Jason Auh

Somewhere in a Sensible dashboard right now, a delivery order scanned at an angle, a commission statement from a vendor nobody has ever heard of, and a five-page invoice with line items buried in a footer are all becoming clean rows of typed data. Jason Auh's job is to make sure the customer on the other end trusts that it happened.

Jason Auh runs sales and customer success at Sensible, the San Francisco company that turns documents into structured data through an API and a configuration language called SenseML. It is a deceptively simple promise - extract data from any document - and Jason spends his days on the exact spot where that promise meets reality: the customer who has a pile of paperwork, a workflow they want to automate, and a healthy dose of skepticism about whether the machine can really read a fax from 2004.

What sets him apart in a category full of confident demos is that he refuses to oversell the demo. At most document-AI companies, the go-to-market person quotes the accuracy number and moves on. Jason writes about why that number is a marketing artifact. In his own words, "Most IDP systems reach 80-90% accuracy on a heterogeneous document set without significant tuning." The next sentence is the one that matters: production accuracy, on the full mix of poor scans and issuer variants never seen at evaluation time, tends to be lower. He does not hide the gap. He maps it.

That instinct - to name the caveats out loud - is unusual for someone whose title is about closing deals and keeping customers happy. But at Sensible the two roles are fused into one, and that fusion is the whole point. The person who sells you the platform is the same person who has to make it work when your real documents show up. There is no handoff, no "that's the support team's problem." Selling honesty is cheaper than selling a surprise.

From the trading floor to the document layer

Jason did not start in document AI. Almost nobody did - the category barely existed a decade ago. He started with data of a very different kind. Early in his career he was a Senior Consulting Analyst at Fidelity Investments and a Sales and Trading Winter Analyst at Citi, the sort of roles where spreadsheets and market feeds are the native language. Numbers on a screen, decisions on a clock.

Then came the long stretch at NetBase Quid, the market and consumer intelligence company. He arrived as an Associate in New Business Sales and left as a Director - Enterprise Account Executive, Director of Strategic Growth, each step deeper into the business of selling analytics to large companies that wanted to make sense of enormous, unruly datasets. It is the through-line of his whole career: helping people turn a mess of information into something they can actually act on. First it was trading data, then consumer intelligence, now the humble document.

The Dartmouth College degree sits at the front of that story. It is a liberal-arts foundation applied to some of the most technical corners of enterprise software - a reminder that the ability to explain a thing clearly can matter as much as the ability to build it. And explaining clearly is, functionally, most of his job. A customer evaluating Sensible is rarely asking "can you extract this field." They are asking "will this hold up when I stop watching it." Answering that honestly requires knowing both the product's mechanics and the customer's fear.

How Sensible actually reads a page

To understand what Jason is selling, it helps to understand what he is not selling: a single large model pointed at a PDF with a hopeful prompt. Sensible's approach runs on SenseML, a declarative configuration language that combines deterministic, layout-based extraction with LLM-based parsing. The result is typed, schema-validated output rather than a paragraph of guessed-at text. In plain terms, the system is built to know when it is looking at a table, a signature block, or a total, and to return data your software can trust rather than prose it has to re-check.

That architecture is the technical answer to the accuracy-gap problem he writes about. Deterministic rules handle the parts of a document that behave predictably; the language model handles the messy, human parts that rules cannot anticipate. Prebuilt parsers cover common document types out of the box, while custom configurations handle the long-tail vendors and one-off formats that make document automation so stubborn. The platform is API-first by design, built to be dropped into a developer's existing workflow rather than lived in as yet another dashboard. When Jason talks to a customer, he is really translating this architecture into a promise about their Monday morning.

Look closely at what he actually publishes and a pattern emerges. He does not write hype. He writes field notes. A piece on AP automation for QuickBooks does not claim the native tool is bad - it says plainly, "Bill Capture works well for simple invoices reviewed inside the QuickBooks interface. The limits show up in automated workflows." Then it names the limit: no line-item extraction, no programmatic handling of multi-page documents. "That manual step," he writes, "is what keeps AP from being fully automated." Sensible, in his framing, is not a replacement. It is the extraction layer that finishes the job.

Reliable enough to trust with the boring stuff

The aspiration is not flashy, and that is rather the point. Document extraction is the plumbing behind proptech, insurance claims, financial services, healthcare intake, accounts payable - the unglamorous machinery of business. If it works, nobody notices. If it fails at 3 a.m. on a batch of odd invoices, everybody does. Jason's whole posture is aimed at closing the distance between the demo that dazzles and the pipeline that has to survive Monday morning.

It is a strange kind of ambition - to be the person who tells you what breaks before you find out the hard way. But in a market built on confident screenshots, honesty turns out to be a competitive edge. Jason Auh is betting the boring, trustworthy version wins.

There is a quiet logic to how his career keeps circling the same problem from new angles. Trading data at Citi and Fidelity taught him what happens when numbers have to be right and fast. Selling consumer and market intelligence at NetBase Quid taught him how large organizations actually buy analytics - slowly, skeptically, and only after someone has answered the awkward questions. Document AI at Sensible sits at the intersection: high-stakes data, cautious enterprise buyers, and a technology that is genuinely good but not magic. He has spent his whole career learning to sell the truth about complicated data, which is exactly the skill this category was missing.

The industries that lean on Sensible make the stakes concrete. Proptech firms pulling numbers off leases. Insurance teams automating claim intake. Financial-services companies parsing statements. Healthcare operations digesting intake forms. Accounts-payable departments trying to close the loop on invoices from vendors they have never met. In each case the document is not the product - it is the friction standing between a business and the thing it actually wants to do. Jason's role is to remove that friction without pretending the friction was never there.

If there is a signature to his approach, it is restraint. He does not promise perfection; he promises a smaller, more useful thing - a system that gets most of the way there automatically and is honest about the last stretch that still needs a human in the loop. In a field crowded with people selling the dream, Jason Auh has staked out the more durable position: selling the reality, and making the reality good enough to bet a workflow on.