Visual summary of operating lessons from Ian Wong.

Lessons from Ian Wong

As the co-founder and former CTO of Opendoor and the first data scientist at Square, Ian Wong specialized in turning manual, operations-heavy businesses into automated systems at scale. This collection gathers his perspectives on data science, technical leadership, and building software that interacts with real-world workflows.

Part 1: The Operations-First Phase

  1. On early-stage survival: First Round's notes summarize Wong's view that in a startup's early innings it can make sense to be operations heavy: in his account a low-fidelity end-to-end process, with operators standing in for an algorithm, comes first and is made more sophisticated bit by bit. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  2. On starting from the key risks: Wong says the first thing a company has to think through is the key risks that could keep it from realizing its vision, and only then whether data science is part of addressing them; at Square the risk pillar is where data science came in, and at Opendoor a robust pricing and data function was one of the capabilities needed to make competitive offers. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  3. On manual learning: At Opendoor, Wong says, pricing data scientists were required to start by valuing homes manually alongside the in-house experts, because that builds the domain intuition: "I don't care if you have a PhD from some fancy school, it doesn't matter." — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  4. On starting with internal tools: Wong says that when he arrived at Square there were only about 50 suspected fraud cases, hardly meaningful, so rather than building bespoke algorithms he built internal tools and then the data warehouse before working up to the machine learning stack, which he calls a blessing in disguise because it gave him deep grounding in the domain. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  5. On the cost of operations: Wong explains why founders under-invest in data science: on a marginal-cost basis it is cheaper and faster to hire operators than to build a pricing data capability, but leaning purely on operations yields only a certain scale, so the executive team needs conviction in the value of data science up front. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  6. On bridging bits and atoms: Wong describes Opendoor's valuation model as combining machine intelligence with human expertise: operators identify location-dependent factors such as road noise, the machines estimate how much they matter, and past human adjustments are used to compute average adjustment values. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020).
  7. On operational empathy: Wong says he shadowed a ton of operators at Square and Opendoor and walked houses with the home estimators, and that Opendoor bused employees to its Sacramento market to shadow every operational function. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  8. On operations as a differentiator: Wong says that as a technologist he has come to appreciate operations as a real strategic differentiator, and that some operational touchpoints are essential for customers: selling a home is the largest transaction most people make, so a human voice on the phone should be treated as something to make great rather than as a defect. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  9. On operations data as a moat: He says the data operators generate is proprietary first-party data that accumulates into a data moat: every home Opendoor inspects yields hundreds of data points, and what it learns about one home reveals something about others in the neighborhood, a loop that works only when operations, data science and engineering talk to each other. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)

Part 2: Transitioning to Algorithms & Scale

  1. On the tipping point: Wong says that if you lean purely on operations and humans it will yield only a certain scale, and you will not reach the precision and scalability the company needs. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  2. On the threshold for production: Wong says an algorithm "doesn't have to be strictly better than a human being" to be rolled into production: it just has to be no worse, at least for a subset of decisions. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  3. On iterative deployment: Wong describes an evolution at Square and Opendoor: first humans armed with good tools, then the tool integrates the algorithm's suggestion and disagreements become feedback, then the algorithm acts as a shadow agent reviewed like an operator; at Square most cases are now auto reviewed, leaving unusual ones for humans. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  4. On controlling the operator tools: Wong says to lean on human experts early but keep very tight control of the tools they use, including the UI and the data going in and out, so their decisions produce structured data that algorithms can plug into; at Square this was the account review tool and at Opendoor the valuation tool, whose decisions were first guided and then largely automated. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  5. On building trust in models: To get both teams on board, Wong says to paint a maturity curve for the operations and data science teams, set pretty aggressive milestones, and decide how accuracy will be measured and how to degrade gracefully if an algorithm is not up to par. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  6. On longevity: First Round's summary of the interview says Wong's view is that in the early innings a startup may be operations heavy, but as it scales data science becomes a critical component of running a business with longevity in mind. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  7. On feedback loops: Wong says that when the human operator disagrees with the model's output, that becomes interesting data fed back into the algorithm, which is then iterated on. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  8. On letting humans inspect the algorithm: Wong says one does not simply deploy a trained algorithm, because it might whiff on things that are obvious to a human, so a tool that lets people inspect its outputs before it goes into the wild is the obvious first step. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  9. On human bias: He adds that having humans review everything was not feasible at Square and, more perniciously, brings human biases, which is why at some point algorithms can handle things better than a human can. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)

Part 3: Structuring Data Science Teams

  1. On practical research: The Data Chief episode notes summarize that data scientists need business skills as well as technical ones, and that research is not helpful if it does not meet the immediate needs of the business. — The Data Chief Podcast, Ep. 28: Opendoor's Ian Wong (2021).
  2. On skin in the game: Wong says to find pockets where data science goes on the front lines so that data scientists "feel the heat" and have real skin in the game, which they like. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  3. On domain understanding: Wong says a data scientist generalizes human intuition into data and algorithms and tests it with a backtest, so real domain understanding matters: Opendoor's data scientists sit alongside the pricing operations team, visit houses and are on the hook for doing some valuations when onboarded, because you cannot have an ivory tower in data science. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  4. On data fidelity: The Data Chief episode notes summarize that great algorithms start with great data and that high-fidelity data is a key differentiator, especially when first-party and third-party data are mixed. — The Data Chief Podcast, Ep. 28: Opendoor's Ian Wong (2021).
  5. On model ensembling: Wong says ensembling lets Opendoor use different pricing models together to compute a weighted average of home values, and that it generally results in more accurate pricing than any single model. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020).
  6. On communication: Wong advises analysts to develop business acumen and communication skills, being able to meet hard business challenges, break them into quantifiable sub-problems and present findings so the audience can engage with them. — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017).
  7. On organizational structure: Wong says ideally everything aligns to a business unit led by someone who knows the disciplines, but realistically Opendoor ran a matrix: business-unit leaders working with functional leaders for data science and engineering who mentor on the craft. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  8. On evaluating models: Wong says to work backwards from the business KPI: at Square fraud loss, at Opendoor turning a profit at the unit level, which then translates into an engineering objective such as minimizing the error of a house valuation. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  9. On fraud and risk: Wong says reviews at Square held the team accountable both for how much fraud was happening and for the false positives, meaning merchants unfairly frozen, which he calls a terrible customer experience. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  10. On simplicity: Wong says advanced techniques often yield only incremental lift over a simple solution while adding complexity that becomes a heavy tax, and suggests implementing the algorithm that papers make fun of in their introductions and forgetting the rest of the paper. — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017).
  11. On predictive and descriptive data science: Wong splits data science into two buckets: a predictive function whose output is a prediction that interacts with the real world, where errors are bad, such as fraud detection or valuing houses, and a descriptive function, the analytics work of understanding where the business is going and running experiments. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  12. On what the company wants from data science: He warns of a trap in which an ambitious data scientist wants the predictive work while the company wants descriptive work, producing an impedance mismatch and burnout, so an organization should first be clear about what it needs out of the data science function. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  13. On distilling human intuition: In classical machine learning, Wong says, much of the work is distilling human intuition, such as the curb appeal of a house, into features and seeing which hold up to generalization, using computer vision and crowdsourcing to get structured data out of high-dimensional data like photos. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  14. On crowdsourcing: In a 2017 talk Wong said Opendoor turned photos into structured data with a combination of deep learning and a lot of crowdsourcing, which he called a totally unappreciated technique. — Modernizing Real Estate with Data Science (Ian Wong at Data Driven NYC, 2017 talk)
  15. On debugging models: Wong says what delivers results is an error-focused, iterative process of continuous model improvement, starting with the model's biggest outliers and reasoning from first principles to eliminate them, and adds that "model debugger" describes the role better than "data scientist." — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017)
  16. On taking a position: Wong says that with 80 percent of the information, if lucky, teammates are counting on a recommendation; earlier in his career he hesitated to form a strong perspective, but he now calls that an abdication of responsibility, since the job is to think through the implications and come up with a point of view. — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017)
  17. On the value of an analysis: He says the value of an analysis is measured by whether it influences decisions, and that even a brilliant analysis is ineffective if it is not delivered to the audience in an accessible manner. — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017)
  18. On data versus algorithms: Wong says more data beats better algorithms most of the time, and that because tools and infrastructure are increasingly commoditized the industry is ready to move up the application stack rather than focus on the base layer. — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017)
  19. On the shape of the organization: At its peak, Wong says, Opendoor's engineering and data science teams numbered roughly 150 to 200 people, organized around four capabilities he tracks as CTO: consumer experiences, operations (enabling the operators on the ground), pricing and infrastructure, with data scientists spread across them in different mixes. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)

Part 4: Hiring, Talent & Culture

  1. On the T-shaped professional: Wong writes that industry needs "more of a T-shaped individual (deep expertise in one area, complemented by breadth in adjacent ones so they can collaborate effectively with teammates), vs. an I-shaped individual." — A Few Reflections on Data Science (Ian Wong, Medium, 2017).
  2. On hiring early data scientists: Wong says early data scientists need to be self-starters able to work without large-company infrastructure and be a sponge deeply curious about the business, falling in love with the customer and the business problem rather than the method. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  3. On generalists first, specialists later: Wong says the early data science team leaned on generalists, but he did not fully appreciate until later the importance of specialists; the signal to bring them in is when problem definition starts to stabilize and he has a clearer map of which disciplines transfer to the problems at hand. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  4. On the ability to unblock yourself: He says that early on being a self-starter is much more important than existing domain expertise, since there are other ways to come up to speed quickly, and that Opendoor's hiring mistakes came from calling in specialists or experienced hires too early when they lacked that ability to unblock themselves. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  5. On mirroring the job in interviews: To assess early data scientists Wong mirrored the interview to the job: candidates built a model in two or three hours on real, anonymized housing data while he sat side by side, ready to unblock them, to see whether they could go end to end. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  6. On cross-functional empathy: Wong says one of his best friends in the company has to be the COO or chief customer officer, and a VP of engineering's best friend has to be her counterpart in operations, so engineering and operations are aligned at the organizational level. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  7. On naming values: Wong says it is important to name the values, because implicit ones lead to clashes when people disagree, and that values are tested in hard moments, such as a slow winter when engineers asked one by one what was happening with the business; he turns values like build openness into rituals such as one-on-ones, all-hands and shared board decks. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  8. On starting and ending with the customer: Opendoor's first core value was to start and end with the customer, and Wong says the early team was flown to Phoenix to run customer studies, talking with prospective shoppers and sellers, because many employees had never bought or sold a home. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  9. On being boxed in by titles: Wong says titles should enable, not constrain, that everyone is a problem solver first, and that a title can become an escape hatch for avoiding things that matter, such as a data scientist treating productionizing an insight as an unnecessary implementation detail. — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017)
  10. On being scrappy and pragmatic: He writes that scrappy plus pragmatic plus business acumen beats technical expertise, and that the best data scientists are relentlessly resourceful, shifting from "I need to gain skill x" to "I am going to solve problem y" and from "not my job" to "run towards where the impact is." — Interview with a Data Scientist: Ian Wong of OpenDoor (Peadar Coyle, 2017)
  11. On the data science Venn diagram: In a 2017 Q&A Wong said people take the hacker, statistician and domain Venn diagram too far: it underplays software engineering and business acumen, and people box themselves in as data scientists who only ship models when the job is to solve really hard problems. — Modernizing Real Estate with Data Science (Ian Wong at Data Driven NYC, 2017 talk)
  12. On the hiring hierarchy: Wong described a hierarchy of hiring criteria: can you do the job, do you align with the company's values, being open and customer-centric, and are you on board with the vision. — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020)
  13. On teaching the vision: He says that in a start-up nothing is more important than the vision, and that the founder's job is to constantly preach and teach it and make sure every member of the team is aligned, which is why screening for vision in hiring is critical. — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020)

Part 5: Engineering Leadership & Execution

  1. On internal tools: Wong says internal tools can easily fail when the process they encode is not legible and aligned among stakeholders, and that they tend to succeed in small teams because alignment on the process is easier. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  2. On a mismatched spec: Wong recalls an engineer on an Atlanta tools team who nearly quit; on flying out he found the engineer was working from a PM spec that differed from the engineering manager's, the operator's and the process central operations prescribed, a triple matrix of product, engineering and ops that makes internal tools hard. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  3. On building proprietary and buying commodity: His tagline is to build proprietary and buy commodity, though he cautions that stakeholders often want to buy software for the process it prescribes rather than for what it does, which creates trouble when the process is not explicit or agreed. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  4. On managing managers: Wong says the common Valley advice to delegate as much as you can can turn into abdication; he wishes he had stayed more hands-on and created a culture where managers and skip levels are acceptable in the weeds. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  5. On being inspected: Wong says people feel micromanaged when expectations are set so that nobody inspects their work; every business unit leader, including the CTO, gets inspected through business reviews, and the tone should be that your work is going to get inspected and it will make it better. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  6. On layers of strategy: He says everything comes from the business strategy, and the engineering leader maps both the software system and the people system to meet it, creating alignment up and down the decision stack because you cannot operate at a 10,000-foot level. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  7. On alignment: Wong says a CTO needs to make sure every roadmap is not just a business roadmap but a business and technology roadmap, collaborating with the exec team so technology keeps playing an outsize role. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  8. On mechanisms: At a 2018 CTO Summit Wong said: "Good intentions don't scale — mechanisms do," describing the meta work of the leader as the work that enables great work to happen as Opendoor grew from 4 to 650-plus employees in four years. — 6 Mechanisms That Enable Engineering Teams to Scale (2018 CTO Summit write-up)
  9. On aspirational and committed key results: Wong split OKR key results into aspirational ones, where reaching 70 to 80 percent can be a success, and committed ones where completion is non-negotiable, kept to about 3 to 5 objectives per team, after early OKRs left half the goals red and the rest no longer applicable. — 6 Mechanisms That Enable Engineering Teams to Scale (2018 CTO Summit write-up)
  10. On OKR check-ins: After several iterations he settled on 30-minute bi-weekly check-ins against OKRs with a lightweight template covering plans, problems and progress, and on also listing objectives the team is deliberately not prioritizing. — 6 Mechanisms That Enable Engineering Teams to Scale (2018 CTO Summit write-up)
  11. On blameless post-mortems: Opendoor's post-mortem template opens with the statement that regardless of what is discovered, everyone did the best job they could given what they knew at the time, then covers what happened, impact, timeline, five whys, prevention ideas and action items ticketed in Jira. — 6 Mechanisms That Enable Engineering Teams to Scale (2018 CTO Summit write-up)
  12. On monitoring the monitoring: After two unexpected Sev0 incidents landed the week a new VP of Engineering started, Wong said they learned they had no monitors on the quality of their incident response, and began tracking bugs by type, severity, breached SLAs and on-call to make a dashboard for operational excellence meetings. — 6 Mechanisms That Enable Engineering Teams to Scale (2018 CTO Summit write-up)
  13. On defining quality: He says the way to keep quality as a company scales is to define it proactively, for instance in a design doc template that makes engineers articulate impacts on other stakeholders, and to form habits early such as lightweight design critiques. — 6 Mechanisms That Enable Engineering Teams to Scale (2018 CTO Summit write-up)
  14. On pre-mortems: Wong says Opendoor's leadership team runs pre-mortems: suppose an initiative is launched and does not work, list all the reasons why, and work through that list to prevent it. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  15. On the fears you tell yourself: He describes a founders' exercise of listing what everyone thinks is true but may be a fear or lie they tell themselves, such as whether selling a home nearly instantly is something people really want, to bust the ego and stay real about the problems to solve. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  16. On learning to learn: Wong says the steepest learning curve over eight years as CTO was learning how to learn more quickly: once you solve one set of problems you earn the right to solve the next, and he built a mentorship network of coaches, advisors and peers and interviewed people about how they do jobs he was getting up to speed on. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  17. On writing your job spec: He suggests writing down, as if hiring your replacement, what the company needs from your role over the next 18 months, because writing forces clarity; his own spec went from building the algorithm and hiring the first team to thinking 12 to 18 months ahead on team structure and platform. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  18. On shock-absorbing weaknesses: Wong says he works both on strengths and weaknesses but accepts that some weaknesses are hard to fix, so he constructs a team around them, for instance recruiting a chief people officer to cover processes like promotion calibration that he finds his mind drifting from, rather than trying to turn every weakness into a strength. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)

Part 6: Problem Solving & Technical Philosophy

  1. On distillation: "I cannot do without a blank sheet of paper and a pen. When I come across something complicated, I will try to distill the core idea until it fits on a single, hand-written page." — HousingWire: 2019 HW Vanguard, Ian Wong
  2. On framing: Wong quotes Alan Kay, "A change in perspective is worth 80 IQ points," and adds: "Framing a problem well is probably the most important part of the solution." — A Few Reflections on Data Science (Ian Wong, Medium, 2017).
  3. On iteration speed: The Data Chief episode notes summarize that when launching machine learning prototypes quickly and iterating, it should not take a quarter to make and deploy a new algorithm. — The Data Chief Podcast, Ep. 28: Opendoor's Ian Wong (2021).
  4. On ignoring noise: In the RBC write-up of his TechTO talk, Wong says Opendoor stays focused on its north star of a seamless customer experience: "If you're distracted, you won't make progress. Keep the main thing the main thing." — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020).
  5. On cross-disciplinary learning: Wong says the team came to see analogies between asset management at hedge funds and what Opendoor was building, so quantitative finance backgrounds turned out to be a great source of talent for the company. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).

Part 7: Product-Market Fit & Strategy

  1. On scaling operations: Wong says deviations from a centralized practice should be controlled and monitored, because otherwise it is hard to scale or to build tools and software that encode the process, and that making the process blueprint explicit between operations and engineering is an overlooked secret sauce. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  2. On centralizing and decentralizing: Wong says centralization brings scalability and standardization that lets you build software and algorithms, while decentralization is valuable for learning quickly in an uncertain environment, so the art is deciding when the learnings have been had and how fast to centralize. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  3. On operations and data science cadences: He notes that operations is often measured on outputs over days while data science and engineering are measured over a quarter or two, and that operational playbooks are easier to change than tools, so engineers end up building to a spec several versions behind unless the process is made explicit. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022)
  4. On adverse selection: On pricing, Wong says Opendoor has to keep its spread very low because of the lemons problem: if offers have random errors, sellers take the ones that undervalue their home, so error is not normally distributed in outcomes. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023)
  5. On unit economics: Wong says Opendoor needs to turn a profit at the unit level, which means it cannot buy high and sell low; but it also cannot undervalue much, because sellers are savvy about their largest transaction. — It Shipped That Way, Ep. 7: Effective and Scalable Machine Learning (2023).
  6. On capital efficiency: Wong writes that running a company is kind of like solving an NP-hard resource optimization problem, so they have to be rigorous about the ROI of each initiative they spend energy on. — A Few Reflections on Data Science (Ian Wong, Medium, 2017).
  7. On soliciting extremes: On messaging, Wong said: "It's not optimal if the response is lukewarm. We wanted either an extremely engaging or extremely disapproving response. We are soliciting extremes." — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020)
  8. On assuming the hypothesis: According to the write-up, Opendoor's product-market fit hypothesis was that the home selling process is broken and that some people will always want to sell right away; until it was proven the team took a leap into the unknown and acted as if it were right. — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020)
  9. On being scientific in a lull: When the company hit a lull over the winter, Wong said they had to dig deep to make sure they were building the right experience: "From product operations to marketing to execution, we had to be really scientific about breaking problems down." — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020)
  10. On playing a repeated game: Asked why a seller should trust Opendoor's price, Wong said sales are recorded publicly so anyone can check accuracy over time, and that Opendoor is playing a repeated game, building a reputation as the provider of fair market value rather than optimizing the margin on any one deal. — Modernizing Real Estate with Data Science (Ian Wong at Data Driven NYC, 2017 talk)

Part 8: The Evolution of PropTech & AI

  1. On physical integration: Wong says what makes Opendoor special to him is using technology, data science and operational excellence on real-world pain points, and that this "marriage of the online and offline worlds has never been done." — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020).
  2. On real estate data: Wong says real estate data is messy and its sources fragmented across recorder offices, assessor offices, MLSs and geospatial datasets, so Opendoor's data engineers focus on ingestion and entity resolution to build a clean dataset. — First Round Review: Operations vs. Algorithms, with Ian Wong (2022).
  3. On cleaning labels: Wong says poor-quality data hurts Opendoor's valuation models, so the company used a Markov Random Field to standardize text and improve labels such as subdivisions, scoring labels jointly on how well they match the original text and how spatially continuous they are among neighbors. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020)
  4. On homes that did not sell: He says Opendoor also uses homes taken off the market without transacting, which it calls delistings: it cannot observe days-on-market for them, but including them gives a more comprehensive picture of supply and demand. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020)
  5. On enriching third-party data: Wong says that while Opendoor combines large proprietary and third-party datasets, the most important step is to enrich readily available third-party data with its own proprietary data, so location-dependent signals improve dramatically over time. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020)
  6. On the evolution of valuations: Wong says Opendoor needs to understand liquidity per home and uses survival analysis to predict how long a house will take to sell, which helps set a risk threshold for each home. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020).
  7. On space and time: In a 2017 talk Wong said the same home can have a very different value depending on whether you ask in 2008, 2012 or today, so pricing has to account for heterogeneity in space and in time, and for liquidity, meaning how ready the market is to absorb a home Opendoor buys and resells. — Modernizing Real Estate with Data Science (Ian Wong at Data Driven NYC, 2017 talk)
  8. On pricing risk: He explained that an Opendoor offer has two numbers, the fair market value and a fee that flexes with the risk of the home, so the price has to capture both the value and a good grasp of the risk profile, since suitable hedging instruments are not liquid enough to rely on. — Modernizing Real Estate with Data Science (Ian Wong at Data Driven NYC, 2017 talk)
  9. On standing behind the model: Wong said what distinguishes Opendoor from companies publishing automated valuations is that it puts its own money to work: it stands behind the model by buying the home and reselling it, so what matters is whether it is accurate at volume, not whether valuation models become commoditized. — Modernizing Real Estate with Data Science (Ian Wong at Data Driven NYC, 2017 talk)
  10. On liquidity: Wong asks what if homeowners could tap into the liquidity of their homes the way you withdraw from a checking account, and what if buyers and sellers could skip months of stress and uncertainty. — Unite.AI: Ian Wong, Co-founder & CTO of Opendoor, interview (2020).
  11. On consumer expectations: HousingWire's profile says Wong's goal is to make buying and selling a home as easy as hailing a ride, buying something online or booking a flight, by rebuilding every part of the process to be automated, frictionless and low-cost. — HousingWire: 2019 HW Vanguard, Ian Wong.
  12. On the ultimate goal: The RBC write-up says Wong learned at Square the value of using technology to eliminate friction and wondered about applying it to the single largest transaction of people's lives. — Tips for Launching, Running and Scaling a Start-Up (RBC Royal Bank, 2020).