Getting multi-platform e-commerce orders into your ERP: three decisions to make before integrating

“We just want orders from Shopee, momo and our own website pulled into the ERP automatically. No more keying them in one by one.” I’ve heard this many times, and it is almost never a technical problem. Integrations stall because nobody makes three decisions: whether inventory is shared, whose product codes are the standard, and who is the single source of truth for returns and reconciliation. This post lays out those three decisions, what each option costs, who should own it, and why talking about APIs before the decisions are made is wasted talk.

Johnny Yang12 min read
Getting multi-platform e-commerce orders into your ERP: three decisions to make before integrating

“We just want orders from Shopee, momo and our own website pulled into the ERP automatically. No more keying them in one by one. That can’t be hard, right?”

On that video call, they kept talking while sharing their screen. Three tabs open: Shopee Seller Centre, the momo (a Taiwanese marketplace) vendor backend, the order list from their own website. Next to them, an Excel file: the shipping list that goes to the warehouse at 3 p.m. every day. Someone had reordered its columns by hand, because the warehouse only accepts that order. In the bottom-right corner, a LINE group was flashing. Customer service was asking, “Has this return actually been refunded or not? The customer keeps chasing.”

I’m Johnny Yang of Brightstream Technology in Taiwan. I’ve led my team through implementations for seven years: 252+ projects across 32+ industries. I know this screen well. And I’ll start with something that may be the opposite of what you expect: among the integration projects I’ve handled, almost none got stuck because the API wouldn’t connect. The real difficulty of e-commerce ERP integration is three decisions only you can make. First, whether inventory is shared — whether every channel sees the same number for the same stock. If it’s shared, who carries the overselling risk? If it isn’t, where are the allocation lines drawn, and who has authority to move stock between them? Second, whose product codes are the standard — the ERP item number, each platform’s product ID, the label on the warehouse box. In the end only one set can be the master. Everything else is a mapping. Third, who is the single source of truth for returns and reconciliation — when platform payouts don’t match the receivables in your system, which one wins, who adjusts, and below what difference you stop chasing. Without answers to these three, integration just moves your existing mess into the ERP faster. With answers, what’s left is an engineering problem, and the engineering problem has always been the easiest of the lot to solve.

First, you’re right: manual order entry should be eliminated

I’m not going to talk you out of integrating. What makes manual entry truly expensive is where the errors land. On a normal day of a few dozen orders, your sales support staff type them slowly and don’t make mistakes. When volume jumps during a promotional period, the same person is typing until 10 p.m., and the odds of a wrong address or a missed order shoot up. Those are exactly the days you can least afford mistakes. Late shipments go straight into your platform rating and your future visibility.

The problem is the next step. Most people then ask “which system can connect to Shopee?” instead of “once orders come in, who handles the things my sales support staff used to handle in their heads?”

With manual work, your sales support staff member is an invisible translation layer. She knows which box in the warehouse that Shopee product name refers to. She knows a bundle has to be split into three item numbers when deducting inventory. She knows to set aside any order she can’t make sense of and ask the boss in the afternoon. None of these judgments are written down anywhere, but they happen every day.

What integration does is remove that translation layer. Once it’s gone, every unwritten rule is exposed, and the system won’t guess for you. It will execute the logic you never thought through, very faithfully.

Decision one: should inventory be shared

This is the one most easily skipped, because most people think the answer is obvious — “Of course it’s shared, otherwise why integrate?” First, let’s be clear about what each option costs.

One shared inventory pool: the warehouse has 100 units, and all three channels see 100 units. The upside is that stock doesn’t sit in a channel where it isn’t selling. The cost is that sync lag becomes your risk. No integration is truly real-time. There are request rate limits, update delays on the platform side, and your own system’s processing time. On normal days you don’t notice. In the peak hours of a promotional period, the same unit being sold on two channels at once absolutely happens. And on a platform, overselling isn’t settled with a cheap apology. Cancelled orders cost you points and affect how much traffic you get afterward.

Separate allocations: the 100 units are split into portions, and each channel only sees its own. The upside is far less overselling. The cost is dead stock — channel A is flashing out-of-stock while channel B’s allocation sits untouched, and the system won’t move it on its own. You need someone who checks and adjusts on a fixed schedule. If that person doesn’t exist, allocations slowly drift until it becomes “nobody trusts the system’s numbers anyway, call the warehouse before shipping.” Most people end up in the middle: a shared pool plus a reserve. The condition is that someone reviews the reserve regularly. Run the same numbers for a whole year and you get the downsides of both.

So the real question here isn’t “shared or not.” It’s three concrete questions: Who has authority to adjust allocations or reserves? How often are they reviewed? When overselling does happen, who handles it, and in what order are orders cancelled? If you can answer these three, either option works. If you can’t, either option breaks. I take apart the trade-offs of multi-channel inventory itself in more detail in how to integrate multi-channel inventory without both sides going wrong.

There’s an even earlier question: is the inventory number in your ERP accurate right now? In most of the projects I see, it isn’t — the books show stock nobody can find, or the warehouse has goods while the system shows zero. Integrating in that state just pushes wrong numbers to three platforms faster, and on the big sale day you face complaints from three channels at once. Inventory has to reconcile first. That’s a precondition. If you haven’t even decided whether your system should manage inventory, what you need to read first is what actually separates an inventory system from an ERP, not integration specs.

Decision two: whose product codes are the standard

This decision feels the least like something to discuss. In practice it’s the one most likely to sink the project.

First, look at what it’s like on the ground. The same product is called “[In stock, free shipping] XX Plus Size Black L” on Shopee, carries the vendor part number assigned by the buyer on momo, uses a SKU someone made up when the website was built, still has the old code on the warehouse box, and in the ERP it’s A-1023. The mapping between all of these currently lives in one sales support staff member’s head, and in the “mapping_table_latest_v3.xlsx” on her desktop.

You’ll hit this on day one of integration: an order comes in, and which item number should the system deduct inventory from? There’s only one solution: designate one master. In most cases it should be the ERP item number, because inventory, cost and purchasing all hang off it. Platform listings are just the display layer. Once the master is set, every platform’s product ID becomes a mapping, and the mapping table has to move off the sales support staff member’s desktop into the system, as data someone is responsible for maintaining.

Sounds simple. The hard part is the cases that follow. Bundles blow up the most. A “3-pack deal” on the platform is one product with one product ID, but it’s three boxes in the warehouse and three item numbers in the ERP. Should the order be split automatically when it comes in? Where is that rule written? If the customer returns only one of the three, how do you record that return? There’s no standard answer, but you must have an answer — and you must have it before the spec goes out.

Free gifts and product variants are the same kind of problem. A gift added during a promotion may be just a note line on the platform order, but the warehouse has to ship it, inventory has to be deducted, and the cost has to be booked. Colour and size are options under one product on the platform, but usually separate item numbers in the ERP. If these two layers of mapping aren’t built cleanly, the typical disaster goes like this: orders come in automatically, the variants don’t match, the warehouse ships the wrong item, the customer sends it back, and everyone decides “let’s not integrate this product for now, just type it by hand.” Exceptions pile up one by one, and the integration degrades into half a system.

There’s also a process issue that matters more than the technology. If marketing can create new products directly on the platform without first creating an item number in the ERP, the mapping table starts springing holes from that moment. The order has to be reversed: the ERP gets the item number first, then the platform can list it. Marketing will feel boxed in, but this is the key rule for whether the integration survives long term.

Decision three: returns and reconciliation, who is the single source of truth

The first two decisions affect whether shipments go wrong. This one affects whether you’re actually making money.

E-commerce money flows work nothing like regular B2B. What the customer pays doesn’t go straight into your pocket. It passes through the platform, which deducts the transaction fee, the payment processing fee, its share of promotional discounts, shipping subsidies and platform coupon cost-sharing, then pays you a lump sum after some settlement cycle. Order amounts and the money you actually receive are mismatched by nature, and each order may be mismatched for a different reason.

Returns make it more complicated, because they come with time lags and several states: the customer requests, the platform approves, the goods arrive back at the warehouse, inspection confirms they can be resold, the platform claws the money back. These points often straddle month-end. So on closing day, finance holds one batch of orders where “the goods came back but the money hasn’t been deducted yet,” and another batch where “the money was deducted but we haven’t seen the goods.”

So you have to decide first: which data is the single source of truth. A sensible split is usually this. For money, the platform payout statement is the standard — that’s the money you actually received, and there’s no arguing with it. For inventory and cost, the ERP is the standard — those are your own books. Differences between the two aren’t errors. They’re normal, and they need to be explained, not eliminated. What has to be settled are three finer points: Who investigates differences — finance, e-commerce operations, or customer service? How often — daily, weekly, or all at once when it blows up on closing day? Below what amount do you just accept it and stop chasing?

What happens without answers to these three? Here’s what I’ve seen: the integration is done, orders really do flow into the ERP automatically, the boss is happy. Then month-end arrives and finance is still doing the same thing — downloading payout statements from three platforms, pasting them into Excel, and matching them against ERP receivables one by one. You paid for integration, and the most labour-intensive part of the whole company wasn’t solved at all, because it was never written into the requirements in the first place.

Only after the three decisions does technology get its turn

Once the three are settled, you’ve earned the right to talk specs — and suddenly specs are easy to write. Sync frequency is no longer a number picked by feel; it follows from the inventory decision. You don’t have to guess the product master fields; the mappings are already defined. Which interface reconciliation needs was already answered in decision three.

There are only two more things I want to flag, because they get missed almost every time. One: exception orders need somewhere to go. No matter how complete the spec, some orders won’t get in — the item number doesn’t match, the address format is odd, the platform pushes a field nobody has seen before. These orders can’t quietly vanish, and they can’t all be dumped on the engineers. The system needs a clear place for them, and someone has to check it every day. If you don’t arrange that, exception orders drift back into the LINE group and the integration slides back to semi-automatic. Two: failures have to be visible. The scheduled jobs running behind the integration will stop one day. If nobody gets notified when they stop, you’ll find out when a customer calls asking “why hasn’t my order shipped yet?”

Here are the three decisions side by side.

What to decide firstWhat a clear decision looks likeWhat happens if you don’t decide
Should inventory be sharedChoose shared or allocated; name who can adjust, how often it’s reviewed, who handles oversellingAll three channels oversell on the big sale day, or stock dies in a channel where it doesn’t sell
Whose product codes are the standardERP item number is the master, platform product IDs are mappings, split rules for bundles and gifts are fixed in writingException orders keep piling up; bestsellers become “don’t integrate for now, type it by hand”
Who is the truth for returns and reconciliationMoney follows platform payouts, inventory follows the ERP; who investigates differences, how often, and what size gets let go are all written downFinance still closes the month by matching three platforms’ payouts in Excel

So what should you do now

Tomorrow, find three people and ask nine questions

No big meeting. No vendors. Get the three people who actually deal with this every day into one room: the sales support staff member who takes orders, the warehouse keeper who manages stock, and the finance person who closes the month.

Ask sales support: What is the same product called on each platform? Where is the mapping table? How do you handle bundles and gifts? Ask the warehouse keeper: Are the inventory numbers accurate? How far off was the last stocktake? Who do you want adjusting allocations? Ask finance: How do you reconcile platform payouts now? What do you do when they don’t match? Below what difference are you willing to let it go?

Write the nine answers on one sheet of paper. You’ll find that on several questions the three people answer differently — and those differences are exactly where things will blow up after integration. Talk them into a single version first. That matters far more than which vendor you pick.

When you shouldn’t integrate yet

There are a few situations where integration won’t help you. I have to be honest about that.

If order volume hasn’t reached the point where your people are overwhelmed, you have only one main channel, and the platform’s own backend is good enough, don’t integrate yet. The value of integration scales with the number of channels, order volume and exception complexity. When all three are still small, maintenance costs more than the time it saves. If your inventory numbers are inaccurate right now, fix inventory first.

The other case is when nobody in the company has the authority to decide the three things above, and marketing, warehouse and finance each say something different. Then what you need to fix first isn’t the system. It’s who has the final say. If that isn’t resolved, integration just turns three departments’ disagreements into a real-time fight every single day.

If you can answer all three decisions, go talk specs. If you can’t, don’t circle around a meeting room on your own. Thirty minutes with someone who has done implementations is usually more useful than three months of thinking it through yourself.

Found this useful? Pass it on
Prices published · free range estimate

So what does a build to your process actually cost?

From NT$300,000, priced against your spec, with same-scope changes free during the maintenance period. The prices and the estimator are in the first screen — get a range first, you do not have to decide anything yet.

Written scope, then a fixed price · staged acceptance · the code and the data are yours · or just book a free 30-minute diagnostic

Johnny Yang (熊董)
Johnny Yang (熊董)
Founder, Brightstream Technology

Serial founder leading a small team that builds systems into an owner's everyday work.

As a founder
7 yrs
Projects delivered
252+
Industries served
32+
About Johnny