What is intelligent process automation (IPA)? How it works, with examples
For operations leaders whose work starts with documents. See how intelligent process automation handles one insurance file, and where it fits next to RPA.
Key takeaways
- Intelligent process automation (IPA) adds AI to process automation. The AI reads documents and decides routine cases, and people handle the exceptions.
- Intelligent automation (IA) is another name for the same idea. You'll also see smart automation and cognitive automation.
- The difference from RPA is reading and judgment. RPA repeats fixed steps on structured data, while IPA also reads PDFs, scans and emails.
- In the insurance example below, four checks pass and one fails. The loss run covers four of the five policy years the guideline asks for, so only that gap goes to a person.
- Start with one process where documents delay the work. Measure it first, then track the share of items that go through with no manual work.
On this page
- What is intelligent process automation?
- How intelligent process automation works
- One insurance submission, from inbox to underwriter
- The technology behind intelligent automation
- Intelligent process automation vs RPA
- Intelligent process automation examples
- Benefits of intelligent process automation
- Intelligent process automation tools and platforms
- How to start with intelligent process automation
- If your process never touches a document
- Frequently asked questions
Intelligent process automation (IPA) uses AI for the parts of a process that used to need a person to read or judge. Rules-based tools such as robotic process automation (RPA) run the routine steps. People handle the exceptions.
You'll also see it called intelligent automation (IA). The two names mean the same thing. In finance and insurance, the steps that need judgment usually start with a document, such as an invoice, a loan file or a claim.
What is intelligent process automation?
Most processes mix two kinds of work. Take an invoice. Someone reads it and decides whether it's right, which takes judgment. Then someone keys it into the accounting system and sends it for approval, which follows fixed rules. Older process automation, such as an RPA bot or a workflow tool, could only do the second kind. IPA does both, so people step in only for the exceptions.
IBM names AI, business process management (BPM) and RPA as the three parts of intelligent automation. In IBM's words, the aim is "to streamline and scale decision-making across organizations."
The idea goes by other names too. IBM notes that intelligent automation is sometimes called cognitive automation, and some vendors call it smart automation. A newer label is agentic process automation. It adds AI agents, which choose their next step from what they find, within limits you set. The goal is the same as IPA, with more of the judgment given to AI.
How intelligent process automation works
- Invoices
- Loan files
- Claims and submissions
- 01AI sorts and reads each document
- 02Rules check the data
- 03A person handles exceptions
- 04A bot or API enters the data
Work comes in by email, upload or an API call. AI sorts each document by type and reads its fields. It gives each value a confidence score, which shows how sure the AI is of that value. Rules then check the values against the rest of the file and your own policies. Items that fail a check, or that the AI isn't sure about, go to a person with the reason. Items that pass, and the ones a person has fixed, go into the system of record through an API or a bot.
One insurance submission, from inbox to underwriter
Here is one case, followed from start to finish. A broker emails a submission for Cedar Ridge Storage LLC, a commercial property and general liability account. The new policy would start on November 1, 2026. The email holds the application (an ACORD 125), the prior carrier's loss run and other documents. The software splits the email into its documents and reads the loss run claim by claim.
| Field | Extracted value | Confidence |
|---|---|---|
| Carrier | Quillmere Mutual Insurance Co. | |
| Named insured | Cedar Ridge Storage LLC | |
| Policy number | QCP-4471082 | |
| Lines | Property, general liability | |
| Valuation date | 2026-09-15 |
| Field | Extracted value | Confidence |
|---|---|---|
| Claims on the report | 2, both closed | |
| 26-01193 · date of loss | 2026-03-04 (general liability) | |
| 26-01193 · incurred | 7,150.00 (paid 7,150.00, reserve 0.00) | |
| 25-00418 · date of loss | 2025-02-08 (property) | |
| 25-00418 · incurred | 31,250.00 (paid 31,250.00, reserve 0.00) |
| Field | Extracted value | Confidence |
|---|---|---|
| 2025–26 | 1 claim · 7,150.00 | |
| 2024–25 | 1 claim · 31,250.00 | |
| 2023–24 | No claims | |
| 2022–23 | No claims | |
| Total incurred | 38,400.00 (2 claims) |
| Field | Extracted value | Confidence |
|---|---|---|
| Paid + reserve = incurred | Yes, both claims | |
| Terms add up to the total | 38,400.00 · yes | |
| Losses on the ACORD 125 | 2 reported, 2 on the run | |
| Valuation to effective date | 47 days | |
| Terms against a 5-year guideline | 4 of 5 · 2021–22 missing |
The loss run covers four policy years and lists two closed claims, with $38,400 incurred in total. Then the checks run. On both claims, paid plus reserve equals incurred. The amounts for the four years add up to the total. The application reports two losses, and the loss run shows the same two. The loss run was valued 47 days before the policy would start, so it's recent enough to pass.
One check fails. The underwriting guideline asks for five years of loss history, and the file has four. The 2021–22 year is missing. Instead of rereading the submission, the underwriter sees the missing year next to the guideline and the four years the loss run does cover.
In Docsumo, the checks against the application and your guideline are cross-document validation. This feature compares each document with the rest of the file and your rules. It's on the Enterprise plan. The arithmetic checks inside the loss run are steps you set up in the workflow. The Document Collection and Email agents can email the broker to ask for the missing year.
Our take. Judge an IPA setup by what still reaches your team. The software should check the whole case and send a person only what failed, with the evidence attached. In this file, that's one missing year and the guideline behind it. If people still open every file to be safe, the process is only partly automated.
The technology behind intelligent automation
IPA combines several kinds of software. One platform may include a few of these parts and connect to the rest.
AI and machine learning
Predict and classify from data, such as which support requests are urgent.Document AI
Sorts documents by type and pulls checked data from invoices, bank statements or ACORD forms. It's also called intelligent document processing.Language models
Read and write text, such as pulling the request out of an email or drafting a reply.Workflow and BPM
Run the steps and approvals, such as sending an invoice to the right approver.RPA
Software bots that operate applications with no API, such as keying data into an old claims system.Integrations and APIs
Move data between systems, such as posting an approved invoice to the ERP.People
Review exceptions and low-confidence results, such as a flagged field on a loan application.
Intelligent process automation vs RPA
RPA repeats fixed steps. IPA adds the ability to read documents and make routine decisions, so it can take on work RPA can't.
| RPA | Intelligent process automation | |
|---|---|---|
| Inputs | Structured data on screens and in files | Structured and unstructured, including PDFs, scans, emails and images |
| Decisions | Fixed rules only | Rules plus predictions and classification |
| Change | Breaks when screens or formats change | Adapts to new layouts and formats more easily |
| Scope | Single tasks | Whole processes |
| Best for | Repetitive steps in systems without APIs | Processes that start with documents or need routine judgment |
RPA isn't obsolete. In IPA, it's one tool among several, used where a system has no API. For how bots and document AI split the work, see RPA vs AI for document data extraction.
Intelligent process automation examples
Each of these processes starts with documents. Each still ends with a decision that a person makes.
| Process | What IPA reads and checks | What a person still decides |
|---|---|---|
| Accounts payable | Invoices, matched to purchase orders and receipts, then routed for approval | Invoices that don't match, and the approval itself |
| Lending | Bank statements, pay stubs and tax returns, with income cross-checked | The credit decision |
| Insurance underwriting | ACORD forms and loss runs, checked against the application and the guideline | Whether to quote, and on what terms |
| Insurance claims | Loss notices, sorted and read, then sent to the right adjuster | Coverage and the claim payment |
| Customer onboarding | ID documents and proof of address, checked against the application | Cases that fail a check |
| Customer service | Emails, sorted by request and routed or answered | Complex or sensitive requests |
Benefits of intelligent process automation
Manual work or RPA alone
- A person reads each PDF and email before anything happens
- Work waits in inboxes between steps
- Values are keyed by hand, and errors show up later
- More volume means more people
Intelligent process automation
- Documents and emails are read and sorted as they arrive
- Each step starts as soon as the last one finishes
- Data is captured once and checked, and every step is logged
- People handle the exceptions, so volume can grow faster than the team
AI use is now common, but fewer organizations have spread it across the business. In McKinsey's 2026 State of AI survey, 89% of respondents said their organizations use AI regularly in at least one business function. Among organizations that use AI, only 44% said they were at least scaling it across the enterprise. The others were still experimenting or running pilots. McKinsey surveyed 1,719 people in 97 countries in May and June 2026.
McKinsey counts 6% of respondents as AI high performers. They say 5% or more of their organization's EBIT (profit before interest and taxes) comes from AI use, and that AI has brought significant value. Nearly three-quarters of these high performers said they had fundamentally redesigned workflows because of AI. Only a quarter of the other respondents said so. IPA is that kind of redesign. It changes the whole process, rather than adding an AI tool to one step.
Intelligent process automation tools and platforms
There isn't one IPA product. The tools come from different starting points, and a working setup often combines two or more. Start with the kind that fixes what slows your process down most.
What slows your process down most?
Docsumo is an intelligent document processing (IDP) platform, so it covers the document steps. Files come in by email, upload or API. On the Business plan, it classifies and splits mixed files, such as one PDF that holds several documents. Documents it can't classify wait as Unclassified for a person to assign. It reads every field, and fields below the confidence threshold you set go to your own reviewers.
On the Enterprise plan, cross-document validation checks the documents in a case against each other. Checked data then goes to your other systems through the API and webhooks. Docsumo runs in the cloud only.
How to start with intelligent process automation
- Pick one processHigh volume and document-heavy, with clear rules and a visible cost today.
- Measure a baselineCycle time, cost per item and error rate, before anything changes.
- Automate the document step firstIt's usually where the work waits. Docsumo's document AI reaches 99% field-level accuracy on 250+ document types and 95%+ straight-through processing.
- Connect the rest of the workflowSend data to the system of record through API and webhooks. In Docsumo, you can add your own checks as workflow steps that call a large language model (LLM) or run your own Python code.
- Design the exception pathDecide who reviews what, with the source document in view.
- Track the straight-through rateThe share of items that go through with no manual work. Add more processes once the rate is stable. See straight-through processing.
If your process never touches a document
A process that never touches a document may not need much AI at all. A workflow tool or an RPA bot can run a process whose inputs are already clean data. IPA is worth it where a person has to read something before the work can move. If that describes most of your queue, start with that step and test it on your own files.
Book a demo with a few of your own documents, or start a free trial.
Frequently asked questions
What is intelligent process automation?
Intelligent process automation (IPA) adds AI to process automation. The AI reads documents and decides routine cases, so a whole process can run with people handling only the exceptions.
What is the main purpose of intelligent process automation?
Its purpose is to run a whole process with as little manual work as possible. AI does the reading and the routine checks, so people spend their time on exceptions and decisions.
Is intelligent process automation the same as intelligent automation?
Yes. Both terms describe AI combined with process automation to run whole workflows. IBM notes that intelligent automation is sometimes called cognitive automation.
What is an example of intelligent process automation?
Insurance submission intake is a common one. The software splits the broker's email into documents, reads the loss runs and checks them against the application and the underwriting guideline. The underwriter sees only what failed, such as a missing policy year. Accounts payable and loan file review are other common examples.
What are the benefits of intelligent process automation?
Work moves without waiting for someone to read and retype each document. Data is checked before it reaches your systems, and every step is logged. People handle the exceptions, so volume can grow faster than the team.
What is an intelligent automation platform?
It's software that brings the parts of intelligent automation together, so one process can be built and run in one place. It usually includes AI models, document processing, workflow, connectors or bots, and a review queue. Platforms differ mainly in where they started, whether in RPA, workflow or document AI.
Sources
- IBM: What is intelligent automation?
- McKinsey: The state of AI in 2026: On the road to ROI (August 2026)
- UiPath: An Introduction to Intelligent Automation
First published . Last updated .