Intelligent RPA with generative AI
Zapper Edge builds software robots for systems that have no API, and adds generative AI so they can read unstructured input, handle exceptions and prepare work for a person to approve, all inside the harness.
Software robots for systems with no API, using generative AI to read documents, handle exceptions and draft work for human review.
- Desktop and browser automation
- Scheduled and event triggers
- Attended and unattended robots
- Email and attachments
- PDF and scanned documents
- Web forms
- REST APIs where available
- Any chosen model
- Document extraction
- Prompted judgment steps
- Human review queue
The challenge
A great deal of back-office work still runs through screens: legacy applications, partner portals and desktop tools with no API. People copy values between windows, key in data from email and PDF attachments, and check results by eye. The work is repetitive, slow to scale and prone to quiet mistakes.
Classic robots help, but they are brittle. They follow fixed scripts, stop when a screen changes, and cannot read a free-text email or judge an unusual case. Adding generative AI fixes that, yet a robot that decides and acts on its own needs clear limits, human approval and a full record.
What it does
Screen and browser automation
Robots operate desktop and web applications the way a person does, through screens, fields, files and email, so systems without an API can still be part of an automated process.
Document and email reading
Generative AI reads unstructured input such as emails, attachments and forms, pulls out the fields the process needs and passes them to the robot in a structured shape.
Judgment steps
Where a fixed rule is not enough, a model helps decide, for example which queue a case belongs in, within boundaries your team sets and with the reasoning kept on record.
Exception handling
When a screen changes or a case does not fit, the robot can interpret what it sees, try an approved alternative or hand the case to a person, instead of simply failing.
Drafting and summaries
The robot drafts replies, case notes and summaries of what it did, so a human reviewer sees a clear proposal rather than raw activity and can approve it quickly.
Attended and unattended robots
Robots can work alongside a person on their desktop, or run on their own on a schedule or trigger, with the same rules and the same record in both modes.
How it works
Map the process
We walk through the process with your team, record the steps, systems and exceptions, and decide which steps are rule based, which need AI judgment and which stay with people.
Design the guardrails
We set what the robot may open, change and send, which steps need approval, and when a case must go to a person. These limits are agreed before anything is built.
Build and test
Robots and AI steps are built, then tested on non-production systems with realistic sample cases, including messy inputs and changed screens, until the exceptions are handled as designed.
Run and monitor
The robots go live under your approval, with each action recorded and reviewed. We tune the AI steps and fix drift as the applications and the work change.
Under the hood
- Automation layer
- Desktop and browser automation
- Scheduled and event triggers
- Attended and unattended robots
- Inputs and outputs
- Email and attachments
- PDF and scanned documents
- Web forms
- REST APIs where available
- CSV and JSON
- AI layer
- Any chosen model
- Document extraction
- Prompted judgment steps
- Human review queue
- Operations
- Secure credential vault
- Run logs and screenshots
- Git repositories
Business outcome
Systems without APIs, automated. Legacy and partner applications that cannot be integrated through an API can still take part in automated processes, without replacing the systems themselves.
Fewer broken runs. Robots that can read what is on screen and handle unusual cases stop less often than fixed scripts, so fewer processes need a person to restart them.
People focus on judgment. Repetitive keying, copying and checking move to the robot, while staff review proposals and decide the hard cases where their experience matters.
Automation you can explain. Limits, approvals and a full record of each run mean you can show how an automated decision was made, which supports your own compliance reviews.
Where it is used
- Reading supplier emails and invoice attachments, then keying the details into a finance application that has no API.
- Updating records across several legacy screens from one customer request, with a summary for the reviewer.
- Checking a partner portal on a schedule, downloading new files and routing them to the right internal queue.
- Drafting responses to routine case emails and holding anything unusual for a person to approve.
Frequently asked questions
How is this different from workflow automation?
Workflow automation connects systems through their APIs and generates integration code. Intelligent RPA works at the screen level, for applications that offer no usable API, and can run with a person alongside.
Do we need to buy a specific RPA platform?
Not necessarily. We design around your process and choose the automation approach that fits your environment and policy. The AI steps are model agnostic, so you pick the provider that suits you.
What happens when a screen or application changes?
The robot can interpret what it sees and try an approved alternative, and cases it cannot resolve go to a person. We also monitor runs so drift is found and fixed early.
Does a person approve what the robot does?
Yes where you want it. You decide which steps need approval, and the robot holds those for a reviewer. AI proposes, people confirm the sensitive steps, and every action is recorded.
See it on your own documents and data
Tell us what you want to automate and we will show you how it runs inside the harness.