Bernstein tracks new game sales with a three-stage AI-assisted model, with early signals for Resident Evil: Code Veronica standing out

Overview

The report introduces Bernstein's three-stage AI-assisted sales forecasting system for Japanese video game research and applies it to several upcoming or recently released games. Its core conclusion is that Steam follower and review data can provide more timely signals of demand for new games beyond traditional installed-base and franchise-history data, with the early performance of Resident Evil: Code Veronica appearing particularly positive.

Core views

The report seeks to address a recurring challenge in research on Japanese game companies: how to estimate sales more objectively when information is limited and a new title has not yet launched. Bernstein designs the forecasting process as a three-stage funnel from announcement through post-launch and emphasizes that the framework pursues not only forecast accuracy, but also processing speed, reproducibility, and data auditability. The first stage establishes a placeholder sales forecast when a game is announced. The model combines three inputs: the console penetration of prior titles in the same IP or comparable games from the same publisher, the addressable console installed base at launch, and the share of PC sales estimated by publisher and IP. The report expects the PS5's approach toward the end of its cycle to slow growth in the total console installed base by FY3/27E, although Switch 2 will carry certain IPs for the first time, potentially generating incremental demand relative to predecessors released several years earlier. The number of commercially meaningful new games on Steam, broadly defined as titles with more than 1,000 reviews, increased from approximately 600 in 2017 to approximately 1,000 in 2025, indicating the PC platform's rising importance. Capcom's PC sales share reached 54% in FY3/26; Ubisoft's PC revenue share increased from approximately 25% in FY3/21 to more than 40% in FY3/26, while EA and Take-Two were generally in the 20% to 30% range. A uniform assumption therefore cannot be applied to the PC mix across publishers. Console penetration examples also show that franchise recognition alone is insufficient. The report estimates console penetration of approximately 4% for Monster Hunter Wilds, below the approximately 6% to 7% recorded by Monster Hunter World and Monster Hunter Rise. Resident Evil: Requiem has achieved approximately 3.5% to date, above the initial estimate of 2.5% and broadly in line with the average for Resident Evil 7 and Village. These comparisons provide a reasonable initial range when a new title has just been announced and no direct demand data are yet available. The second stage uses Steam Hub followers to gauge pre-launch interest. The current sample comprises 58 games. Follower counts generally rise continuously from the announcement date and jump noticeably around trailers, trade shows, gameplay showcases, and other marketing events. The report focuses on comparing follower counts 90 days, 30 days, and one day before launch with initial sales, finding a generally robust logarithmic relationship between the two, although conversion rates vary widely among games. Historical conversion rates within the same IP or publisher are more informative than a single market-wide model. For example, the conversion from followers to PC sales is relatively stable across the Resident Evil franchise, whereas the conversion rates for Silent Hill 2 Remake and Silent Hill f differ more substantially. Because the sample still contains only 58 games, the report currently uses targeted comparable samples rather than constructing a complete statistical model for this stage. Follower growth curves are also used to interpret marketing events. The report notes that Steam followers for Resident Evil Village nearly doubled after Lady Dimitrescu's debut became a social media sensation, while Resident Evil: Requiem accelerated markedly around the reveal of Leon Kennedy and Capcom's showcase event. For Resident Evil: Code Veronica, the report recommends monitoring the response to the reveals of Chris Redfield and Albert Wesker. Approximately 70 days after its announcement, the game had accumulated about 65,000 Steam Hub followers, 27% and 49% more than Resident Evil: Requiem and Resident Evil 4 Remake, respectively, at the same post-announcement stage. The report initially assumed approximately 3.5% console penetration based on prior franchise titles. Although the game is not expected to launch until Capcom's FY3/28E, making a T-90 comparison impossible at present, and acceptance of horror games on Switch 2 remains uncertain, current signals support optimism regarding this flagship product and help underpin expectations for Capcom's pipeline. Signals for other games are not uniformly positive. Steam follower performance for The Blood of Dawnwalker is considered relatively solid, a positive sign given the report's previous concerns about the risks associated with a new IP. Steam momentum for Silent Hill Townfall is relatively weak, but Konami has positioned it as a smaller-scale spin-off than Silent Hill 2 Remake and Silent Hill f. The report therefore expects that even if sales fall short of expectations, the impact on overall company earnings will be limited. For games already released, the report extrapolates first-year sales of approximately 9 million copies for Resident Evil: Requiem and 3 million copies for Pragmata based on the pace of review growth. The third stage uses cumulative post-launch Steam reviews as a proxy for sales momentum. The database matches 385 official sales announcements across 68 games with the corresponding number of Steam reviews on each announcement date. The first set of review-count regressions uses variables including the current number of reviews, days elapsed since launch, launch year, and release platforms to forecast review counts at D+30 and D+365. The second set of sales-forecast regressions then uses review counts as the primary explanatory variable and adds the effect of Steam platform expansion over time through a separate track to estimate PC sales and total sales. The two forecasting tracks are averaged to reduce the influence of any single specification. The report's example is a game released in 2026 on PC, PlayStation, and Xbox, with an assumed PC sales share of 45%. If it has received 10,000 reviews 10 days after launch, the model projects an average of approximately 16,069 reviews at D+30 and 36,604 at D+365, corresponding to total sales of approximately 1.6 million copies at D+30 and 2.8 million copies at D+365. The R² values for the two D+30 review-regression tracks are 95.0% and 93.9%, respectively, while their leave-one-out mean absolute percentage errors are 10.4% and 11.3%. The D+365 R² values fall to 81.5% and 78.9%, reflecting greater uncertainty in longer-term extrapolation. AI is primarily used to build and maintain the process rather than replace final judgment. The team uses the Hermes agent framework and models including GPT-5.5/5.6, GLM-5.2/5.3, and Kimi K3 for prototyping, data extraction, filtering, cross-checking, and event interpretation. Different bots red-team one another, after which humans resolve the remaining data issues. The regression models retain a more manual Excel-based construction because the report believes this approach is less error-prone, easier to audit, and more conducive to understanding the model's mechanics. The principal limitation is the uneven sample composition. The data useful to the model effectively begin only when SteamDB started collecting reviews more systematically in 2017. Capcom accounts for 8 of the 58 games in the second stage and 173 of the 385 official sales announcements in the third stage, with Resident Evil alone accounting for 97 records. The model may therefore resemble a "Capcom forecasting machine." The report plans to continue expanding the database through scheduled tasks or manual triggers, but explicitly states that human validation will remain a permanent part of incremental data additions, with a particular need to increase the sample of publishers other than Capcom. The ratings table shows Outperform ratings for Nintendo, Capcom, Konami, and Nexon, with respective target prices of JPY 10,300, JPY 4,700, JPY 27,200, and JPY 3,300. Bandai Namco and Square Enix are rated Market-Perform, with respective target prices of JPY 5,300 and JPY 3,200. The report does not indicate any changes to these ratings or target prices relative to prior coverage.

Analysis framework

The report first establishes an initial sales assumption at announcement using the console installed base, penetration of comparable prior titles, and PC sales share. It then tracks Steam followers and marketing events to update its assessment of pre-launch demand, before extrapolating D+30 and D+365 sales using review-growth and sales regressions. AI agents handle prototype iteration, data extraction, cross-checking, and high-frequency event interpretation, while humans perform final data validation and model audits.

Methodology notes

• Industry/Sector Analysis FrameworkVolume-price decomposition

  Decomposition of console installed base, franchise penetration, and PC sales share

  The report combines the potential console user base with penetration rates for comparable games, then adjusts for the PC share of different publishers and IPs to form an initial sales estimate when a game is announced.

• Quantitative/Factor/Portfolio Theory

  Two-stage multivariate linear regression

  The first set of regressions extrapolates current review counts to D+30 and D+365, while the second set converts the forecast review counts into PC and total sales. The model includes both a base track and a platform-adjusted track and averages the two.

• Event-Driven Strategy and Behavioral FinanceEvent-driven analysis

  Analysis of Steam follower growth before and after marketing events

  The report observes changes in follower curves around character reveals, trailers, showcase events, and demo announcements, and compares them with prior titles in the same franchise to distinguish organic growth from event-driven changes in demand.

• Quantitative/Factor/Portfolio Theory

  Leave-one-out error and goodness-of-fit testing

  The report uses R² to measure how much sample variation the regressions explain and leave-one-out mean absolute percentage error to test forecasting stability when individual observations are excluded sequentially.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Capcom (9697.JP)

  Early Steam followers for Resident Evil: Code Veronica are significantly ahead of comparable franchise titles, while the pace of post-launch reviews for Resident Evil: Requiem and Pragmata also supports strong sales estimates.

  Strengths

  Capcom's PC sales share reached 54% in FY3/26, and the company consistently and frequently discloses sales milestones for major games, providing a deep data foundation for the model.

  Weaknesses

  The model sample is highly concentrated in Capcom, and the positive results may partly reflect its data availability and long-term consistency of execution.

  Comparison

  Approximately 70 days after announcement, Resident Evil: Code Veronica had 27% and 49% more followers than Resident Evil: Requiem and Resident Evil 4 Remake, respectively, at the same stage.

  Risks

  The release date has not yet been confirmed, with the report only estimating FY3/28E; the acceptance of horror titles on Switch 2 is also uncertain.

• Konami (9766.JP)

  Steam momentum for Silent Hill Townfall is relatively weak, but the game is positioned as a smaller-scale spin-off, and the report expects any potential sales shortfall to have a limited impact on overall earnings.

  Strengths

  Silent Hill 2 Remake and Silent Hill f provide same-IP comparison samples for the new title.

  Weaknesses

  The conversion rates from Steam followers to sales differ substantially between the two existing Silent Hill games, reducing the reliability of a single comparable relationship.

  Comparison

  Silent Hill Townfall is smaller in scale than Silent Hill 2 Remake and Silent Hill f.

  Risks

  Pre-launch Steam momentum is currently relatively weak.

Key data

• Pre-launch follower sample58 gamesUsed for second-stage Steam Hub follower comparisons; the sample is too small to support a complete statistical model.
• Post-launch sales database68 games, 385 official sales announcementsUsed to match Steam review counts with official sales and construct the third-stage regressions.
• Resident Evil: Code Veronica Steam followersApproximately 65,000Approximately 70 days after announcement, 27% and 49% more than Resident Evil: Requiem and Resident Evil 4 Remake, respectively, at the same stage.
• Initial console penetration for Resident Evil: Code VeronicaApproximately 3.5%Benchmarked against historical franchise performance; the report expects a possible launch in Capcom's FY3/28E.
• Monster Hunter Wilds console penetrationApproximately 4%Below the approximately 6% to 7% for Monster Hunter World and Monster Hunter Rise.
• Resident Evil: Requiem console penetrationApproximately 3.5%Above the report's initial estimate of 2.5% and close to the average for Resident Evil 7 and Village.
• Number of commercially relevant new Steam titlesIncreased from approximately 600 to approximately 1,000Refers to new titles with more than 1,000 reviews, increasing from approximately 600 in 2017 to approximately 1,000 in 2025.
• Capcom PC sales share54%FY3/26 data illustrating differences in publishers' PC mixes.
• Estimated first-year sales of released gamesResident Evil: Requiem approximately 9 million copies; Pragmata approximately 3 million copiesEstimated based on the pace of Steam review growth.
• Illustrative game D+30 forecast16,069 reviews, approximately 1.6 million copiesAssumes 10,000 reviews 10 days after launch, release on PC/PlayStation/Xbox, and a 45% PC sales share.
• Illustrative game D+365 forecast36,604 reviews, approximately 2.8 million copiesOne-year average forecast for the same example.
• D+30 review-regression performanceR² 95.0%/93.9%; LOO MAPE 10.4%/11.3%Corresponding respectively to the review-driven track and the platform-adjusted track.
• D+365 review-regression goodness of fitR² 81.5%/78.9%Below the D+30 model, indicating greater uncertainty in long-term extrapolation.
• Capcom sample concentration8/58 games in the second stage; 173/385 records in the third stageThe Resident Evil franchise accounts for 97 records in the third stage.

Impact & implications

The report believes the three-stage funnel can continuously update sales assumptions as information about a new title accumulates, reducing reliance on subjective judgment or static comparable relationships. The most direct company implication at present is that Resident Evil: Code Veronica's follower performance strengthens confidence in Capcom's FY3/28E product pipeline. Although Silent Hill Townfall's momentum is relatively weak, its smaller project scale means the expected impact on Konami's overall earnings is limited. The model's explanatory power for publishers other than Capcom still requires validation with additional data.

Risks

• The second stage includes only 58 games, so the report uses targeted comparable samples rather than constructing a complete statistical model.
• Capcom accounts for 173 of the 385 official sales records in the third stage, while the Resident Evil franchise accounts for 97, so the model may not be equally applicable to other publishers.
• Conversion rates from Steam followers to final sales vary widely, and comparisons across publishers or IPs may be distorted.
• Social media sentiment, trailer views, and responses to marketing events are highly context-dependent and provide only indirect signals of demand.
• The release date for Resident Evil: Code Veronica has not yet been confirmed, and market acceptance of horror titles on Switch 2 remains uncertain.
• The Blood of Dawnwalker still carries the new-IP risks previously highlighted by the report.
• Human validation remains a permanent component of the data pipeline, and automated extraction and agent cross-checking cannot eliminate all data errors.

What to watch

• Track changes in the Steam follower curve for Resident Evil: Code Veronica following the reveals of Chris Redfield and Albert Wesker.
• Once the release window is confirmed, compare Resident Evil: Code Veronica's follower levels at T-90, T-60, T-30, and one day before launch with prior franchise titles.
• Continue monitoring Steam follower growth for The Blood of Dawnwalker to determine whether the early positive signal persists.
• Track pre-launch momentum for Silent Hill Townfall and its conversion performance relative to Silent Hill 2 Remake and Silent Hill f.
• Monitor the pace of review growth for Resident Evil: Requiem and Pragmata and whether their first-year sales approach the estimates of approximately 9 million and 3 million copies.
• Watch whether the database adds more samples from non-Capcom publishers and improves the model's ability to draw inferences about other companies.