Andrew Joiner’s origin story in software contains the sort of cheerful disorder that career advisers are paid to prevent. He studied pre-med. He worked on Wall Street. Then, by his own telling, he started a software company one afternoon. The line earns a laugh, which is part of its purpose. The more consequential detail is what he had noticed: email was changing the balance of power inside finance. Messages once treated as disposable chatter were surfacing on the front page of The Wall Street Journal. Firms accustomed to tidy ledgers had acquired a vast new archive of sentences, attachments and liability.
Joiner and his brother founded Singlecast Technologies to classify that unruly material. This was the early 2000s, before “unstructured data” became a phrase that could fill a conference hall. The business problem was already plain. Human beings produce information for other human beings. They write messages, fill out forms, move totals around invoices and leave meaning scattered across context. Computers prefer obedient rows and columns. The distance between those habits has supplied Joiner with a career.
“I was a pre-med student who worked on Wall Street then started a software company one afternoon.”Andrew Joiner, describing his route into technology
In 2006, information-management company ZANTAZ acquired Singlecast’s classification technology and retained its employees, including its founder. Joiner stayed aboard as the corporate nameplates changed. ZANTAZ led to Autonomy; Autonomy led to Hewlett-Packard. He ran a high-growth Autonomy unit focused on customer experience and marketing technology, then became worldwide head of HP Software’s roughly $250 million customer-experience business. A founder who entered through the side door had learned to operate at enterprise scale.
One problem, five corporate chapters
Chapter twoUtah, with no contacts and a growth brief
When Joiner became chief executive of InMoment in 2017, he moved his family to Utah without knowing a single person in the state. He later wrote that they loved it, praising both the welcome and the emerging technology scene. The move also put him in charge of a company that converted another unruly human artifact, customer feedback, into something businesses could act upon.
At the time of his appointment, InMoment served more than 350 brands across 95 countries. Its founder, John Sperry, shifted toward product innovation and the chairmanship while Joiner took the operating wheel. Joiner described the attraction in terms that could have applied to Singlecast: unusual technology related to his past, a people-centered culture and a location where a business could scale economically.
The six-year chapter became an education in combination. Madison Dearborn Partners acquired and recapitalized InMoment. In 2020, InMoment and MaritzCX agreed to join, with Joiner leading the combined company. More acquisitions followed. By his 2022 farewell note, Joiner said the business had moved from roughly $40 million in revenue and a little more than 200 employees to nearly $200 million and more than 1,400 people worldwide.
Numbers this large invite a heroic portrait of the chief executive. Joiner’s own note was more crowded. He thanked a 14-person management team, employees, investors and partners. He praised his successor, John Lewis, as a close partner and confidant. The instinct matters. Scale in software appears on a graph, but the work is conducted by people enduring reorganizations, acquisitions, new systems and the occasional video call that should have been an email.
Chapter threeThe paperwork beneath the AI show
In April 2023, Joiner became CEO of Hyperscience. The timing was almost indecently neat. ChatGPT had made artificial intelligence legible to a mass audience, and companies everywhere were rushing to attach “AI” to products, pitches and perhaps the office kettle. Hyperscience had been founded in 2014 by machine-learning engineers working on a quieter question: could software read documents with something closer to human flexibility?
Joiner had spent his career arriving at variations of that question. Email at Singlecast. Mixed media and customer communications at Autonomy and HP. Voice, video, surveys and reviews at InMoment. At Hyperscience the specimens are invoices, bank statements, claims, applications and handwritten forms. The content is different; the irritation is the same. It is information a person can understand at a glance and a conventional database cannot.
Joiner draws a useful distinction between systems that generate human-friendly information and systems that read it. The first category writes the memo. The second finds the account number on a crooked scan, recognizes a handwritten field, notices that the invoice total moved to an unexpected corner and produces data another system can trust. The glamour gap is immense. So is the commercial opportunity.
“We use models to essentially read human-friendly information, and that’s really the opposite of generative AI.”Andrew Joiner on Hyperscience’s role
His pitch is deliberately grounded in operations. Enterprises need a measurable return. They need accuracy that holds up in claims, benefits, payments and mortgages. They need to know where information came from, how a model handled it and what a transaction cost. In 2024 he argued that this provable return had been missing from much of the generative-AI conversation. By 2026, his vocabulary had shifted toward the “inference inflection point,” the moment when AI moves from experiments into production systems expected to reason and act.
The operational argument
Chapter fourThe bartender in the loop
Joiner can explain this architecture with an evening out. At Google Cloud Next in Las Vegas, he said his group checked the walk to dinner. Google Maps estimated 22 minutes. Apple Maps offered eight. They asked a bartender to settle it. His joke was that they had brought a bartender into the loop. The episode is small enough to remember and exact enough to reveal the problem: two inferences can be individually plausible and collectively useless.
His preferred future has humans “on the loop.” In a mature system, people should not have to rescue every document that confuses a machine. They supervise performance, investigate the meaningful exceptions and adjust the process. Different models take different jobs. Specialized models handle high-volume tasks efficiently; expensive frontier models are reserved for the work that needs broader reasoning. Affordability joins accuracy and automation as an engineering constraint.
This is where Joiner’s long apprenticeship becomes relevant. Email supervision taught that new information creates new obligations. The acquisition chain taught him how technical products enter larger corporate systems. InMoment taught scale, combination and the value of contextual signals. Hyperscience places all of it inside a market currently tempted to use one expensive hammer for every job.
The company’s 2026 product release emphasized routing work across processors and model types, with controls for quality, observability and sensitive information. That spring, Hyperscience was named a Leader and Customer Favorite in a Forrester evaluation of document-mining platforms. In September it was named a Leader in Gartner’s intelligent-document-processing Magic Quadrant for a second consecutive year. Such recognitions belong to a company and its team. They also fit the argument its CEO has been making: enterprise AI will be judged by whether it works repeatedly, affordably and under scrutiny.
The continuing lineA career arranged by the data
Joiner’s résumé looks varied only from a distance. Up close, it is almost stubborn. He keeps returning to information that carries meaning for people and inconvenience for machines. The formats have evolved from email to rich feedback to scanned documents. The consequences have expanded from regulatory discovery to the systems that process public benefits, insurance claims and financial transactions. His job titles grew grander while the essential task remained wonderfully humble: read this correctly.
There is something reassuring in that persistence. Technology fashions change at a pace that makes every quarter feel like a geological era. Joiner’s career suggests that durable work often hides beneath those labels. Big data, customer intelligence, hyperautomation and agentic AI all eventually meet the same old world, with its smudges, eccentric layouts, incomplete fields and people who place a decimal point wherever they please.
The aspiration now is to turn that old world into dependable fuel for the new one. Joiner wants enterprise AI to operate in consequential settings with security, governance and a sensible bill. He wants models chosen for the task instead of prestige. He wants human judgment elevated from endless correction to active supervision. It is a product entrepreneur’s view of the future: ambitious in scope, suspicious of waste and permanently interested in what the software does after the demonstration ends.
A quarter century after email sent him looking for a classification system, the volume of human-friendly information has only grown. So has the appetite of machines. Between them stands the work Joiner chose one afternoon, still unfinished and finally fashionable.