Operations without spreadsheets: orders, inventory and deliveries coordinated by AI agents
Run operations without spreadsheets: AI agents coordinate orders, deliveries, inventory and quality, propose restocking and warn you before something breaks.
The problemYour operation depends on a shared spreadsheet that everyone edits and nobody fully understands.
Tuesday, 6:50 a.m. The warehouse supervisor opens the shared orders spreadsheet, the one called something like "Orders_FINAL_v7," to build the day's routes. What he doesn't know is that last night someone in sales sorted one column without selecting the rest. Three orders now carry another customer's address. The truck leaves at 8. At 11 a customer calls: their production line is down because the material never arrived.
Operations without spreadsheets aren't a tech fad. They're the answer to a very concrete problem: when coordinating orders, deliveries, inventory and quality depends on a file everyone edits, the system is only as reliable as the most distracted person of the day.
How the spreadsheet becomes the system
Nobody decides to run their operation on a spreadsheet. It happens gradually. The ERP didn't have a field for something, so a column got added. Logistics needed to see sales orders, so the file got shared. Quality started logging rejections in another tab. Five years later, the spreadsheet is the company's real operating system.
The problem isn't the tool, which is excellent for analysis. The problem is using it to coordinate. A spreadsheet doesn't alert anyone when something changes, doesn't validate that data makes sense, doesn't know who needs to act and doesn't leave a clear record of who changed what. On top of that, the real logic of the operation, like which customer always orders more in December or which supplier runs late, lives in the heads of two or three people. That's what we call tribal knowledge.
What an agent-coordinated operation looks like
An agent is an AI model inside a harness: the system that gives it context, tools, permissions and oversight. In operations, the typical design is one agent per process, all reading from and writing to the same business memory:
- Order agent: receives orders by email, web portal or messaging apps, validates them against the catalog and the customer's terms, and records them without data-entry errors.
- Delivery agent: drafts the route plan, confirms delivery windows with customers, tracks every shipment and records proof of delivery.
- Inventory agent: cross-checks stock, committed orders and supplier lead times, spots shortages before they happen and proposes restocking.
- Quality agent: records inspections and rejections, links them to the lot and the supplier, and opens a corrective action when a pattern repeats.
The key difference from the shared spreadsheet is that these agents notify each other. If quality rejects a lot, inventory deducts it, delivery knows which orders are at risk and the affected customer gets a heads-up before they call. That coordination is the idea behind a Company OS: moving from disconnected tools to a system that operates as one.
Operations without spreadsheets: a practical framework
You don't have to shut off the spreadsheet on Monday. The sensible approach is to migrate in stages:
- Inventory the spreadsheet. List every tab and every column, who fills it in and who reads it. You'll be surprised how many exist for a single use.
- Pick one workflow. Start with the one that hurts most, usually orders to deliveries, because that's where errors reach the customer.
- Write down the rules. What makes an order valid, what delivery windows exist, when a shortage counts as critical. What lives in the supervisor's head today has to be written down.
- Connect, don't replace. The agent reads from and writes to your ERP and current systems. The spreadsheet becomes a read-only report.
- Define who approves what. Routes, restocking, quality exceptions: every important decision has a named owner.
- Measure and expand. Once the first workflow works, add the next one.
Illustrative example: before and after
Imagine an industrial supplies distributor with several hundred orders a week and a central warehouse.
Before. Sales enters orders in the shared spreadsheet. Logistics copies them to another tab to build routes. Inventory gets checked on Fridays, and shortages are discovered after the order has already been promised. Quality logs rejections in a separate file almost nobody looks at. When the supervisor goes on vacation, the operation becomes fragile.
After. Orders come in through any channel, and the order agent records them already validated. At 6 a.m. the supervisor receives the day's proposed routes, with at-risk orders flagged. On Tuesday the inventory agent warns that a supply will run out the following week and proposes an order to the usual supplier, which the purchasing lead approves in two clicks. When quality rejects a lot, affected customers get a notice with a new date. The spreadsheet still exists, but only as a report.
If you're interested in the procurement side, like getting quotes and comparing suppliers before approving, we cover it in procurement with AI agents.
What to measure
The indicators that show whether the change is working are operational, not technical:
| Indicator | What it tells you |
|---|---|
| Orders with data-entry errors | Whether intake validation works |
| On-time, complete deliveries | Whether coordination between orders and inventory works |
| Shortages detected before the order is committed | Whether the inventory agent is getting ahead |
| Time from quality rejection to customer notice | Whether the agents are talking to each other |
| Restocking proposals approved without changes | Whether the inventory rules are well calibrated |
Common mistakes
- Automating the spreadsheet as it is. If today's logic has errors, the agent will repeat them faster. Review the rules first.
- Letting the agent buy. Proposing, yes; committing spend, no. Keep human approval in purchasing.
- Ignoring the exceptions. An agent that flags discrepancies only helps if someone handles them every day.
- Trying to migrate everything at once. One workflow done right is more convincing than five done halfway.
- Not documenting what the supervisor knows. If his judgment doesn't make it into the system's memory, you'll still depend on him.
What to do on Monday
Open your most-used operations spreadsheet and count how many people edit it and how many times something was corrected last week. Then pick one workflow, write its rules on a single page and name the owner of each decision.
At Addabra we build custom operations systems with agents: one agent per process, connected to your current systems, with human approval for purchasing and exceptions, and a trail for every order and every delivery. We start with a diagnosis to choose the first workflow worth taking out of the spreadsheet.
Frequently asked questions
Do I have to replace my ERP to use agents in operations?
Not necessarily. Agents connect to the systems you already use, such as your ERP, your warehouse system or even existing spreadsheets during the transition. What changes is that coordination no longer depends on someone updating a cell on time.
Do the agents buy inventory on their own?
No. The inventory agent proposes restocking with the suggested quantity, the supplier and the reason. A person with purchasing authority approves or adjusts the proposal before any order is issued.
What happens if a piece of data in the system is wrong?
A good harness includes validations: if an inventory count does not match the recorded movements, or a delivery is marked complete without proof, the agent flags it as a discrepancy and alerts the responsible person instead of moving on.
