Published May 4, 2026
The modern enterprise has more data than it has ever had, and less judgement to show for it.
Dashboards update in real time. Analytics teams surface patterns across every business function. AI systems summarise documents, draft plans, classify risk, and answer questions against vast internal corpora. Planning rituals multiply: QBRs, forecast reviews, pipeline meetings, launch gates, steering committees, strategy offsites. And yet, for all this instrumentation, most companies still struggle to answer a surprisingly small set of forward-looking questions well. Will this launch happen on time? Will the regulator move this quarter? Will the competitor ship first? Will the quarter close where the official forecast says it will? Will the supply disruption actually hit us, or are we overreacting?
The information exists. People across the company carry fragments of it in their heads — what a sales team senses about a competitor’s launch, what a country manager feels about a regulatory shift, what a supply-chain lead suspects about a vendor’s reliability. None of it shows up on a dashboard. Most of it never makes it into a planning meeting. The enterprise has built systems for everything except the layer where this judgement could be captured, weighed, and acted on.
That layer is what WePredict Enterprise is designed to become.
1
The enterprise has data. It still lacks judgement.
The paradox of the AI-enabled enterprise is that information abundance has not translated into better forward-looking decisions. Every official forecast is distorted by four familiar forces.
The first is sandbagging. Forecasters who carry accountability for hitting numbers learn to bias their estimates downward, building headroom into every commitment. The forecast is no longer a prediction — it is a negotiation, and the negotiation distorts the signal long before the number reaches a leadership review.
The second is HiPPO bias — the highest-paid person’s opinion. Once a senior executive has expressed a view in a room, the people in that room with better local information stop disagreeing. The official forecast calibrates to seniority, not accuracy.
The third is political adjustment. Forecasts get nudged to fit narratives the organisation has already committed to publicly. A miss becomes a story about external headwinds; a hit becomes a story about strategic execution. Either way, the underlying probability is buried.
The fourth is diversity collapse. The same five people argue the same five positions in every planning cycle, because they are the only ones in the room. The contrarian view from a junior analyst, the warning from a regional team, the pattern a customer-success manager has been seeing for three months — none of it makes it into the official process.
The four pathologies that distort the official forecast.
Companies do not merely suffer from imperfect information. They suffer from information that gets filtered, softened, delayed, and made legible only after it is safe. It is not data poverty. It is judgement poverty.
Deloitte’s recent observation is worth taking seriously: internal prediction markets, dormant as a category since the 1990s, may have something useful to contribute to surfacing strategic signals amid data noise — particularly for sensing, hedging, and what they call evergreen insights. The point is not nostalgic. It is that none of the four pathologies above are solved by adding more data, more dashboards, or more AI. They are solved by building a discipline for capturing distributed judgement and giving it consequence.
The problem is no longer lack of information. It is lack of organised judgement.
2
Why corporate prediction markets never quite arrived
The category has a longer history than most enterprise software buyers realise, and the history is more interesting than the standard dismissal suggests.
The mechanism worked more often than the institutions did. Hewlett-Packard ran internal sales forecasting markets between 1996 and 1999 with twenty to thirty handpicked participants. In six of the eight markets where an HP official forecast existed, the market was closer to the actual outcome — a 75% win rate against professional forecasters with access to internal data. Ford’s prediction markets in the late 2000s reduced mean squared error on weekly vehicle sales by approximately 25% versus the company’s expert forecast. Google’s Prophit, launched in 2005 with design input from Hal Varian, ran for over six years, attracted twenty per cent of Google’s employee base as traders, and produced well-calibrated forecasts across thousands of questions. Siemens forecast a software project’s delay before management’s planning systems were ready to admit it. The Intelligence Community Prediction Market, hosted on Cultivate Labs’ infrastructure on the IC’s classified network from 2010 to 2020, processed over 190,000 predictions from 4,300 cleared analysts and was directionally accurate on roughly 82% of question-days.
The institutions did not.
Most of these deployments stayed experimental rather than becoming infrastructure. They depended on a champion. They asked too many casual questions and not enough consequential ones. They lived on a separate URL that employees forgot to visit. When the champion left, the markets quietly faded. Google’s first market shut down. The ICPM was decommissioned in 2020. The mechanism produced research-grade calibration data; the institutions never built the operating habits to use it.
The vendor migration tells the same story in a different language. Inkling Markets, the leading enterprise platform for a decade, became Cultivate Labs in 2016 and has now explicitly discontinued prediction-market mechanisms in favour of scoring-rule-based forecasting. Consensus Point pivoted to Cipher, a different category entirely. Metaculus and Good Judgment moved up-market into curated superforecaster panels rather than internal trading. The category quietly evolved away from fetishising the mechanism and toward what buyers actually wanted: better judgement, better calibration, better decision support.
Five reasons explain the stall. Too many casual or low-value questions. Management ignoring the outputs the market produced. Markets living outside the workflow where decisions actually got made. Internal markets becoming politically dangerous because they embarrassed plans or exposed fragility. And, most decisive of all, the field sold a mechanism when buyers actually wanted decision advantage. “Prediction market” is how the machine works. “Collective intelligence” is why a buyer cares.
The single most instructive lesson comes from Google’s second internal market, Gleangen, which ran a forecasting contest in 2022 on whether Google would integrate large language models into Gmail by particular dates. Bo Cowgill, who designed the original Prophit, observed that the market would have been more useful asking about Microsoft and Outlook than about Google and Gmail. External questions carry almost none of the political risk of internal ones. They are easier to resolve. They are easier to act on. And they are not entangled with the career incentives of the people forecasting them. That asymmetry may be the most important thing the field learned in twenty years.
3
What’s changed: AI, Slack, and the rise of collective intelligence
Three things are genuinely different in 2026 than in 2015.
The first is that Slack and Teams have become the operating surface of knowledge enterprises. Collective intelligence can finally live where work actually happens, rather than on a separate URL that employees forget to visit. Many earlier systems died not because the core idea was wrong, but because participation sat outside normal work. Slack changes that geometry. Market cards in channels. One-click participation. Threaded discussion attached to each market. Weekly rituals embedded in the cadence the team already runs. The forecasting layer becomes part of the operating system rather than a side-quest.
The second is that AI participants change the participation economics. Questions can be drafted, rationales summarised, disagreement mapped, and base rates estimated at near-zero cost. The cold-start problem that previously made internal markets feel sparse and lifeless has a credible solution — synthetic forecasters seeded against base rates ensure every market has a starting price, and AI digests turn overnight movement into a few legible sentences for senior reviewers. AI does not replace judgement here. It lowers the operating cost of collecting, explaining, and routing it.
The third is that public-market legitimacy from Kalshi at a $22 billion valuation and Polymarket at $15 billion has made probabilistic thinking legible to executives in a way it simply was not when Bo Cowgill and Patri Friedman were defending Prophit inside Google. The world has, in a real sense, started thinking in bets. Even for enterprises that will never use public betting products, that familiarity matters. It makes the underlying grammar of probabilities, odds, and changing beliefs easier for executives to absorb.
The institutional validation goes further still. In April 2026, Jamie Dimon publicly said it was “possible one day” JPMorgan could offer a prediction-market service of its own — not for sports or politics, and with strict guardrails on insider information. Goldman Sachs is reported to be examining the space; Robinhood already runs a prediction-markets hub it describes as its fastest-growing business. When the most powerful CEO in global finance frames the goal as harnessing collective intelligence for risk assessment and decision-making, the category has crossed a legitimacy threshold that no academic paper or retail platform alone could deliver. The mechanism is no longer fringe. The remaining question is who builds the right configuration of it for whom.
Sooth Labs is worth pausing on, because it sits adjacent to what WePredict Enterprise is building rather than competing with it. Sooth’s models train on cross-industry datasets and structured market data. They produce a probability that the WHO declares another pandemic by 2028, or that Anthropic goes public this year. These are useful numbers. But they cannot access what the organisation knows. They cannot read the signal a country manager is picking up from her dealer network in Indonesia. They cannot weigh the suspicion a regulatory affairs team has been carrying for three weeks about an unannounced inspection. They cannot integrate the texture of how three different sales leaders feel about the same customer. AI-only forecasting and collective organisational judgement are not substitutes. They are complements — and a serious enterprise will eventually want both.
The category language has also matured. The serious enterprise vendors no longer talk about prediction markets. They talk about collective intelligence, decision hygiene, forecast culture, calibrated judgement. This is not cosmetic. It reflects a shift from selling a mechanism to solving a buyer’s decision problem.
The modern opportunity is not enterprise prediction markets as a novelty. It is enterprise collective intelligence as infrastructure.
4
Inside WePredict Enterprise: four building blocks
WePredict Enterprise is the third surface in the WePredict architecture. The first is Public — open markets on the web. The second is Private — closed markets running in WhatsApp groups, where reputation accumulates inside an existing social context. The third is Enterprise — closed tenancy, admin-curated questions, on Slack. One engine. Three configurations.
The product rests on four building blocks.
Questions are management-curated. Question quality is destiny. Open bottom-up question creation produces noise, political landmines, and low-relevance fun markets that dilute the signal until the system stops being taken seriously. The question library is owned by a named admin or a small curation team. Every market that goes live answers a question that someone has already decided is worth answering. This is discipline, not censorship. It is the line that separates a forecasting system from a betting site.
Three mechanisms are used selectively. Different questions need different aggregation modes. A poll with receipts is the lightest weight — anyone can submit a probability, every prediction is logged against the forecaster’s track record, the consensus is a calibrated average. This works for organisational sensing where the goal is to surface a distribution of views. Parimutuel pools work for multi-outcome questions where stakes are pooled and redistributed at resolution — useful for bounded forecasts on competitive moves or product timing. LMSR markets, with continuous pricing driven by an automated market maker, work for high-stakes binary events where live probability matters and traders need to update prices in response to new information. The point is not to choose one mechanism ideologically and impose it on every question. It is to apply the right mode to each question type.
WeCoins is the internal currency. The enterprise admin allocates WeCoins to employees on a regular cadence — the tenant funds the float. WeCoins are earned through accuracy, burned through staking on markets, and redeemable for non-cash benefits the organisation already values: training budgets, conference allocations, time-off flexibility, internal recognition. Critically, WeCoins are structurally inconvertible to external cash by design. This single architectural choice resolves securities law, gambling law, and HR-compensation exposure at a layer below policy. Pure play money decays at month three because nothing is at stake. Real money triggers a legal review the company will not survive. WeCoins thread the needle by giving participation real consequence inside the organisation’s existing benefits economy without crossing any of the lines that matter externally.
Forecast Score is the compounding reputation layer. Every prediction every employee makes is scored against the eventual outcome using Brier mechanics. The score accumulates. It compounds. Over months and years, the system produces a record of who the organisation’s most accurate forecasters actually are, on which kinds of questions, at which time horizons. Most enterprise software assets do not compound. Forecast Score does. For the individual, it is a record of judgement that formal review processes rarely capture. For the team, it identifies whose view to weight when stakes are high. For the organisation, it builds an internal calibration capability that no external vendor can sell. Forecast Score is per-tenant by design — it lives only inside the company that owns it. Leaderboards are the surface employees see; Forecast Score is the substrate that makes them mean something across cycles.
The product lives where the work is. Slack is the primary surface in 2026, with Teams as the obvious second deployment. Market cards appear in dedicated channels. Participation is one click. Discussion threads attach to each market. AI-generated digests summarise overnight movement and explain why the consensus shifted. Weekly rituals — a Monday-morning question post, a Friday-evening resolution — create the habit loop. Admin controls handle question curation, audience scoping, and resolution authority. Executive dashboards roll up calibration views and disagreement maps for leadership review.
The engine is the same as Public and Private WePredict. The configuration is built for the enterprise: closed tenancy, admin-distributed currency, management-curated questions, decision-quality outputs. Public WePredict lives on the open web. Private WePredict lives in WhatsApp groups and closed social contexts. Enterprise WePredict lives in Slack. One engine. Three configurations. Different surfaces. Different incentives. Different jobs.
Three surfaces on a shared engine — Public, Private, Enterprise.
5
Two wedges: external signals and internal truths
The architecture supports two distinct categories of question, and the deployment sequence matters.
External-event markets come first. These are externally resolved events where domain insiders inside the company have better judgement than generic public markets. The four families that matter most in practice:
Regulatory timelines. Will the FDA approve this drug class by Q3? Will the EU AI Act amendment pass before year-end? Will the data-protection authority enforce the new rule before April? Internal regulatory affairs teams, government-relations staff, and country leads carry rich tacit understanding of these processes that shows up nowhere on a dashboard.
Competitive moves. Will a named competitor launch in this category before our planned launch? Will the pricing change we have been hearing about materialise this quarter? Will the partnership announcement we have heard rumours of be signed? Sales teams, channel partners, and analyst-relations functions accumulate this signal continuously.
Macro and geopolitical. Will the central bank cut rates this quarter? Will the trade dispute escalate? Will the election outcome shift the policy environment in our largest market? These questions sit at the intersection of public information and private interpretation — public markets like Kalshi can produce a number, but a multinational’s country managers and policy advisors have texture the market does not.
Supply, weather, and operational. Will the hurricane season materially disrupt logistics? Will the strike at the supplier resolve before our production schedule slips? Will the chip shortage extend into the second half? Operations leaders and procurement teams carry the early signals on all of these, weeks before they show up in financial models.
The reason to lead with external-event markets is simple. They are externally verifiable. They have clean resolution. They do not turn the market into a referendum on any individual employee’s performance. They produce decision-relevant signal that complements rather than threatens the organisation’s existing forecasting and planning processes. They are the wedge that lets the system establish credibility before it tackles the harder ground.
Internal markets come second. Once the system has resolved enough markets for Forecast Scores to mean something, and once leadership has built trust in the discipline, internal markets become valuable. Will Sprint 14 ship by Friday at six? Will the product launch happen this month? Will the partnership close by quarter-end? Will churn for the new cohort cross threshold X by year-end?
Internal markets do two jobs, both valuable. They produce better forecasts than the official forecast, which is what the academic literature on Ford and HP and Google demonstrates. And they produce alignment — a structured way for teams to surface what they know but cannot easily say in a status meeting where the senior person has already expressed a view. Internal markets are not just about being right. They are about making it easier for an organisation to tell itself the truth.
But internal markets must observe a hard line. They are about projects, outcomes, and collective efforts. They are never about individual people. No performance reviews. No personal sales targets. No questions that turn the platform into a scoring system for human beings rather than projects, outcomes, and collective efforts. A market on whether a particular salesperson will hit their personal quota is the move that ends the system. A market on whether a particular project will ship on time is the move that grows it. Every documented failure of corporate prediction markets in the academic literature traces back, in some form, to crossing this line. Every successful deployment respected it. The discipline is not optional.
Two wedges — external signals first, internal truths later, and the line that must never be crossed.
The first deployment template that has the highest probability of working is narrow and concrete. Ten external questions where the organisation has clear domain insight. Five internal questions on collective outcomes only. One named admin who owns the question library. One named decision-maker who commits, in writing, to acting on the signal — to bringing market outputs into the relevant planning meeting and explaining when the decision diverges from the consensus. Three months of running the system and resolving questions before any judgement is made about whether to scale. The instinct to launch with breadth is wrong. The instinct to launch with discipline is right.
Coda
From noise to judgement
Every era of enterprise infrastructure has been defined by the layer it added. Mainframes added systems for data. Networks added systems for communication. Cloud added systems for computation. AI is adding systems for reasoning over text and structured information.
None of these layers, on their own, captures human judgement. The expert knowledge a regulatory affairs team carries about an upcoming decision. The pattern a customer-success manager has been watching for three months. The competitive signal a country manager picks up at a regional industry dinner. The texture three sales leaders carry about the same prospect. All of this is information the organisation already owns, distributed across the people inside it, and almost none of it makes it into the formal systems that drive decisions.
That is the layer WePredict Enterprise is designed to become. Not a betting platform. Not a gimmick. Not a side tool. A collective-intelligence infrastructure for the modern enterprise — one that aggregates dispersed belief, makes dissent safe, rewards calibration, and gives the organisation a compounding record of who actually sees the future best.
What compounds in such a system is not only the forecasts. The question library compounds. Forecast Scores compound. Base-rate libraries compound. AI summaries improve. Organisational trust in calibrated forecasters compounds. Most of all, the habit of thinking in probabilities about questions that matter compounds. That may be the most valuable asset of all.
The companies that win in the AI era will not just have better models. They will have better ways to turn distributed human judgement into decision advantage.
