Morsa publishes this guide. Morsa is software with an AI agent that works in your ERP, email and Teams; it chases supplier dates under rules a person approves. Estimates are ours.
Quick answer
Unit of work: one decision, such as “is every purchase order behind a customer order confirmed?”, not a whole plant.
First pick: a decision that happens daily, costs real money when late, uses data already in the ERP, and ends in a message rather than a machine move.
Payoff: about $23,000 a year in buyer time at 150 people and about $35,000 at 250, before counting shortages caught early (our estimate; arithmetic below).
After day 90: add one decision a quarter. Machine-level work (robots, sensors, a manufacturing execution system or MES) comes later, as capital projects with their own business case.
Definitions, the industry’s 0 to 5 scale and adoption data: autonomous manufacturing: what it is and the levels.
The five levels, for one decision
Industry scales grade a whole plant 0 to 5; this table grades one decision 0 to 4, because you implement one at a time. Example: a purchase order (PO) line with no confirmed ship date.
Level | Who does what | The unconfirmed PO line |
|---|---|---|
0. Manual | A person notices, decides and acts | The buyer finds out when the part is late |
1. Visible | Software shows the problem; a person does the rest | A daily report lists PO lines with no confirmed date |
2. Drafted | Software drafts the action; a person sends it | The buyer edits and sends a drafted email |
3. Under a rule | A person approves a kind of action once (“remind suppliers 2 days before the due date”); matching actions then go out without asking | Reminders go out; anything new waits for the buyer |
4. Closed loop | Software acts, checks the result and escalates when proof is missing | The reply updates the ERP date; silence goes to the buyer by noon |
If your buyers chase from a report of late lines, that decision sits at level 1. At level 4 people still set the rules and handle every exception.
What to automate first
Score each candidate 1 to 3 on four tests, 3 being best: it happens daily, lateness is costly, the data is in your ERP, a mistake is easy to undo. Start with the highest total of 12.
Decision | Score | Why |
|---|---|---|
Confirm supplier dates on POs behind customer orders | 12 | Daily; a miss stops a line; ERP data; a wrong email is easy to fix |
Enter customer orders from email and PDF | 11 | Daily; errors are caught at acknowledgment; costs less than a line stop |
Answer “where is my order?” | 10 | Daily and easy, but a slow answer costs little |
Re-plan after a shortage or breakdown | 9 | Costly, but needs correct routings, and a bad plan is hard to undo |
Track promises made in meetings and chat | 9 | Constant, but the data is in messages, not the ERP |
Adjust machine settings or run a robot cell | 8 | Needs sensors and safety work; a mistake scraps parts |
Our scores for a typical plant; use your own costs. In our experience the first row is the best start for most plants of 50 to 300 people, because it needs only email and ERP entries. Step by step: how to automate supplier follow-up on open purchase orders.
The 90-day plan, week by week
Weeks | What happens | Done when |
|---|---|---|
1 to 2 | Pick the decision and its owner. Measure the baseline. Fix missing supplier emails and due dates. Send IT the access request. | 9 in 10 open PO lines have a due date and supplier email (our rule of thumb) |
3 to 4 | Connect the software to the ERP (Epicor, JobBOSS2, Global Shop, Business Central, NetSuite or another) and the owner’s mailbox. It drafts (level 2); the owner sends. | IT has approved access; most drafts are right |
5 to 6 | The owner approves each draft. Write the edits down as rules. | Fewer than 1 in 5 drafts edited (our rule of thumb) |
7 to 9 | Approve kinds of action as standing rules (level 3): reminders, follow-ups on silence. Anything new waits. | Exceptions reach the owner the same day |
10 to 12 | Close the loop (level 4) on routine lines: replies update the ERP, silence escalates. Start the second decision at level 2. | ERP dates match supplier replies with no typing |
13 | Compare with the baseline; widen, change or stop. | A one-page result, signed off by finance |
IT’s security review is usually the slowest step: allow two to three weeks from the request (our estimate), so send it in week 1.
What software does at levels 2 to 4, on supplier dates
When a PO behind a customer order has no confirmed date, Morsa asks the supplier from the buyer’s mailbox and writes the reply’s date into your ERP.
The buyer approves each kind of message once; after that it runs under that rule. Anything new waits.
When a date slips, the planner sees the customer orders it hits and picks the fix.
At J4S, a 120-person glass plant, on-time completion of floor commitments (dated promises such as “material by Tuesday” or “dispatch on Friday”) rose from about 30% to about 75% across about 900 commitments in four weeks (J4S customer story) [2]. Book a 30-minute walkthrough.
The data you need
How software that acts sits beside your ERP and MES is in Morsa vs MES and ERP. Before week 3, check:
Open sales orders: every line has a date the customer agreed to.
Open POs: every line has a due date and the supplier’s email.
Bills of materials: parts on your top 20 orders match what the floor uses.
Access: IT has approved mailbox and Teams read and send for the owner.
Owners: one name per supplier, customer and work center.
Who owns what, at 150 and 250 people
Role | Owns | Hours over 13 weeks (150 / 250 people) |
|---|---|---|
Plant manager or GM | Picks the decision, clears blockers, runs a weekly review | 13 / 13: 1 a week |
Process owner (buyer or planner) | Approves drafts, writes rules, handles exceptions | 27 / 27: 4 a week in weeks 1 to 2, 3 in weeks 3 to 6, then 1 |
Second buyer, when suppliers are split | Approves drafts for their own suppliers | 0 / 11: 1 a week from week 3 |
ERP admin or IT | Access, security review, data fixes | 18 / 24: more suppliers to clean |
Finance | Agrees how the savings are counted, signs off at week 13 | 2 / 2 |
What it costs and what it pays back
Item | 150 people | 250 people |
|---|---|---|
Your people’s time, 13 weeks | About 60 hours, about $3,600 at $60 an hour loaded | About 77 hours, about $4,620 |
Software for the first decision | Morsa starts with a pilot on one live problem; it is free when no implementation work is needed, then priced as a share of the savings measured in your own numbers. Get any vendor’s price in writing first | Same |
For comparison: an entry-level cobot (collaborative robot) welding cell | $52,000 to $70,000, as reported by welding-equipment maker Binzel-Abicor [1] | Same |
Payback, our estimate: assume each buyer chases supplier dates 6 hours a week, with 2 buyers at 150 people and 3 at 250, and levels 3 to 4 remove two thirds of it: 8 and 12 hours a week, times 48 weeks at $60, about $23,000 and $35,000 a year in buyer time. Use your own two-week log instead of the 6.
Often larger: shortages caught before release. Count them at week 13 and multiply by what one line stop or rush shipment costs you.
How to measure it
Measure three numbers in weeks 1 to 2 and again in week 13:
The outcome: confirmed PO lines behind customer orders ÷ all such lines; shortages found before release ÷ all shortages; then on-time delivery, as on the manufacturing KPI dashboard.
The work: hours a week spent chasing, from a two-week log.
The trust: drafts sent without edits ÷ drafts sent, and actions run under a rule ÷ all actions.
No movement by week 13 means the decision or the data is wrong: pick again.
After day 90: what to tell the CEO
In our view, the “autonomous factory” a CEO asks for means, in 12 to 24 months, that most routine decisions around the ERP run at level 3 or 4. Get there one decision a quarter, in score order: supplier dates, order entry, “where is my order?”, then re-planning after a shortage once routings are right.
Machine-level work comes after. Add an MES or machine sensors when re-planning needs live machine status, so that what they report feeds software that already acts. Treat robots as capital projects, one cell at a time, each with its own payback on labor and output.
What goes wrong
Starting with a dashboard. It moves a decision to level 1. Nothing acts.
Wrong dates in the ERP. Acting on a due date nobody confirmed is fast and wrong. Fix the top 20 suppliers’ lines first.
No owner. Drafts go stale and the team gives up.
Rules too loose or too tight. Something new goes out unapproved, or everything waits.
What to do this week
List five daily decisions and score each on the four tests.
Pick the top one and name its owner.
Send IT the ERP and mailbox access request.
Count the open PO lines behind customer orders with a confirmed date: your baseline.
Book a one-hour weekly review for 13 weeks.
Sources
Binzel-Abicor, Cobot welder price: how much will it really cost? (24 March 2025): https://blog.binzel-abicor.com/usa/cobot-welder-price-how-much-will-it-really-cost (read 8 November 2026)
Morsa, J4S customer story: https://morsa.ai/customers/j4s (read 8 November 2026)
