SIC codes assigned
at application time.
Send the filled application and bank statements the moment a submission lands. Get back the 2-digit and 4-digit SIC code with a confidence score, the evidence behind the call, and a review flag when the signals conflict — in seconds, not underwriter-minutes.
- Built for intake volume — thousands of applications a month, one REST call each
- Confidence thresholds you control: auto-accept the clean ones, queue the rest
- Every classification ships with its evidence — audit-ready, not a black box
- No labeled training data needed to start
Drop a funding application and/or bank statements here
PDF, JPG, or PNG · up to 4 files · 20MB total · processed transiently, never stored
Try it with a real application or bank statement — demo uploads are processed transiently and never stored.
Drops into your intake flow, not on top of it
No model training, no OCR pipeline to build, no new UI for your team. Classification happens between submission and underwriting.
Submit at intake
Your application system POSTs up to four filled applications or bank statements to the shared /api/sic/classify route when a submission lands. No pre-processing or client-side OCR step is required.
Get codes + confidence in seconds
Structured JSON back: 2-digit major group, 4-digit industry code, calibrated confidence, the evidence for the call, and plausible alternates when the business straddles lines.
Auto-accept or route to review
You pick the confidence threshold. Clean classifications flow straight into underwriting; conflicts and thin-signal files land in your review queue with the reason attached.
Product update: read the SIC classification launch announcement →
Structured output your system can act on
Not a label in a spreadsheet — a JSON payload your LOS or CRM ingests directly. Both taxonomy levels, calibrated confidence, cited evidence from each document, and alternates when a business genuinely straddles industries.
- 2 + 4 digitMajor group for portfolio buckets, industry code for pricing and restricted-industry screens.
- confidenceA calibrated 0–1 score — the knob your auto-accept threshold turns on.
- evidenceThe specific application field or bank-statement descriptor behind the call, per document.
- needs_reviewSet when the application and bank activity disagree or signal is thin — with the reason spelled out.
{
"business_name": "Summit Peak Plumbing LLC",
"business_description": "Residential and commercial
plumbing contractor",
"sic": {
"code_2digit": "17",
"title_2digit": "Construction — Special
Trade Contractors",
"code_4digit": "1711",
"title_4digit": "Plumbing, Heating and
Air-Conditioning"
},
"confidence": 0.94,
"needs_review": false,
"evidence": [
{ "source": "application",
"detail": "Business type listed as
'plumbing services'" },
{ "source": "bank_statement",
"detail": "Recurring deposits from
ServiceTitan; supply purchases
at Ferguson Plumbing" }
],
"alternates": []
}Priced per application, not per seat
Pay for what you classify. At underwriting volume that's roughly 35¢ per application — against minutes of underwriter time per file, and a fraction of the $1–3 per document comparable underwriting-data vendors charge.
Run 100–200 recent applications through; your underwriters spot-check the codes before any commitment.
Volume pricing for high-throughput intake through the synchronous /api/sic/classify route. Responses include confidence and review signals for your own routing. Monthly minimums apply.
Need full document extraction too? SIC classification plus the entire Dokyumi platform — custom schemas, white-label portals, and a credit volume sized to your intake. Send us a month of representative files and we'll quote against them.
Underwriting teams ask
What documents does SIC classification need?
Just what you already collect at application time: the filled application and business bank statements. The application usually states the business type directly; bank-statement activity (payees, processors, suppliers, payroll patterns) is used to verify it and to resolve the 4-digit industry code.
Do we need to provide labeled training examples?
No. The classifier works out of the box against the 1987 SIC taxonomy. To validate accuracy on your book, run a pilot: submit 100–200 recent applications and have your underwriters spot-check the assigned codes against what they would have coded. No upfront labeling work required.
What granularity is supported?
Both levels used in underwriting: the 2-digit SIC major group and the specific 4-digit industry code. Every result includes both, plus a calibrated confidence score, the evidence behind the classification, and up to two plausible alternates.
What happens when the model is not confident?
Every result carries a confidence score and a needs_review flag. You set the threshold: high-confidence classifications flow straight through, and anything below the bar is routed to your review queue with the reason spelled out — application/bank-statement conflicts, thin signal, multi-line businesses. Nobody has to trust a black box.
How do we integrate this into our intake flow?
One REST call. Your intake system POSTs up to four application or statement files to /api/sic/classify and receives a synchronous JSON response — SIC codes, confidence, evidence, and alternates. For asynchronous workflows, queue calls in your own system.
What does it cost at underwriting volume?
Pricing is per application — typically around 35 cents per submission at 8,000 applications a month, with volume-based minimums. Comparable underwriting-data vendors charge $1–3 per document for adjacent analysis, and manual coding costs minutes of underwriter time per file. Pilots are free, and full-platform extraction is quoted against your actual document mix and volume.
Are the documents stored?
Demo uploads on this page are processed transiently and never stored. In production, documents live in your own organization with org-scoped access controls; contact Dokyumi to arrange or confirm retention requirements.