How to automate copy-paste between applications, even when there is no integration

Rank the options by how little you have to build and maintain. Start at the top, and move down only when an option cannot do the job.

The effort column is a rough order of magnitude, not a quote. It grows with the number of fields, exceptions and systems involved.

Check first, because it is the cheapest option and someone else maintains it. Look in each app's integrations or marketplace page for the other app. Then check the details: which fields it carries, which direction it syncs, how often it runs, what it does with a record that already exists and what happens when it fails. Some native integrations are limited to certain plans.

Fits when it moves the fields you need in the direction you need. Breaks when it is one-way and you need two-way, it ignores custom fields or it runs too slowly. A native sync is also hard to change: if your rules differ from its defaults, you work around it or replace it.

Many systems can export a report as CSV or Excel, sometimes on a schedule, and most can import a file. For data that moves in batches, such as a daily list of orders or a weekly payments file, this is plain and reliable.

Fits when timing is not critical and the import can match on a unique key, so a re-run updates a record instead of duplicating it. Breaks when you need near real time, the file format drifts or someone has to remember to run it. A scheduled script between the export and the import removes the person, but then you have built a small integration, and it needs the same care as one.

Often, yes. These tools join apps through ready-made connectors. A trigger, such as a new email, a new row or a new deal, starts a flow, and actions write to the other app. Zapier calls a flow a Zap, Make calls it a scenario and Power Automate calls it a cloud flow. All three handle simple branching, filters and basic data cleanup. A typical flow for email to spreadsheet: a labeled email arrives, the tool pulls out the fields and adds a row to a sheet (the data-entry guide covers the steps).

Fits when both apps have connectors, the flow is a straight line and a missed run is cheap to retry. Breaks when no connector exposes the field you need, a failed step leaves a half-finished record across systems, the same record is created twice or the bill grows with every run: Zapier counts tasks and Make bills credits. Zapier or Make vs a custom integration sets out the criteria. Testing the failure cases takes longer than building the happy path.

RPA (robotic process automation) tools such as Power Automate for desktop and UiPath drive the screen of a desktop program the way a person does: they open windows, click, type and copy values. They exist for software that has no API and no connector, such as an old Windows application.

Fits when the program has no other way in. Breaks when a screen layout changes, a pop-up appears, a session times out or the machine that runs the robot is off or locked. Each of those stops the run, and a robot that clicks the wrong thing can write wrong data without raising any error. If the system also has a web interface or an API, one of those is usually more reliable than clicking a desktop screen. Licensing usually differs between attended runs, where you are at the machine, and unattended runs, so check the current terms.

When both systems have an API and the job needs more than a straight line: two-way sync, rollbacks, volume, rules from your accounting or CRM, or code and hosting you own. A custom integration calls each system's own endpoints, so it reads and writes exactly the fields you choose, retries safely and keeps a log.

Fits when reliability matters and the rules are specific. Breaks when one side has no API, or the job is so small that a connector would do. APIs also change, so plan for versioning, rate limits and credentials that expire.

Two builds on this site show the range. Aid Through Trade keeps orders, customers and its product catalog in Brightpearl, its ERP, while sales works in HubSpot. HubSpot changes reach Brightpearl in real time through webhooks, Brightpearl data reaches HubSpot in daily jobs, and loop protection stops a note synced one way from coming back as a duplicate (Aid Through Trade). Zuniga's reps priced each pool in one tool and then retyped the proposal into HubSpot; now, when the client accepts, the app creates the contact, deal and line items in HubSpot (Zuniga).

Some systems offer no API and no connector: a vendor portal, a class-management system, a government site. Before you automate the screen, ask the vendor whether an export, a scheduled report or an API exists, sometimes on another plan or on request. If not, browser automation drives the same web pages a person would, using a tool such as Playwright or Selenium, and runs on a server on a schedule.

A reliable flow looks like this:

Two builds work this way. PlanYear's benefits consultants needed plan data across medical, dental, vision, life and more, and were copying it into spreadsheets by hand. A Playwright-based extractor logs in, keeps session state, navigates every plan type, handles the portal's Ant Design components, pagination and conditional fields and outputs clean normalized datasets (PlanYear). DeMont Family Swim School's enrolment data sat in iClassPro with no API the school could use. A service signs in, pulls the families, students and drop-list reports and syncs linked parent, household and child records into HubSpot on a schedule, with no manual exports (DeMont).

What breaks, and what to do about it:

• A changed layout. The robot cannot find the field. Build checks that fail loudly when an expected field is missing, and keep a sample page to re-test against.
• Logins and MFA. If the site requires a second factor or a CAPTCHA, an unattended run may not be possible. Ask the vendor for an API key or a service account. Do not work around a security control.
• Terms of use. Check that the vendor's terms allow automated access to your own data, and keep the request rate gentle.
• Where it runs. On a server on a schedule, not on someone's laptop.
• What a person checks. The failure alerts, and a short summary of each run: items read, items written, items set aside. If the counts do not match, someone looks. Make retries safe and match on a key, so a re-run updates a record instead of duplicating it.

No-code tools bill by usage: Zapier counts tasks and Make bills credits, so the bill follows the number of runs. Plans change, so read the current ones. RPA adds licenses and a machine to run on. API and browser builds cost a one-off build, hosting and some time when a system changes. A custom build here is a fixed price agreed in writing before any code is written.

When the options above have run out: no connector and no API, a failure that must undo itself, two systems that must stay in step both ways, or data you need to own. If that sounds like your case, an automation consultant who also writes the code can scope it on a 15-minute call. The guide to automating repetitive work covers how to choose what to automate in the first place.