

  • The choice of which outcome to change.
  • Accountability for results and for errors.
  • The authority limits and the control design.
  • The decision to promote a pilot into production.

What is the series in one paragraph?

AI improves individuals. ERP improves organisations. Processes need control as well as intelligence, and the database is the asset underneath. Productivity comes from process design, and agents need guardrails that only a governed system can provide. The architecture puts AI in the top layers and control in the bottom ones. The winning model is AI embedded in a well-designed ERP, which means AI should amplify a good ERP, not stand in for one. If you would like to read the series from the start, begin with Part 1.

Frequently asked questions

Should an SME buy ERP or AI first?

If your data is scattered and your process is manual, ERP first. If you already have a sound system and your pain is individual speed, an AI copilot first. Most SMEs benefit from fixing the backbone before adding automation on top.

How much should I budget for an AI pilot on top of ERP?

Keep the first pilot small: one workflow, one team, a fixed review date and a measurable target. Check current Singapore support schemes, such as the Enterprise Development Grant, for eligible projects. Our guide to the EDGE Grant explains how SMEs approach it.

How do I know if the AI pilot worked?

Compare the process KPIs to the baseline from the ERP: cycle time, error rate, rework and cost per transaction. If they did not move, the pilot did not work, however good the demo looked.

Who should own an AI and ERP programme?

A business owner of the process, not the IT team alone. IT supplies the platform. The process owner is accountable for the outcome.

What if we already have an ERP but nobody trusts the data?

Treat it as a data and adoption project first. Pick one process, clean its master data, make the required fields mandatory, name the owners, and retire the parallel spreadsheets. Trust returns when the system is the only place the answer lives.

Can the 90-day plan run on Odoo?

Yes. Odoo’s apps share one database, so a process such as order to cash can be mapped, standardised and measured across sales, inventory and accounting in one place, which makes the baseline and the AI pilot easier to compare.

A question for you: if you could change only one business outcome this quarter, which would it be, and does your current system let you measure it?

By Willie Loo, Managing Director, Mxgsoft Pte Ltd and Moxogo Pte Ltd, Singapore. Part 7 of 9 in the series Why AI Will Not Replace Proper ERP.

Two companion articles extend this series: why agentic tools cannot replace enterprise data governance and audit trails, and how personal AI workflows differ from cross-department ERP control.

Next step: talk to us about reviewing one end-to-end process and your data readiness, so you can decide whether to start with ERP, AI or both.