Systems and integration

How to integrate a legacy system that has no API, without replacing it

By · Published · Updated

A legacy system with no API can still be integrated by one of four routes: import and export files, reading its database, the software maker's own connection tool such as Sage Data Objects, or approved automation of its own screens. Which is right depends on what the maker permits, how much data moves and how often. Every route needs the system owner's written approval first.

Every time a door needs ordering, Phil types an email. He is the office manager at a door and window retailer, and the order is already on the sales system in front of him. A composite front door, anthracite grey, glazed top panel, chrome handle, hinges on the left. He types it all out again for the supplier, reads it back twice because a wrong door is an expensive door, and presses send. Two days later he rings to ask whether they received it. When the confirmation comes, he types the delivery date into his own system. He once asked whether the two systems could pass orders between them. The supplier said theirs was a legacy system with no API. It “cannot be connected to”.

Phil is made up. The typing is not. One project further down this page is real, and it is labelled as real.

Key facts

  • An API is a way for one program to talk to another. “No API” usually means software installed on your own PCs or server, written before anyone expected it to talk to a website.
  • The four routes are import and export files, reading the database, the software maker’s own connection tool, and approved automation of the screens.
  • Sage 50 is the common UK example: no web API, Sage Data Objects under the Developer Programme, and ODBC, a way of reading its database, that is read-only and unsupported.
  • Sage 200 has an API, but it signs in through Sage ID, needs that permission renewed from time to time, and cannot run fully unattended.
  • Better Systems Co. only uses a route the system owner has approved in writing, and proves it against a copy before it touches live data.

What does the typing cost?

Say Phil orders ten doors a day and each email takes six minutes to type and check. Both figures are assumptions. That is an hour a day copying out what is already written down, before he chases a single reply. A slip costs more. If he types the hinges on the wrong side, the firm pays for a door nobody can use.

What counts as a legacy system with no API?

Any system where the only official ways in and out are a person at a screen and perhaps a file it can load. “Legacy” only means older software a business still runs on. Phil deals with two: the supplier’s ordering system and, in this example, his own firm’s Sage 50. Sage 50 is desktop software with no web API; Sage’s knowledgebase and Hyperext both set out the position, and our Sage 50 page summarises it.

Pegasus Opera 3 has an API only on the SQL SE edition; the classic edition relies on its built-in import facility (TSG). Add the stock system a retired developer wrote fifteen years ago, and the older ERP, one system for orders, stock and accounts, that exports a spreadsheet and nothing else.

“No API” tells you the front door is shut. There are usually other doors, as long as the owner agrees to their use.

Even “has an API” needs checking. Sage 200 has had one since 2018, but it signs in through Sage ID using a method called OAuth, the permission has to be renewed every so often, and it cannot run fully unattended (Knit). So ask what the maker permits, and what will run on its own at 6am.

What are the four routes into a system with no API?

Route How it works Direction Typical use
Import and export files The system saves a file on a schedule, usually a CSV, a plain spreadsheet file. Other software reads it and writes files the system can load Both, in batches Prices and stock out each night; orders in
Direct database access Other software looks things up where the system keeps its records Read only, in nearly every case Looking up customers, products, prices and stock; reports
Vendor SDK or COM interface A connection tool the software maker supplies, running on a PC with the software installed Both, with the system’s own checks Creating orders, invoices and records as if typed
Approved screen-level automation Software works the system’s own screens as a person would, under written permission Both, slowly The last resort when no other route exists

How do import and export files work?

Most legacy systems can export a list as a CSV and load a file laid out to their own template. Phil’s door orders could leave as one file at 4pm, with a file of confirmations coming back. The limits are that orders move in batches, something has to start the load at the other end, and a line the system rejects must come back to Phil with the reason, not vanish.

How does direct database access work?

Other software reads the records where the system stores them. Many older systems keep their data in SQL Server, Access, Pervasive or a format of their own, which a standard connector called ODBC can read. Phil’s customer, door and price could come straight from his own system, so nobody copies them out. Reading is often tolerated and sometimes documented. Writing almost never is, because the program keeps totals, links and running numbers in step and the database alone does not, so a line written straight in can leave the books wrong for weeks. Sage 50 is typical. Its ODBC is read-only and unsupported. We use database access for lookups and reports, and do not write through it unless the maker documents that it is safe.

How does a vendor SDK or COM interface work?

An SDK or COM interface is a connection tool supplied by the software’s maker, running on a Windows PC alongside it. Sage Data Objects is the UK example, licensed under Sage’s Developer Programme, which Sage prices at £3,438 a year plus VAT from 1 August 2026. A record created this way passes the same checks as one a person typed, which makes it the most reliable way to put information in. The costs are the licence, a dedicated PC, a user licence taken up while it is connected, and re-testing when the maker releases a new version.

How does approved screen-level automation work?

When no other door exists, software can work the system’s own screens. It runs on a dedicated PC under its own login, doing exactly what a member of staff would do by hand and nothing more. It is slower than the other routes, it needs watching because a changed screen layout breaks it, and it is only ever done with the system owner’s written permission. For Phil, that means the supplier’s. This is not scraping and not a way round anyone’s terms. It is a documented method agreed with the owner for a system that has no other way in.

When is each route the right one?

Start with the highest one in the table that the owner permits and that does the job.

  • Prices, stock and completed orders moving daily or nightly, with nobody waiting: import and export files.
  • Something a website or order reader must look up there and then, such as whether a product exists and the account’s price: reading the database.
  • Creating orders or invoices as they arrive, with the system’s own checks: the maker’s own tool, where one exists and the volume justifies the licence.
  • A system with no export, no documented database and no maker’s tool, where the owner agrees in writing: screen-level automation.

Volume changes the answer. At ten orders a day a nightly import file may be fine. At fifty a day with a cut-off, each order needs to go in as it is approved, and the maker’s tool pays for itself. Our post on getting emailed PDF orders into Sage 50 works this through for the most common case.

What are the risks, and how are they managed?

The risks are the same whichever door is used, and so are the safeguards.

What can go wrong The safeguard
The same door is ordered twice after the software tries again Every order carries a reference, such as the customer’s order number, and the software checks for it before sending
Half an order goes across because the connection dropped The order is written in one go where the route allows. Otherwise it is checked afterwards and anything incomplete waits for a person
An update changes a file layout, the database or a screen The new version is tested before it reaches the live system, which is what a support plan is for
The system refuses the connection because someone has the record open The software waits and tries again, and tells a named person if it still cannot get in

Everything is built against a copy of the system or a test company first. Live data is touched only after the test run has been checked by someone who knows what the records should look like. In Phil’s office, that is Phil.

Why does every route need written approval?

Three reasons. The maker’s terms often say what may and may not connect to the software, and breaking them can void support. If the maker’s helpdesk finds a connection nobody told them about, the support conversation changes. And it is their system, or their customer’s, so they should know what is being done with it.

So before any route is used, the system owner confirms in writing that the method is permitted. For software you own, that is you, and sometimes also the maker. For a supplier’s or customer’s system, it is them, so Phil starts with a conversation with his supplier. If the answer is no, we pick another route or we do not do the work. There is no version of this where we proceed and hope.

What did this look like on a real project?

This part is real, and it is not Phil’s story. Better Systems Co.’s one completed project is a bespoke ordering website for a retailer of composite doors. Before it, a member of staff typed each confirmed order into the supplier’s ordering system by hand. That system is a legacy system with no API.

Customers now configure and place orders on the site, and those orders reach the supplier’s system by a method the supplier permitted. We kept a record of that permission. Every order is checked before it goes across, and any failure is held and reported, never lost. The case study has the detail.

To be clear about what that was, the supplier permitted the method. It did not commission the work and has not endorsed it. It is the one real example of the approach above, and it shows that the approval conversation can be had with a third party, not only with the maker of your own software.

What would Phil’s Wednesday look like afterwards?

Phil is still made up, so this is a picture of what the work is for, not a result.

The customer agrees the door on Saturday. The order goes to the supplier that afternoon, by the route the supplier approved, carrying exactly what is on the sales system. Nobody copied it out, so nobody could put the hinges on the wrong side. The delivery date arrives in Phil’s system without him. On Wednesday he rings the customer with a date. The only orders he handles himself are the ones the software has stopped, each with its reason beside it.

Neither firm bought a new system. A retailer that connects the systems it already has gives customers a date sooner than one still waiting for somebody to replace their software.

What we would do

We would begin with an automation audit at £500 to £1,250, credited against the build. We test against a copy what each system can save out, load and show to other software, tell you in pounds what the typing costs, and ask each system’s owner for written permission for the route we recommend. The report sets out the route, the risks and a fixed price. The build is our system integration service, from £3,000 for two systems with supported routes to £15,000 and above for several systems kept in step both ways. For orders, see what sales order processing automation is, our sales order processing automation page and the ROI calculator. If re-keying is the problem, start with how to stop re-keying customer orders. If the real problem is that nobody can see what all the systems know without logging in to each one, the answer may be one screen with one login rather than moving data between them. ODBC, SDK and COM are defined in the glossary.

Sources

Questions people ask

Our main system is 15 years old and the supplier has gone. Is it hopeless?

Rarely. Most older systems keep their records in a database that can be read, accept an import file, or can be worked through their own screens in a controlled way. The audit tests each route against a copy of your system and picks the most reliable. The only hard stop is a system owner who refuses permission for every route.

Is screen-level automation the same as scraping?

No. Scraping usually means pulling data from a website without its owner's agreement. Screen-level automation as Better Systems Co. uses it works a desktop program's own screens, on your machine, under your account, with the system owner's written permission, to do what a member of staff would otherwise do by hand. Without that permission we do not do it.

Will you need our passwords?

We need the access the job needs and no more. Where possible that is a separate login with limited rights, created by you. Passwords are never sent by email. They are kept in an encrypted password store and never written into the software itself. Your IT provider can see the set-up.

Can a legacy system be integrated in both directions?

Often, but putting information in is the harder half. Reading a database or an exported file is low risk. Putting information in needs a route the software maker supports, such as an import file or the maker's own connection tool, because the program applies rules the database on its own does not. Two-way flows also need a clear rule about which system wins when both change the same record.

Related guides

Find out what the typing is costing you

A 20-minute call. You tell us about your Sandra, your Dave or your Phil, and how the job gets done today. We tell you honestly whether automating it would pay, and what it would cost.