COI Tracking API vs Building Your Own ACORD 25 Parser
Read any certificate of insurance in seconds. Upload an ACORD 25 and the AI pulls the insurer, policy numbers, limits, dates and endorsement status into structured data.
PDF, JPG, PNG, BMP, HEIC, TIFF
Upload your certificates of insurance
Drop files here or click to upload
Up to 50 files
Uploading...
No certificate on hand? Loading the sample certificate...
For most product teams, calling a COI tracking API beats building an ACORD 25 parser, because the parser is the cheap 20 percent of the work and the maintenance is the other 80. Building only wins when certificate reading is your core product. If it is a feature inside a property, procurement or compliance tool, an API gets you to a working feature in days instead of quarters.
This is the decision behind the certificate of insurance API page, laid out with numbers you can replace with your own. If you want the endpoints themselves, they are in the API documentation.
What a COI parser actually has to do
The ACORD 25 is a standard form published by ACORD, but standard does not mean uniform. Agency management systems print it with different fonts, spacing and page breaks. Certificates arrive as clean PDFs, flattened scans, faxes turned into images and photos taken on a phone. A usable parser has to handle all of that and then do five jobs.
- Header fields. Named insured, certificate holder, producer, certificate date.
- Insurers. Up to six insurers with NAIC numbers, matched to the letters used in the coverage table.
- The coverage table. General liability, auto, umbrella or excess, workers compensation, each with policy number, effective and expiration dates and a set of limits.
- Checkboxes and wording. Occurrence or claims-made, additional insured, subrogation waived, plus the description of operations box.
- Everything attached. Endorsement pages that sit behind the certificate and change what the form means.
The first two are the part a weekend prototype handles. The last three are where production parsers spend their lives.
Worked example, build versus call
Take a property management software team that wants to add vendor certificate intake. The figures below are assumptions for illustration, not quotes. Swap in your own salaries and volumes.
| Line | Build in house | Call an API |
|---|---|---|
| First working version | 2 engineers for about 10 weeks, which is 20 engineer weeks | 1 engineer for about 1 week |
| Test set needed | Hundreds of real certificates across insurers and scan quality | Ten to fifty of your own to confirm fit |
| Handling a new layout | Rule or model change, retest, redeploy | Nothing on your side |
| Ongoing owner | Roughly one engineer day a week once customers upload real files | Integration code only |
| Cost of a wrong date | Your support team and your customer's claim | Same risk, so you still review exceptions |
At 20 engineer weeks, a team paying a fully loaded 4,000 dollars a week spends 80,000 dollars before the first customer uploads a scan. One engineer day a week of upkeep is another 800 dollars a week, about 41,600 dollars a year. Those assumptions are yours to change, but the shape does not: the build cost is front loaded and the upkeep never ends. The last row matters too. An API does not remove the need to review exceptions, it reduces how many reach a person.
When building is the right call
Build if reading certificates is the product you sell, if you need to run on your own infrastructure for a regulatory reason, or if your volume is so high that per-document cost dominates everything else. Outside those three cases, the engineering time usually has a better home. The same logic applies far from insurance. A team that wants to watch competitor prices and site changes does not write its own scrapers, it uses a service such as competitors.io, and the time goes into acting on the data.
What to test before you commit to any COI API
Whether you choose ours or another, run the same checks.
- Scans and photos. Feed it the worst five certificates in your inbox, not the clean ones from a demo.
- Coverage rows. Confirm general liability, auto, umbrella and workers compensation come back as separate rows, each with its own dates and limits.
- Failure behavior. Send a file that is not a certificate. You want a clear status and message, not a confident wrong answer. The COISoftware extraction returns a status of pending, processing, completed or stopped, and a stopped document carries a sentence explaining why.
- Endorsement pages. Send a certificate with endorsements attached and check whether they count as part of the same document. Here, one uploaded certificate counts as one whatever its page count, endorsement pages included.
- Token control. You should be able to limit a token to upload only or read only, and revoke it alone.
- Plan access. Check which plan includes the API. For COISoftware it is the Pro and Enterprise plans, listed on the pricing page.
Common objections from engineering teams
"We will lock ourselves in." Keep the vendor behind one internal function that takes a file and returns your own certificate object. Swapping providers then touches one file. Store the raw extraction beside your normalized record so you can reprocess later.
"Our customers want their data to stay with us." Certificates already travel by email between agents and vendors, so the sensitivity is real but familiar. Ask any provider where files are stored, whether they are encrypted and whether they are used for anything beyond your request. COISoftware encrypts stored certificates and does not sell or share them.
"Accuracy has to be perfect." It will not be, from any reader, human or machine. Plan the exception path first. Show the extracted value next to the source page, let a person correct it, and flag anything outside your expected range, such as an expiration date more than three years out. A good API makes that review short, it does not pretend to remove it.
"We only have a few hundred certificates." Then you may not need an integration at all. Upload them in the dashboard and export to Excel or CSV, which every plan allows, and revisit the API when volume or workflow justifies the engineering.
What the three calls look like
The flow is deliberately small. You POST the certificate file to the upload endpoint and receive a file ID. You POST that ID to the extract endpoint and receive an extraction hash. You GET the extraction by hash until the status reads completed, then read the fields. Header values come back as labeled fields, and the coverage lines come back as a table field with rows and cells, one row per coverage. Authentication is a Bearer token created in your dashboard. Full request and response samples are in the API documentation.
What happens after the data is extracted
Reading the certificate is step one. Step two is deciding whether it meets your requirements, and that is where most teams underestimate the work. Minimum limits differ by vendor type, and the endorsement wording decides whether an additional insured box means anything. If you would rather not build that layer either, the dashboard side of the product handles it, see COI management software and the endorsement verification breakdown. For the OCR step on its own, see certificate of insurance OCR.
Run the test on your own certificates
The fastest way to decide is to take ten of your real certificates, including a scan and a phone photo, and look at what comes back. Upload them to the free 10-document demo on the COI tracking software homepage with no credit card, and compare the fields to the paper. If the extraction fits, the API on a Pro plan returns the same result to your code.
Stop tracking COIs by hand
COISoftware reads every certificate of insurance, checks coverage and endorsements against your requirements, and flags expirations automatically. Test it on your own certificate, no account needed.