The clue was hiding in a car-rental curve. Demand would jump, fall or behave strangely, and the team at Online Republic could see the movement without seeing its cause. Sometimes the reason was a summit. Sometimes it was weather. Sometimes several ordinary events had landed on the same weekend and quietly rearranged a city's appetite for cars, rooms, rides and meals. Campbell Brown, then the travel company's chief marketing officer, kept returning to a simple frustration: by the time the business understood the surprise, the useful moment had passed.
Brown had seen a related version of the problem at GrabOne, the New Zealand daily-deals company where he was part of a small early team. A promotion could generate a rush of demand almost immediately. The operating lesson was visceral. Demand was not a smooth line politely extending itself from last Tuesday. It was a living response to whatever people were doing nearby.
At Online Republic, Brown and his colleagues tested whether an explanation could change behavior. A booking widget told travelers why inventory was tight or prices were high in a place: a concert might be colliding with a basketball game. The company reported that the added context improved conversion by 35 percent. The result suggested something larger than a travel feature. If customers made better decisions when they knew the reason behind a number, perhaps businesses would too.
“I had a thesis if we could understand real-world events, we'd be able to better predict demand.”Campbell Brown
Signal 01 / The founder's raw material
A geographer follows the crowd
The idea sounds inevitable only in reverse. Brown had studied geography at Victoria University of Wellington, then spent seven years in London working across geospatial technology and advertising technology. He returned to New Zealand in 2008 and entered startups, bringing together an unusual bundle of skills: maps, data, customer acquisition, product journeys and conversion. PredictHQ would eventually need all of them.
His time at GrabOne supplied speed. The daily-deals company grew from a handful of people into a large operation in a short period, teaching Brown what obvious product-market fit feels like and how much work it demands. Online Republic supplied the problem. Global car and camper-van rentals exposed the team to thousands of local demand patterns that history alone could not explain.
Brown founded PredictHQ in Auckland in 2014 with Robert Kern and Mike Ballantyne. Their task was more difficult than assembling a global calendar. A Taylor Swift concert, a school holiday and a flood are all “events,” but their economic footprints differ. A stadium show can lift ride demand after the encore and hotel demand months before it. Rain can help one retailer while hurting another. A small trade show may matter more to a nearby restaurant than a huge match across town. Useful context requires location, timing, category, attendance, relevance and an estimate of impact.
The first PredictHQ product looked like familiar software: users could sign in, create projects and explore events in an interface. Sign-ups arrived. So did a more important message. Customers wanted the intelligence inside their own notification systems, pricing tools and forecasting models. A separate destination added work. An API could disappear into the work already being done.
Signal 02 / Follow the pull
The product became less visible and more valuable
Uber gave the shift a concrete shape. The ride-hailing company was an early inbound customer and had tried to solve event intelligence internally. To show what PredictHQ could do, Brown's team mocked its information inside the Uber app on their phones, flew to San Francisco and demonstrated how a driver might see an important event days before the crowd emerged. The visualization made a headless product legible. Uber could place the intelligence in driver notifications, advertising and forecasting instead of asking employees to consult another screen.
Brown treated Uber's adoption as a geographic instruction too. Roughly 18 months after starting the company in Auckland, he moved his family to San Francisco. The transition stripped away the advantage of a familiar network. His history at GrabOne and Online Republic carried weight in New Zealand, but little recognition in Silicon Valley. He spent about a year meeting founders, learning the venture landscape and improving a pitch he later described as initially “absolutely diabolical.”
One early investment connection had come through a scene too ordinary for a startup montage. While taking out the recycling in Auckland, Brown ran into his neighbor Paul Naphtali, a partner at Rampersand. Brown mentioned the idea; Naphtali connected him with colleague Jim Cassidy; Rampersand went on to lead the seed round. The anecdote fits Brown's style: direct, opportunistic and able to find comedy in the slog.
“Sometimes failure makes the business and idea much stronger than you ever thought it could be.”Campbell Brown
By 2018, PredictHQ had raised a $10 million Series A led by Aspect Ventures. A $22 million Series B led by Sutter Hill Ventures followed in 2020. The company also collected three New Zealand Hi-Tech Awards in 2019, across software, services and emerging company. The more meaningful progress happened inside customer workflows. PredictHQ expanded beyond travel and mobility into retail, restaurants, accommodation and logistics, where the same event can alter staffing, inventory, pricing or supply in different ways.
Brown studies the relationship between people, place and systems before entering geospatial technology.
An early marketing role provides a close view of sudden, promotion-driven demand.
The company forms in Auckland, wins Uber early and follows customers to San Francisco.
Brown announces sustainable profitability and reframes the platform as context for enterprise AI.
Signal 03 / Context compounds
A decade-old question meets the AI boom
Brown announced in March 2025 that PredictHQ had reached sustainable profitability. By then, the vocabulary surrounding the company had evolved. “Event intelligence” became “demand intelligence,” and the current pitch emphasizes “real-world context.” The underlying question remained stable. What outside force caused this number to move, and what should a system do about it?
That continuity matters in the age of large language models. A model can produce a fluent answer about a restaurant in Austin while missing the festival happening three blocks away next weekend. Brown's argument is that intelligence without current, location-specific context can sound convincing while remaining operationally weak. In a 2026 demonstration, he showed PredictHQ data inside a conversational workflow, with plans to bring the company's event features and forecasts closer to the interface where a user asks the question.
The ambition is larger than attaching a calendar to a chatbot. PredictHQ says it processes thousands of sources, cleans changing event records and models their consequences. Events move, cancel and compound. Their relevance changes with distance and business type. The defensibility lives in the accumulated machinery that decides which signal matters to which decision, and in years of demand data used to test that judgment.
Brown's public persona remains more operator than oracle. He is frank about impatience, poor early pitches and the danger of mistaking retrospective clarity for foresight. “People should never confuse hindsight with vision,” he has said. His preferred method is closer to learning in public: make a claim, put it near a customer, notice where it breaks and return with a stronger version.
There is a personal cost hidden inside that repetition. In a 2026 conversation, Brown described founders as people who “eat stress,” then made a distinction between absorbing it and reconciling it. The remark lands because his story contains plenty of the former: leaving a known market, rebuilding a network, enduring rejection and carrying a data infrastructure company through cycles of venture enthusiasm. PredictHQ's eventual profitability offers a less cinematic reward than a launch-day spike. It is evidence of a system that can keep operating.
The useful thing to steal from Brown is an approach to surprise. Do not merely record that the line moved. Search for the outside cause. Explain it in the language of the person making the decision. Then deliver the explanation inside the tool they already use. PredictHQ's history is a series of those translations: from a car-rental anomaly to an event, from an event to a feature, from a dashboard to an API, and now from an API to an AI conversation.
The world will continue to interrupt the model. A match will be rescheduled. A small convention will fill the hotel next door. A storm will change what people buy. Brown has spent a decade making those interruptions computable. The business he built is an argument that the future becomes more useful when it arrives with context attached.