TL;DR. If your legacy ERP or portal still does its job but your team spends hours keying data into it, you probably don't need to replace it to use AI. You have four options: replace the system, integrate through an API, script it with RPA, or put an AI agent on the screens your team already uses. Replacement fixes the system. The other three fix the work around it. Start with one workflow and measure it before deciding anything bigger.
Most "legacy modernization" conversations start with the system and end with a multi-year replacement plan. Meanwhile the actual pain, people copying data from documents into old screens all day, keeps going.
I run Skcript Applied Labs, and we build the fourth option below. So read this as a buyer's guide from someone with a stake, and judge the options on your own workflow.
The four options
| Option | What it changes | Fits when | Watch out for |
|---|---|---|---|
| Replace or rebuild | The core system | The system blocks the business: it can't support a new process, or support has ended | Takes years, and the manual work continues until it lands |
| Integrate through an API | How data gets in and out | The system has a usable API or you can get database access | Many older ERPs and government portals have neither |
| Classic RPA | Who clicks the buttons | Screens never change and inputs are already clean | Scripts break when a button moves, and bots can't judge a document |
| AI agent on the screens | Who does the repetitive entry | No API, messy documents, high volume, rules a person could explain | Someone on your side still owns the rules and the exceptions |
These are not mutually exclusive. A common path is to automate the painful workflow now and replace the system later, on your own schedule instead of under pressure.
What "legacy" usually means in practice
In the UAE and in Europe alike, the legacy systems behind most manual data entry fall into a few groups:
- An older ERP that finance and operations live in, customized over years, with no modern API.
- Government or free-zone portals where applications, renewals and filings are entered by hand.
- In-house tools built a decade ago that only a few people still understand.
The work around them is also similar: documents arrive (passports, invoices, purchase orders, trade licenses), someone checks them, and someone types the details into several screens.
If e-invoicing is what is forcing the question, start with a readiness check before you change systems. We built readiness checkers for the UAE and the EU.
How an AI agent works on a legacy system
The agent works like a trained operator:
- Documents arrive. By upload, email or a shared folder.
- It reads and extracts. It pulls the fields it needs and checks them against your rules.
- It decides. It works out what goes where, or flags the case if something is off.
- It operates the legacy screens. It signs in like an operator and fills each tab.
- It verifies. It reads the record back and confirms every field matches.
- A person approves exceptions. Anything unclear goes to a reviewer with an explanation.
The difference from RPA is in steps 2, 3 and 5. The agent finds fields by what they mean, not fixed screen positions, so small layout changes usually pass. It can tell when a document is wrong. And it checks its own work instead of assuming the click landed.
You can see a step-by-step example (a sample company formation, with fictional data) on our legacy ERP automation page.
Questions to ask any vendor before you start
Whoever you hire, get clear answers to these:
- Where is the data processed and stored? Ask for the specific cloud region or on-premise setup in writing, and check it against your own policies.
- How are system credentials handled? They should sit in your own secrets vault, with a dedicated account that has only the access it needs.
- What happens when the screens change? Small changes should pass. Bigger ones should be flagged, not silently mis-entered.
- What is logged? You should be able to trace every field and decision for any transaction.
- Who owns it after launch? Someone on your side has to keep the business rules current. Agree who before you start.
- What does the first step cost, and what decides the next one? A fixed scope with agreed KPIs beats an open-ended discovery phase.
Where to start
Pick one workflow that is high volume, document heavy and rules based. Then:
- Diagnostic. Map that workflow end to end: the screens, the documents, the volumes, the exceptions.
- Paid proof of concept. Run it on your real system, with two or three KPIs agreed upfront.
- Rollout. Add volume and workflows only after the numbers hold, with monitoring and a clear owner on your side.
That is how we run engagements at Skcript Applied Labs. We have been building software since 2013, and the engineers who do the work are on every call. Our pilot prices are public on the AI consulting pricing page. If you are in the UAE or Europe, those pages cover how we work in each region.
If you want a second opinion on whether your workflow fits, bring one workflow to a short call. We'll tell you if it doesn't.
