Many procurement tools start from one default path: open the form, type the items, click send. That is fine for the simplest case — you know what you need, you have ten minutes, and you are at your desk. But it is not how procurement always happens for an SMB operator running between deliveries, invoicing, and customer service.
Real procurement has at least five distinct shapes. The right tool supports all five, because the wrong shape for the moment is what produces wrong orders, missed reorders, and the spreadsheet you swore you'd retire last quarter.
Here are the five ways to build a PO, when each one is right, and what to look for in a system that supports all of them.
Quick answer
The five practical ways to build a purchase order are: manual entry from scratch, inventory-driven recommendations, automated draft POs from reorder triggers, AI-assisted cart building, and off-platform shopping lists for supplier portals or buying trips. A strong procurement system should support more than one starting point, but every path should end in the same operating record: a living PO that can be sent, updated when the supplier replies, received against the current state, and handed to accounting with context.
1. Build it from scratch
The classic. You sit down, pick a supplier, type out the items and quantities, hit send. Used when:
- The order is unusual (a one-off — new equipment, a special event)
- The supplier isn't in your regular roster yet
- You're testing a new product before adding it to your standard order guide
What to look for: speed. A from-scratch PO should take under two minutes for a five-line order with a known supplier, including all the dispute-prevention fields (delivery date, payment terms, ship-to address). If the form makes you click through six screens, you'll do it wrong out of frustration.
2. Build it from inventory recommendations
The most common shape for replenishment orders. You open the inventory screen, see which items are below PAR or trending toward stockout, add them to a cart, and the cart becomes a PO. Used when:
- It's your regular order day (Tuesdays, every two weeks, end of week)
- You're reordering items you've ordered many times before
- The system can do the math: how much you've sold, how fast you'll keep selling, what your safety stock should be, what the lead time is from this supplier
What to look for: the recommendation has to actually be right. A recommendation engine that uses simple sell-through rate or "min/max" misses items with intermittent demand (specialty bitters, seasonal products, anything that sells in clumps). The right system uses statistical methods that classify each item's demand pattern and forecast appropriately inside the PO workflow.
When this is the dominant shape, the quality of the recommendations determines whether the buyer trusts the system. Bad recommendations push operators back to memory and spreadsheets.
3. Auto-build (the system creates the PO for you)
Step beyond recommendations: the system not only computes what you should order, it builds the draft PO automatically when conditions are met. Used when:
- You have stable, predictable replenishment and want to remove the "build the cart" step entirely
- You want auto-order on a fixed schedule (every Tuesday at 7am, the produce PO is drafted, ready for one-click approval)
- A specific item drops below a threshold and you want an immediate PO ready to send to the supplier
What to look for: trigger flexibility. Different items need different triggers — some are calendar-based, some are threshold-based, some are days-of-stock-based. A system that only supports one trigger type forces you to use the wrong rule for half your inventory.
A strong version: the system drafts the PO automatically, surfaces it for review, and the buyer clicks send. The buyer touches one control moment instead of rebuilding the cart by hand.
4. AI-assisted (talk to the system about what you need)
Newer shape, made possible by language models. You type or say "build me a PO for next week's produce based on last week's sales" or "order more of everything below 5 days of stock from Charlie's Produce" and the system drafts the order. Used when:
- You're under time pressure and need a draft fast
- The order spans multiple criteria (specific supplier, specific stock level, specific category)
- You're new to the system and don't know exactly which screen to click through
What to look for: the model needs to be grounded in your actual data. An AI that hallucinates an item that doesn't exist in your catalog or guesses a quantity is worse than no AI. The right architecture has the model consult your catalog, sales history, and supplier rules — and the operator always reviews the draft before sending.
This shape is useful for operators who are mobile, traveling, or stepping away from the back office. It works on a phone. It works while walking through the warehouse with a clipboard.
5. Off-platform shopping list
The shape many procurement tools forget exists. Some suppliers do not accept email POs — they require you to log into their web store, build the order in their checkout, and submit through their portal. Restaurant Depot, Costco Business, certain specialty distributors, and some import-direct relationships work this way.
For these suppliers, you need a shopping list — the items and quantities you want to order, in the format you can paste or reference when you're on their website. Used when:
- The supplier has a web-only ordering portal
- You want to consolidate items from multiple sources into one weekly shopping run
- You're sending a delivery driver out and need a paper list
What to look for: the shopping list needs to be live and connected to your inventory. Otherwise it is just a printed spreadsheet. The right system generates a shopping list from the same recommendations that would have produced an emailed PO, then brings the order back into the system when the confirmation arrives from the supplier.
Why one shape isn't enough
Operators often end up with a half-broken procurement workflow because they picked a tool optimized for one of these five shapes. Email-PO-only tools may not handle web-portal suppliers. Recommendation-only tools may not help with one-off orders. Manual-only tools can become slow as supplier count grows.
A complete procurement workflow needs all five, because a real SMB has all five — sometimes in the same week. The Tuesday produce order is from inventory. The Friday equipment-repair part is built from scratch. The standing weekly bar order auto-builds. The "what should I order from Charlie's?" question on a phone uses AI. The Restaurant Depot run uses a shopping list.
This is what we mean when we say LineNow is a closed-loop procurement platform — meaning each ordering shape flows into the same operating record, with supplier replies parsed into reviewable order updates and inventory updated through the receiving workflow. All five ordering shapes flow into the same procurement record. Supplier replies from supported channels attach to that record. Accounting handoff follows the same order context regardless of who built the order or how.
A diagnostic for your current setup
Three questions:
- Of the five shapes above, how many does your current system support? Most operators answer one or two.
- When you need a shape your system doesn't support, where do you go? Spreadsheets? Email? A different tool?
- How many tools and tabs are open during a normal procurement morning?
If the answers are "two," "spreadsheets," and "five tabs" — you're paying for a one-shape tool plus the operational glue cost of doing the other four shapes by hand.
The fix
A procurement system that supports all five shapes natively should still land them in the same order record, with the same recommendation engine and supplier-reply AI behind each path. LineNow is $100/month flat, all locations, all five ordering shapes, 90-day free trial.
The five shapes are not only a feature checklist. They are a description of how procurement happens in real operating weeks. The right system supports them because the wrong shape for the moment is one reason orders get delayed, duplicated, or pushed back into spreadsheets.