RPA + AI Tutorial: Build AI Automation Workflows...

Introduction: When Office Workers Meet RPA and AI, the World Suddenly Becomes a Better Place

Folks, have you ever had this experience? Every day you sit down at your computer and face a series of mechanical tasks: copy-paste, fill out forms, organize data, send emails, download files... By the end of the day, your eyes are strained, your hands are sore, your neck hurts, but when you look at what you've actually accomplished, it doesn't seem like much. 😭

I once worked as a "spreadsheet guy" at an internet company for three months, processing all kinds of Excel reports every day. The pain was real. It wasn't until I discovered RPA and AI automated workflows that I felt there was light at the end of the tunnel. Today, I'm going to share my battle-tested approach to building these workflows. I promise that after reading this, you'll be able to set up an AI-powered automated workflow in 3 steps and watch your productivity skyrocket 🚀.

Don't scroll away just yet — this isn't some advanced programmer-only course. I guarantee that even if you're a complete beginner, you can do this by following along. This RPA and AI combo, in plain terms, means: let robots (RPA) handle the "dirty work," and let AI (artificial intelligence) handle the tasks that require "brainpower." When you combine the two, the results are explosive — the effect of 1+1>10 is no exaggeration.

Workflow Concepts: Dissecting Your Work First

Before we start building, we need to understand one concept — what is a workflow? Simply put, a workflow is the process you follow to complete a task from start to finish. Take the "daily report compilation" you might do every day: open group chats → collect reports → copy-paste into a spreadsheet → check formatting → send to your manager. That's a workflow.

What we need to do is break down every action in this process and identify which ones are "mechanical actions" (like copy-paste, opening web pages) and which are "cognitive actions" (like summarizing, analyzing, writing comments). The logic behind combining RPA and AI is simple: mechanical actions go to RPA, cognitive actions go to AI.

Let's use the "daily report compilation" example again. In this workflow, opening group chats, scrolling through messages, copying text, and pasting into a spreadsheet — these are mechanical actions that RPA handles perfectly. But "reading through each person's report, extracting key points, and summarizing them into a few bullet points" — that's a cognitive action that we can delegate to AI.

Once you understand this, you've grasped the underlying logic of automated workflows. Many people try to jump straight to "one-click full automation" and end up making things overly complicated. My advice: don't aim for full automation right away. Start with the most annoying, most repetitive part of your work and test the waters. Once you taste the benefits, you won't be able to stop. 🤣

Core Components: RPA and AI, One by One

Since we're combining RPA and AI, let's get to know these two main characters, along with their helpful companions.

RPA (Robotic Process Automation): Your "Hands Replacement"

There are many RPA tools on the market — international ones like UiPath and Automation Anywhere, and domestic ones like Yingdao (影刀), Laiye (来也科技), and Octoparse (八爪鱼). I personally use Yingdao most often, for three simple reasons: Chinese interface, extensive community tutorials, and a free tier. For individual users, the free version is more than sufficient.

Things RPA can do include, but are not limited to:

Opening/closing software and web pages
Simulating mouse clicks and keyboard input
Reading data from Excel, Word, and PDF files
Scraping data from web pages and filling out forms
Sending emails, WeChat messages, and DingTalk notifications
Renaming, archiving, and moving files

In short, if you can do it with a mouse and keyboard, RPA can learn it. And it doesn't require coding — it's mostly drag-and-drop visual workflow design with an extremely low learning curve.

AI Large Models: Your "Brain Replacement"

The AI we're talking about here refers to the AI tools we commonly use, such as ChatGPT, Claude, ERNIE Bot (文心一言), Tongyi Qianwen (通义千问), and others. Their strengths include:

Text summarization, extraction, and translation
Content creation (writing emails, copy, weekly reports)
Data analysis and information extraction
Image recognition and OCR text recognition
Code generation (we don't need this yet, but it doesn't hurt to know)

Most of these AI tools now offer API interfaces, which means we can have RPA "summon" AI for help at any point during its execution. This is the essence of combining RPA and AI.

The "Glue": APIs and HTTP Requests

Many people get a headache when they hear "API," thinking it's a programmer's thing. But in modern RPA tools, calling APIs has become incredibly simple. You just add an "HTTP Request" component to your RPA flow, fill in the AI tool's API endpoint (usually provided with an official key), put in the text you want to send to the AI, and click run — RPA will retrieve the AI's response. The entire process requires no coding knowledge, just copy-pasting configuration parameters.

If you can't even be bothered with APIs, there's an even simpler method: many RPA tools have built-in "AI Assistant" components that come with pre-configured connections to major large models. You just select the model, enter your API key, and everything else is fully automated.

Building Steps: 3 Steps to an AI-Powered Automated Workflow

Now for the main event — how to build a truly functional automated workflow in 3 steps. Let's use a concrete example: "Automatically collect the latest AI daily news and generate a summary email." This example involves both data scraping (crawling web pages) and AI processing (generating summaries), making it highly representative.

Step 1: Process Breakdown and Design

Before you open any RPA tool, sketch out the workflow you want to automate on paper (or in a memo app). For our "AI daily news collection," the breakdown looks roughly like this:

Open specific "Latest AI News" websites (e.g., QbitAI, Synced, etc.)
Scrape the titles and links of all articles published that day
Click into the top 3 articles and scrape their full text content
Send article titles + full text to an AI tool to generate a 200-word concise summary
Fill the summary and links into an email template
Send to designated recipients via Outlook/QQ Mail

Once the workflow is designed, everything after becomes much easier. Remember one principle: keep each step as simple as possible — don't try to do everything at once. If something goes wrong midway, you'll be able to quickly pinpoint which step failed.

Step 2: Building the Workflow in an RPA Tool

Open Yingdao (or your chosen RPA tool) and create a new application. The component library is on the left, and the canvas is on the right. Drag each step we designed onto the canvas one by one:

Drag in the "Open Web Page" component and enter the URL.
Drag in the "Get Element Information (Batch)" component, select the "title" element, and set the scrape count to 10.
Drag in a "Loop" component to process the first 3 links.
Inside the loop, drag in "Open New Web Page" and "Get Text" components to scrape the article content.
Drag in an "HTTP Request" component — this is the critical part!!! Set the method to POST, fill in the AI service's API endpoint as the URL, add "Authorization: Bearer {Your Key}" in the Headers, and in the Body put "Please read the following article and generate a summary:" + the article content.
Then drag in a "Parse JSON Data" component to extract the summary from the AI's response. AI API responses are typically in JSON format, with the summary in one of the fields.
Finally, drag in a "Send Email" component, fill in the summary + links in the body, and send.

This process looks like a lot of steps, but it's all drag-and-drop. When I first built this, it took me about 40 minutes. If you've never used an RPA tool before, I recommend spending 20 minutes watching the official beginner videos to understand what each component does. Sharpening your axe won't delay your woodcutting!

Step 3: Debugging and Running

Once your workflow is built, don't rush to run it at full speed. First, click "Step Debug" and watch it execute one step at a time. This step is especially important because web page button positions, loading delays, and other factors can affect scraping results. During debugging, you might encounter:

Selector positioning failures (button not found) — in this case, you'll need to manually specify the element.
Web page loading too slowly, resulting in no content scraped — add a "Wait for Element to Appear" component.
AI returning results in an unexpected format — check whether your AI prompt is clear and whether the JSON parsing path is correct.

After debugging passes, you can set up a scheduled trigger, such as automatically running at 9 AM every day. From then on, every morning when you arrive at the office and open your email, you'll find a neatly formatted AI daily news digest with summaries. While others are still browsing websites for news, you're already sipping your coffee. The experience is simply unbeatable. 😎

Optimization Tips: Making Your Automated Workflow "Smarter"

Getting a basic RPA and AI workflow running is one thing, but we shouldn't stop there. If you want to achieve a 10x efficiency boost, you need to know how to optimize. The following tips are all lessons I learned the hard way — take a listen.

Tip 1: Treat Your AI Prompts as Core Assets

With the same AI model, different AI prompts produce wildly different results. In our automated workflow, the AI's performance depends entirely on the prompt. I recommend keeping a separate document for your frequently used prompts, such as "Daily Report Summary Prompt," "Customer Email Reply Prompt," and "Data Anomaly Analysis Prompt." Don't type them in from scratch every time — reference them as variables instead. RPA tools support reading text file contents as variables, so if you ever want to change the AI's tone or style, you only need to edit the prompt document, not the entire workflow.

Tip 2: Exception Handling Mechanisms

What does RPA fear most? Sudden web page redesigns, pop-up overlays, and network disconnections. So be sure to add "Exception Capture" components after critical steps. For example, if a web page fails to open, skip that entry and continue to the next one; if the AI service times out, retry once. When I first built my workflow without exception handling, it would get stuck midway and everything after would crash, requiring manual intervention. After adding exception handling, the success rate jumped from 70% to over 98%.

Tip 3: Use Variable Pools to Share Data

If your RPA workflow is complex and involves multiple sub-processes, I recommend setting up a global "variable pool." For example, store "user nickname," "department," and "common recipient email addresses" in the variable pool. When writing emails or generating reports, you can reference these directly, saving the hassle of repeated input. This is also a go-to method for many RPA veterans to improve operational efficiency and stability.

Tip 4: Combine OCR Capabilities to Process Image Information

In many cases, data isn't neatly formatted text — it's embedded in images. This is where RPA alone falls short. But wait — we have AI! Many RPA tools now come with built-in OCR recognition components, or you can call AI vision APIs (like GPT-4V). Feed the screenshot to AI and have it structure the information into tabular text. With this, you can even process PDF scans, broadening the scope of your workflow several times over.

Case Studies: Three Real-World RPA and AI Applications

All talk and no action is just hot air. Let's look at some real cases from my friends and colleagues to see how much RPA and AI can actually save us.

Case 1: A New Media Editor's "AI Article" Material Collection

A friend of mine who runs a WeChat public account used to dread finding material every day. She previously spent 2 hours a day browsing over a dozen websites, copying promising topics into her notes app. Now she uses RPA to build a crawler bot that automatically scrapes hot article titles and summaries from designated websites every 2 hours. Then she has AI filter out the 10 best topics based on criteria she set, such as "emotional resonance" and "trend relevance," and generate a recommendation blurb for each. She just glances at the AI-generated results each day to decide what to write about. She says this alone saves her 70% of her topic-selection time, and the open rates for her AI articles are actually higher than when she picked topics herself. 😄

Case 2: Finance Staff's Invoice and Expense Reimbursement Processing

The finance department deals with piles of invoices every day. Previously, they had to manually open Excel, verify invoice numbers, amounts, and dates one by one, then enter them into the system. Now they've built an automated workflow: a scanner converts invoices into images → RPA sends the images to AI's OCR interface → AI recognizes key fields (invoice number, amount, payee) → RPA automatically fills them into Excel and imports them into the finance system. The entire process has been reduced from 3 minutes per invoice to 30 seconds, with a recognition accuracy of 99%. The finance manager told me that once this workflow went live, they no longer need to work late into the night at month-end.

Case 3: Sales Team's Customer Information Management

What salespeople hate most is organizing customer information. They used to log into various business information lookup websites and manually copy phone numbers, emails, company sizes, and other details into the CRM system. Now they use RPA to automatically open lookup pages, scrape the information, and have AI clean up the formatting (removing extra spaces, standardizing formats) before directly entering it into the CRM. Even more advanced, AI automatically assigns tags to each customer, such as "high intent," "needs follow-up," and "potential." Salespeople just review the AI-prioritized list each day and get straight to work.

After seeing these cases, are you starting to feel the excitement? The applications of RPA and AI go far beyond these examples. Things like automatically generating meeting minutes, processing customer service tickets, and monitoring competitor pricing can all be handled with this same logic.

Summary and Outlook: RPA and AI — The Future Is Already Here

Alright, after all this writing, let's wrap things up. The so-called RPA and AI is essentially combining "hands" with "brain." RPA solves the "how to do it" problem, while AI solves the "how to think about it" problem. When the two come together, you can make the leap from "process automation" to "decision automation."

This article covered everything from workflow concepts to core components, walked you through the building steps and optimization tips step by step, and showed you several real-world cases. By now, you should have a clear understanding of how to build an AI-powered automated workflow. Remember, don't be afraid of the hassle. Pick the most painful, most repetitive part of your daily work and try the 3-step method outlined in this article. The first build might be a bit rough, but once it's running smoothly, the time and effort you save will absolutely get you hooked.

Finally, I want to say that RPA and AI aren't some distant, inaccessible black technology. They're becoming standard workplace skills, just like Office software. There are already plenty of resources on this topic, and the barrier to entry is lower than you think. The key is to start now — don't wait until "someday." After all, the best time to plant a tree was 10 years ago, and the second-best time is today. Let's get started! 💪

RPA + AI Tutorial: Build AI Automation Workflows in 3 Steps to Boost Productivity 10x