Comparison
Manual entry is not a mistake. It is the correct choice below a volume threshold. This page helps you work out where yours sits, with the arithmetic shown.
Manual order entry means a person opens the purchase order and types it into Shopify. It is the most common way a B2B order arrives, it asks nothing of your customer, and it handles any document a buyer can produce. It is also the only option that gets more expensive every time you grow.
Below about 15 document orders a month, keep doing it by hand. A person is cheaper, simpler and more capable than any software at that volume. Between 15 and about 100 the arithmetic turns, and where it turns depends on your loaded hourly cost and on how expensive a wrong order is in your category. Above 100 a month it is no longer close.
The measured comparison is this. About six minutes an order down to about one. An error rate of about 3% keyed by hand against 0.5% with a draft order a human approves, which is 6x fewer, and about $250 to unwind each erroneous order. On one operator handling one high volume customer, eight hours a day came down to 40 minutes. None of that assumes anyone loses their job, because a person still approves every draft before it becomes an order.
Described accurately and generously, because most readers are doing this right now and it works well enough that nothing has forced a change.
Purchase orders land in orders@ as attachments and someone works the queue, oldest first. No configuration, no vendor, nothing to renew and nothing to break.
A person reads a PDF, a spreadsheet with merged cells, a photographed page, a scan with a note in the margin, and an email with the order written as a sentence. No parser matches that range.
An operator knows which buyer code maps to which variant, that this customer means the case and not the unit, and that "usual" means last quarter's order. Fast, accurate, undocumented.
No licence, no per order fee, no contract. The whole cost is time, which is why it never appears as a line anyone reviews. It sits inside a salary that would exist anyway.
Double the orders and you need roughly double the keying hours. Nothing absorbs a spike except overtime or another pair of hands.
Six minutes sounds trivial until you break it into the eight things actually happening. Every one leaks time, and most leak accuracy too.
Purchase order, quote request, revision of yesterday's order, or remittance advice. Buyers use one thread for all four, and getting it wrong creates an order that should not exist.
On a clean PDF it is obvious. On a scan it is at an angle. On a spreadsheet it is on the third tab, above a totals row that must not be keyed. Sometimes the real order is in the email body.
A line reading 12 CS 6/750ML has three numbers in it and one of them is what you sell. The mistake here is silent, which is what makes it the most common wrong order. Twelve cases instead of twelve units looks plausible until the loading dock.
Large buyers order against their own internal codes, which mean nothing in your catalogue. Others send no code at all, just "the small blue jars, same as before". Either way it is a lookup.
The buying entity and the delivery address are often different, and a chain has a dozen. Picking the wrong store is not a keying error, it is a redelivery. It is also where the price list gets attached, or does not.
The document usually prints a price. It may be last year's, a different tier, or the buyer's aspiration. Someone compares it against that company's price list, or skips the check and finds out at invoicing.
The part everyone pictures when they say order entry, and on a long order the bulk of it. A hundred line order is a hundred lookups, in an interface built for one order occasionally rather than forty a day.
One line in every few orders does not make sense: a discontinued SKU, a quantity an order of magnitude off, a product you do not sell. Resolving it means emailing the buyer and waiting, so the order gets touched twice.
None of these eight steps is hard. That is the problem. They are repetitive and steady, and they consume the attention of the person who also knows your customers best.
Every vendor here shows a labour saving. Almost none show the error cost, which at real volume is the larger of the two. Here is both, using only numbers we can trace.
Take a wholesaler receiving 100 document orders a month. At about six minutes each that is 600 minutes, ten hours a month transcribing what a customer already wrote down. At about one minute each it is 100 minutes. The recovered time is roughly eight hours and twenty minutes a month, which you multiply by your own loaded hourly cost, the one number here we cannot supply.
Now the line nobody counts. At a 3% manual error rate, three of those hundred orders go out wrong every month. At about $250 each to unwind, averaged across the picking, the freight both ways, the credit note and the apology, that is $750 a month and $9,000 a year, never attributed to order entry. At the measured 0.5% rate, with a human approving every draft, the expectation falls to about one wrong order every two months, roughly $125 a month and $1,500 a year. The difference is about $625 a month, $7,500 a year, on errors alone.
Set that against the price. Growth is $250 a month for up to 100 orders, or $2,400 a year on annual billing, which saves 20 percent. The error saving alone covers it about three times over before a single recovered hour is counted. Hours are the number people quote in meetings. Errors are the number that shows up in the accounts.
Now run it backwards, at ten orders a month. Ten orders at six minutes is one hour a month, and automating one hour a month is not a business case. The error expectation is 3% of 120 orders a year, about four wrong orders, against about one at the assisted rate. A difference of roughly $750 a year: real money, but an expectation rather than a line on a statement. You may go a full year without an expensive mistake, the person keying ten orders knows those accounts well enough that their real rate beats the average, and the recovered hour has nowhere useful to go. Keep keying.
One more traced figure is worth having in view. On one operator handling one high volume customer, eight hours a day of order entry came down to 40 minutes.
Three inputs: your order volume, your minutes per order and your loaded hourly cost. The ratios are ours and traced, the inputs are yours, and nothing is sent anywhere.
Estimated yearly savings
$21,000
LevelOps, per year
≈ $4,800
Quoted above 100 orders a month, shown at the Growth annual rate
Return on investment
4.4x
Net gain, per year: $16,200
Nothing is sent anywhere. This runs in your browser.
Based on 6 minutes down to about 1 minute per order, and an error rate of 3% by hand against 0.5% with a human approving every draft. An erroneous order costs about $250 to unwind. Your numbers, our measured ratios.
If the result says manual entry is cheaper for you, that is the correct answer and you should act on it. Come back when the volume changes.
Below about 15 document orders a month, keep keying them. No hedging. The cheapest paid plan costs more than the time it saves, the error saving is too lumpy to bank, and you would be adding a step to a process that is not failing. To test that rather than take our word, the Free plan covers your first 5 successful orders with no expiry. Not five tidy samples: five of your genuinely worst formatted real purchase orders. Time them by hand first, then run them and time the review.
Between 15 and 100 a month is where it depends. On your loaded hourly cost, on how long your orders are, on how expensive a mis-shipped case is in your category, and on whether peak days are already breaking. A wholesaler at 40 orders a month with hundred line orders has a very different answer from one at 90 short orders.
Above 100 a month it stops being a judgement call. Ten hours of keying becomes a day a week, then two, and the error line becomes a number finance would recognise if anyone attributed it to order entry. The choice is no longer manual against automated, it is automated against another hire.
At growing volume the realistic comparison is not manual against software. It is manual, manual plus another pair of hands, and automated intake with a human approving the result.
| Manual entry | Manual entry plus a hire | LevelOps | |
|---|---|---|---|
| Setup time | None | Weeks to recruit, months to full productivity | Under an hour, plus a 2 to 5 minute catalogue sync |
| Cost shape | Labour only, rises with volume | A second salary, permanent | Free for your first 5 orders, then $50 one time or $100 to $250 a month by volume |
| Handles non standard formats | Yes, a person reads anything | Yes, a person reads anything | Yes, no template and no field mapping |
| Needs the customer to change anything | No | No | No |
| Human review before an order exists | The person is the review | The person is the review | Yes, always, a draft awaiting approval |
| Matches lines to your Shopify catalogue | By hand, from memory or a lookup sheet | By hand, less reliably while learning | Exact SKU, then semantic matching, plus learned buyer codes |
| Prices from your Shopify B2B price lists | If the person remembers to check | Often missed in the first months | Always, the document never sets the price |
| Creates native Shopify draft orders | Yes, typed by hand | Yes, typed by hand | Yes, with the source file attached |
| Audit trail of what the document said | The email, if nobody deletes it | The email, if nobody deletes it | Source file attached to the draft |
| Ongoing maintenance | None technically, but the work never stops | Management, training and cover | Vendor managed |
| Who it fails for | Above about 15 document orders a month | Volume that is seasonal, not growing | Below about 15 document orders a month |
| Best fit | Low volume and genuinely unusual orders | Growth that needs more customer facing time | The repetitive majority of emailed orders |
There are situations where a person keying the order is the right call, and no volume threshold changes that.
The truly unusual order. A hand drawn diagram on the back of a page. A photograph of a shelf with three items circled. An order that says "two of the blue ones, same as the March order but in the bigger case, and call me about the rest". These arrive with a phone call attached and need the person who took the call. Nothing in this category reads intent, and pretending otherwise is how automation loses credibility in week one.
Judgement about substitutions. A discontinued SKU on a line, and the closest variant is a different pack size at a different price. Substitute, split the shipment, hold the order, or ring the buyer. That is a commercial decision with a relationship attached, and the person who knows whether this buyer would rather wait makes it correctly. A system flagging the line is useful. A system deciding it is not.
The quantity that should be queried. An order arrives for ten times the usual volume: a promotion, a slip on the buyer's side, or a new store opening. An operator pauses because they remember what normal looks like for that account. That is the highest value thing a human does in order entry and it has nothing to do with typing speed.
Relationship work that only happens because someone is looking. Noticing that an account has ordered half its usual volume three months running, that a buyer keeps ordering a size which always comes back as a return, or that a new contact name has appeared. The first orders from a new account deserve the same attention, because that is when you learn their format and their codes.
And genuinely low volume. A handful of document orders a month is a person's job, not software's. Same arithmetic as the rest of this page, run at a volume where it points the other way.
None of these arrive as an invoice, which is why the manual entry decision usually gets revisited after an incident rather than after an analysis.
Average volume is not the problem. The Monday after a trade show is, when a normal day's queue arrives before lunch. Manual entry has no elasticity, so peaks become overtime, late shipments, or both.
The fastest person at order entry is the one who knows which buyer code means which variant, and none of it is written down. When they take two weeks off the queue does not just slow, it gets less accurate.
When a customer disputes a line three weeks later, the evidence is an attachment somewhere in a shared inbox, if it was not deleted. Nothing links the order to the document it came from.
Nobody joins a wholesaler to retype purchase orders. The repetitive part of the job is what people cite when they resign, and replacing an experienced rep costs more than the keying ever did.
A seasonal peak or one large new account arrives as a step change in keying hours. Overtime is finite and a hire is permanent, so neither matches the shape of the problem.
Verifying every line against the price list is the correct process, and nobody does it at volume because the queue is visible and the risk is not. The 3% error rate is time pressure, not carelessness.
The same eight step anatomy, with seven done before a person opens it. What is left is the review, the part where human attention is actually worth something.
Your customer keeps emailing the same document, in the same format, to the same address, and you add a forwarding rule on your shared inbox. PDF, Word, Excel, CSV, TXT, HTML or email body, plus scans and photographs with handwritten notes.
Line items, quantities, pack configurations, buyer part numbers, the purchase order number, the ship to address and requested dates come out of the document. A new format needs no setup first.
Exact SKU lookup first, then semantic matching from the description and pack configuration when there is no code. Confirm once what a buyer's part number means and it is remembered for that sender. Low confidence lines are flagged rather than guessed.
The product price, or the B2B price list attached to the resolved company, sets the line price. A price printed on the buyer's document is surfaced as a disagreement for a human and never used silently.
The result is a native Shopify draft order with the source file attached, waiting for approval. Nothing becomes an order until someone says so, which is why the measured error rate is 0.5% and not zero.
Seven steps automated, one kept. The person approving the draft is the person who used to key it, and the recovered time goes to exceptions and to customers.
A human approves every draft order. This is not a staff replacement argument and should not be read as one. A draft is not an order until a person accepts it. The colleague who keys the queue today reviews it tomorrow, with the lines already matched and priced. What changes is how many orders one person can carry, and the recovered hours go to the exceptions, the substitution calls, and the accounts nobody has phoned in a while.
The error rate is 0.5%, not zero. Documents are ambiguous and buyers make mistakes on their side, so any system claiming perfection is describing a demo rather than a Tuesday. What 0.5% against 3% buys is six times fewer bad orders reaching a customer, with the last word belonging to someone who can pick up the phone.
Migrating is a forwarding rule, not a project. Start with the two accounts whose documents your team dreads most, because that is where the difference is largest. Keep keying everything else while you watch, widen it account by account as the matching learns each sender's codes, and stop wherever the arithmetic stops working. Almost every team that automates intake keeps a manual path deliberately, for the unusual order and the new account.
An absolute timesaver. We have been manually entering B2B orders into Shopify for 6 years and I am convinced that I would be 3 years ahead if I found this app sooner. It is literally reading my PDF and uploading a 100+ item order in under 4 minutes. This just made us 10x more efficient.
Levelops is truly the app we have been waiting for. Our biggest pain point was manually entering orders from outside PDF POs. Levelops completely solves this nightmare. It has saved us massive amounts of time and frustration.
About 60% of our orders come to us as PDFs and other file types attached to emails. Before adopting LevelOps, our team spent considerable time manually processing orders by moving information from disparate customer files into Shopify. This process was suboptimal due to the time required for processing and the increased possibility of human error.
Manual entry is one of six ways an order gets into Shopify. Each page says where that alternative genuinely wins.
For the retailers who mandate it, and nothing else.
Your biggest accounts will not log in.
A template per sender, or no template at all.
Keep SPS for mandated retailers, automate the rest.
A portal and an EDI bridge. Where each one wins.
An OMS manages orders. Intake gets them in.
The category, and what to look for beyond extraction.
Between the attachment landing and the order existing.
The full grid, and the calculator on your own volume.
Loaded hourly cost, minutes per order, and the error unwind. Same arithmetic as this page, worked as a post.
Time five real purchase orders by hand, the ones your team dreads. Then run the same five and time the review. Five orders is enough to know, and it is the only test that uses your documents rather than our arithmetic.
Your first 5 successful orders are free with no expiry. If your volume is under about 15 document orders a month, we will tell you to keep keying them.
FAQ
Below about 15 document orders a month, keep keying them. A person is cheaper and better at that volume. Between 15 and 100 it depends on your loaded hourly cost. Above 100 a month it is no longer close.
About six minutes for a normal purchase order, which is the figure we measure against. It is eight tasks rather than one, from opening the attachment to resolving the odd line. Long orders take far longer.
About 3% by hand against 0.5% with PDF to Order, which is 6x fewer, and a human signs off on every draft. We never claim zero errors, because a system promising perfection loses your trust on its first miss.
About $250 to unwind, averaged across the picking, the freight both ways, the credit note and the apology. That is the line nobody counts, and at real volume it is larger than the labour saving.
No, and it is not sold as one. A human approves every draft before it exists as an order, so the person keying today reviews tomorrow. What changes is how many orders one person can carry.
Yes. The Free plan covers your first 5 successful orders with no expiry. Run five of your worst formatted real purchase orders through it and compare against what those five cost you by hand.
Keep them manual. An order with a hand drawn diagram, a phone call attached and a substitution decision in it belongs to the person who knows the account. Automation earns its place on the repetitive majority.
Use your own number. Six minutes is our measured average and experienced operators beat it on familiar accounts. Time ten real orders with a stopwatch, including the hundred line ones, and use that average.
