Insurance process automation: what to automate at each stage, from submission to claim

Insurance process automation uses software to move an insurance file through its stages with less manual work. The stages run from a broker's submission to the policy and, later, to any claim. At most stages, the slow part is a document that a person has to read and key in. Automation reads those documents and checks them against your rules. People see only the files that fail a check.

Each stage below links to our full guide on its documents and checks.

What insurance process automation covers

An insurance file goes through much the same stages at a carrier, a managing general agent (MGA) or an agency. A broker sends a submission, and an underwriter reviews it. Then the policy is quoted and bound, and the carrier issues it. During the policy term, the insured asks for changes, and later the policy comes up for renewal. A claim starts when someone reports a loss.

Each stage has its own documents, and many arrive as email attachments from brokers, carriers or claimants. The table shows what arrives at each stage and what your team checks there.

Endorsements can change coverage as well. IRMI, the International Risk Management Institute, publishes an insurance glossary. It defines an endorsement as a policy form that changes or adds to the provisions in other forms. So each endorsement needs a check against the change the insured asked for, just like a new policy.

Certificates of insurance are a separate job for the companies that receive them, such as property managers. They check each certificate against their contract requirements, as our page on certificate of insurance tracking shows.

Which tools automate which part

Insurance automation runs on four kinds of software. Each does a different job, and most teams use more than one.

• Core systems

  Policy administration systems quote and issue policies. Claims systems set reserves and pay claims. Both need clean, structured data.

• Workflow tools

  Workflow and business process management (BPM) tools send each task to the right person or queue. They track deadlines and show where a file is waiting. Insurance workflow automation usually means this kind of tool.

• RPA bots

  Robotic process automation (RPA) bots copy data between screens and systems, for example into a system that has no API. They follow fixed steps, so a new screen or document layout can stop them.

• Document AI

  Document AI reads the documents that come in from outside, even when each sender uses a different layout. It checks what it reads and sends the exceptions to a person. Docsumo works at this step.

Most manual work is the reading these tools leave to people. A bot can paste data into a core system, but it can't read a loss run in a layout it hasn't seen before. A workflow tool can send a submission to an underwriter, but someone still has to read it and record what it holds. Document AI does that reading. It turns the documents into checked data that the other tools can use.

Some vendors call the mix of these tools "intelligent automation" or "intelligent process automation". Our guide to RPA vs AI for document processing explains how RPA and AI work together. RPA in claims processing shows where bots stop on a claim. For named products, see our list of insurance document automation and claims tools.

How the document step works

Take a commercial property submission from a broker. The email holds the ACORD 125 application and its general liability and property sections. It also holds a statement of values and the prior carrier's loss runs. Some of these arrive merged into one PDF. The software sorts each attachment by type, then splits the merged PDF into its documents.

Next, it reads the fields, such as the named insured, the value of each building and each claim on the loss runs. A loss run table that spans several pages is joined into one table.

Then come the checks. The losses reported on the ACORD 125 should match the claims on the loss runs. The loss runs should also cover the years your guideline asks for. Say the guideline asks for five years and the broker sent four. The file waits only for that missing year, and the software emails the broker to ask for it. When the underwriter opens the file, the other checks have already run.

Docsumo does this document step, from ACORD forms and loss runs to policies and claim documents. Our page on intelligent document processing for insurance shows it on a sample submission. Files come in by email, upload or the API, and Docsumo has an inbox for emails. On the Business plan, it sorts files into the document types you name and splits merged PDFs.

On the Enterprise plan, cross-document validation checks the loss runs against the application and against your guideline. It also compares a binder or quote with the issued policy. When the guideline check finds a missing loss run year, Docsumo's Document Collection and Email agents email the broker to ask for it. Checked data goes to your systems through the API and webhooks. Docsumo isn't a policy or claims system, so it doesn't rate risks or pay claims.

Make the automation fit your process

No two insurance teams handle a file in exactly the same way. A carrier that writes contractors asks for different documents than an MGA that writes restaurants. One underwriting guideline asks for five years of loss runs, and another asks for three. Claims teams split work by line of business, and agencies check policies at different points.

Start with your document types. In Docsumo, you name the document types you want, such as loss run or contractors supplement, with no training set. That means no sample files to label first. Docsumo then sorts each incoming file into those types, a step called document classification. AI Split takes a plain-English definition of each type and splits a merged PDF. Documents it can't classify wait as Unclassified for a person to assign.

Then add your checks. Your guidelines set them, for example how many years of loss runs each line of business needs and how recent the valuation date must be. A check that's specific to your team is a step you set up in the workflow. A step can call a large language model (LLM) or run your own Python code.

Then set the review. On the Enterprise plan, case management keeps each submission or claim together in one case. For each field, you set how sure the software must be before a value passes without review. That level is the field's confidence threshold.

Our take. Choose automation that runs your process as it is. Don't pick a tool that makes your team adopt the tool's own process. The checks should come from your underwriting guidelines, and the stages from the way your team already works. If a tool can only run its own fixed steps, your people end up doing the rest by hand, outside the tool.

Which stage to automate first

Start with the stage where files wait longest for someone to read them. Pick the backlog below that looks most like yours.

How to measure insurance process automation

Measure each stage before you automate it, so you can see what changed. Two measures are enough to start. The first is how long a file waits for a person, such as the time from a broker's email to an underwriter's first look. The second is how many documents a person touches. A document counts as touched if anyone opens it to fix, check or key a value.

The share of documents nobody touches is your straight-through processing (STP) rate. Measure it for each stage on its own, because intake and claims have different documents and checks. The calculator works on the second measure. For one stage, enter a month's documents, the documents a person touched and the minutes each touched document took. It returns your STP rate and the hours of hands-on work.

Docsumo's published figure is 95%+ straight-through processing. Your own rate depends on your documents and the thresholds you set, so measure it on your files. Our guide to straight-through processing explains the measure in more detail.

What stays with your people

Automation does the reading and the comparing. People still decide whether to quote a risk and whether to pay a claim. They also handle the exceptions, such as a loss the broker left off the application.

Regulators look at decisions that AI supports, too. In December 2023, the National Association of Insurance Commissioners (NAIC) adopted a model bulletin on insurers' use of AI systems. State insurance departments can issue it to the insurers they license. It reminds insurers that insurance laws and regulations still apply when AI makes or supports a decision or action.

When insurers decide how to control their AI systems, the bulletin asks them to consider how much people take part in the final decision. It also covers AI systems that insurers buy from third-party vendors. A regulator may ask to see the insurer's contracts with those vendors, and the insurer's validation and audit records.

In Docsumo, fields below the confidence threshold you set go to a person for review. When a reviewer clicks a field, Docsumo highlights the line the value came from, and each correction improves the model. Audit logging is on the Business plan. Docsumo supplies the software, not a review team, so your own staff make the decisions. It's SOC 2 Type 2 audited and ISO/IEC 27001:2022 certified, and your documents aren't used to train shared or third-party models. The security page has the details.

List your stages and checks before you buy

Write down each stage your files pass through, the documents that arrive there and the checks your team runs. That list shows which kind of tool each stage needs, and which stage to test first. Then test a tool on that stage, with your own files and your own checks.

Book a demo with a few real files from the stage you picked, or start a free trial.

Frequently asked questions