In most of the companies we get called into, nobody says "we have an integration problem". They say "Chris can't keep up with invoicing", "we keep losing track of stock", "we have no idea what's actually happening in the online shop". Underneath every one of those sentences sits the same thing: two or three systems that don't talk to each other, and the gaps between them are patched by a human. In Excel. In the evenings.
This post is about when "manual gluing" stops being a saving and starts being the most expensive thing in the company — and what to do about it sensibly if you run an SME, not a corporation with an 800-thousand-zloty data warehouse budget.
How to spot that your systems aren't talking
It's usually not one dramatic outage. It's small inconveniences that pile up over years until somebody notices the office spends 60% of its time on data and only 40% on the business. The signals we see most often:
- The same invoice, order or customer number is typed by hand into two or three systems.
- Every person in sales and the warehouse has their own Excel — and each of them believes theirs is "the current one".
- Stock in the ERP is different from stock in the physical warehouse and different again from stock on the online shop. Nobody knows which is right.
- Monthly board reports require two days of data cleaning by someone who should be selling.
- A new hire spends a week learning "our systems" because they have to memorise where to copy what.
Each signal looks innocent on its own. Together they mean your company is paying a hidden tax on manual work — in hours, in errors, and in losing trustworthy historical data.
The real cost of "gluing" — a real example
At one of our clients — a 12-person food production company — pre-integration looked like this: orders came in by email or via a B2B platform, someone retyped them into Comarch Optima, printed a warehouse pick list, the warehouse worker filled in shipment details by hand into Excel, and in the evening a second person loaded that Excel back into Optima. Every day was 2–3 hours of work for the "retypers", plus a distracted warehouse worker scribbling notes on the side. Plus the errors: wrong product, wrong quantity, invoice issued at the old price.
After integrating Comarch Optima with their WMS — over REST API, with simple SKU mapping and a single proper product catalogue — they got back roughly 12 hours a week. Those weren't overtime hours to be eliminated. Those were hours people had previously stolen from "creative" work (talking to customers, planning production) just to keep up with paperwork. Within six months, the number of shipping mistakes dropped enough that the company stopped budgeting separately for corrections. The project itself paid for itself in 4 months.
This isn't an exceptional case. It's the typical effect when you connect two systems between which the same data is already moving daily — only via a human.
Three ways to connect systems
In SMEs, integrations get done in one of three ways. None is "the best" — each is best in a different situation.
- File exchange (CSV, XML). One system drops a file on disk or FTP, the other picks it up. Cheap, simple, sufficient in many cases. The downside is delay (usually hourly or nightly) and the fact that one bad row can blow up the whole batch. Good for reporting, for syncing price lists, for nightly accounting dumps.
- REST API (or SOAP, if the system is older). The systems talk in real time — when an order is created in one, the other knows about it within a second. Comarch Optima, Subiekt GT, enova365 all have documented APIs. So do the e-commerce platforms (PrestaShop, Shopware, WooCommerce). More expensive to build, but it's the standard if you want the company to operate in real time.
- Middleware or iPaaS. A third layer is placed between the systems that translates one format into another, logs, retries on failure and lets you add new integrations without breaking old ones. It makes sense if you have four or more systems to wed together — it doesn't make sense if you only have an ERP and a shop.
In practice, most projects we do for SMEs are a mix: the steady, real-time things go via API, and monthly reports still travel as a file because they only happen once every thirty days anyway.
What's most often missing in integration projects
Integrations rarely fail because of "bad technology". They fail because somebody forgot something that seems obvious:
- No unique identifier. "Jan Kowalski Sp. z o.o." in the ERP and "Jan Kowalski sp. z o. o." in the CRM are two different companies as far as the system is concerned. Without a shared key (tax ID, customer code, internal ID), the integration will pass garbage through, and nobody will clean it up later.
- No logs. When the sync fails at 3 a.m., you want to know on which row and why. If the log is just "OK" or "NOT OK", debugging will take a week instead of an hour.
- No retry and no error queue. A momentary network glitch shouldn't mean you lose 30 orders. A sensible integration retries on its own and parks the records it can't load for manual review.
- No data owner. Product price changes in three places — ERP, shop, sales price list. Which one is "the truth"? Without that decision, the integration will overwrite one with the other every day and create chaos faster than a human ever could.
- No contact at the ERP vendor. Comarch Optima has an API, but sometimes you have to reach out to a partner for access or to clarify a nuance. A project where nobody has that contact can stall for weeks over a single configuration setting.
Where to start
Don't start with "let's integrate everything". Start with the one pair of systems between which data already moves manually today. That will be the integration with the fastest payback, because you're already paying the real cost — you just don't see it.
Concretely:
- Make a map: which systems you have, who copies what to where, every day. That's two hours of conversations with the team, not a quarter-long project.
- Pick the one pair with the most pain — most often it's ERP + warehouse or ERP + online shop.
- Set one data owner for each critical field (customer, product, price).
- Run a pilot on a subset (e.g. a single product line). See whether the logic holds before you let the whole company near it.
- After the pilot, measure: how many hours a week did you save, how many errors went away. Without that measurement you can't prove the value — and the next integration simply won't happen.
This isn't a six-month project. The first sensible integration in an SME — Comarch Optima ↔ shop, or ERP ↔ WMS — is usually 4–8 weeks and a budget in the range of tens of thousands of zlotys. That's far less than the company is losing every year on manual retyping that nobody adds up.
If you don't know which pair to start with in your business, we'll help you make the map. The first call is free and takes half an hour.
"Gluing systems" with Excel isn't a saving. It's a loan you repay every day with someone else's time.