ARENA: one sporting world, a business worth testing

Product vision, competitive research and founder operating plan · 9 September 2026

Recommendation: build the broad experience, activate licensed coverage in affordable waves, and expand when fans repeatedly choose it. Your existing team covers development; external spending still needs to pay for dependable data, distribution and operation. ARENA is a working name, subject to trademark and domain checks.

This report separates verified source facts, proposed product advantages and explicitly hypothetical financial scenarios. The global product vision comes first. The later football-focused research and pilot economics provide one optional starting route; they are not an all-sport licensing quote or a limit on the design.

ARENA should become the place a fan opens to follow their whole sporting world: understand a match, monitor several events, catch up after an absence, investigate a player, or see what is confirmed about a developing story. Its ambition includes every sport. The current thirteen families establish the foundation; they are not a claim to cover every competition worldwide.

The product vision is multi-sport, supported by commercial discipline: your development team can build a broad, sophisticated experience, while public live coverage expands according to affordable licences, dependable data and demonstrated demand. A minimal external budget changes the activation sequence, not the ambition.

ARENA v06: fantasy, collecting and a social exchange

The private preview now connects real football statistics to an account-backed fantasy game, player profiles, collectible cards and fan conversation. This release adds functioning server/database features; the original 13-sport score examples remain illustrative. A wide directory is not a claim of worldwide live coverage.

A reason to trade the same footballer

The proposed hook is a Player Passport. Two managers can own Haaland, but one edition may carry a verified gameweek stamp and several ownership chapters while another is newly issued. A manager can collect a different story through a same-player swap, or trade a spare position for a player they need. The edition's story moves with the card; the underlying real-world performance and fantasy scoring stay the same. This is a product hypothesis worth testing, not a claim that no competitor has provenance or card trading.

Social value comes from the trading terrace, clear public offers and inspectable history. The first test should ask whether fans actually care about another edition's story, understand the offer in a few seconds, and return for the next window. If provenance alone does not create desire, test optional visual collections or club-themed cosmetic albums before changing competitive scoring. Avoid introducing artificial scarcity claims: edition numbers are unique, but the same base player can be acquired again.

The implemented credit economy

A new club receives a playable starter XI and 200 credits. The aspirational blueprint XI costs 298 credits if bought in full; each signing is also available at a fixed published price. A pack costs 40 credits and reveals three different base players. Each draw has a 75% Common, 20% Rare and 5% Elite tier chance; selection within the tier is uniform. These are per-draw odds, not a promised rarity in each pack. Owned duplicates return 2, 5 or 10 promotional credits by tier. Three Elite duplicates refund at most 30, so a pack cannot generate credits merely through duplicate refunds.

The daily grant is 10 promotional credits per UTC day. A valid, fully settled Classic round earns floor(net points) credits clamped to 0–60, once, after all fixtures finish and a 26-hour correction window after the last kickoff. Transfers and captain effects follow the versioned Classic rules. Monthly rankings automatically include eligible locked squads for all settled scheduled rounds; declining to claim a negative round cannot improve a ranking. Card rarity, spending and trade history do not improve that competition score.

Only unspent performance earnings and received trade proceeds can fund exchanges. Signup, daily and duplicate credits stay promotional. Ordinary purchases spend promotional funds first; later promotional grants cannot replace earned coins already spent. Offers reserve a specific edition, and both card and coin movements commit together or neither changes. Simultaneous acceptance, retries, cancellation races and window-end delays are covered by isolated transaction tests. These controls limit simple farming; public launch still needs real-account abuse monitoring and operational review.

Adaptation from BetTheBookie-Rebuild

With the owner's explicit permission, ARENA uses an attributed historical data sample from BetTheBookie-Rebuild, pinned at revision 1a510223ccb0105188883c118d03d1cbe02ebc10. The included data and normalized derivative retain CC BY-SA 4.0 attribution and notices. The upstream application is GPL-3.0; ARENA's elementary inverse-odds and margin calculations were independently implemented and no upstream application source was copied. The replay exposes source timing, market margin and assumptions rather than claiming its past prices predict future profits. Replay data and notices.

Practical launch sequence

Keep the broad design but fund one dependable live-data pilot first. Agree the selected sports/competitions, player events, historical storage, corrections, fantasy derivatives, images and web/app territories with the provider. Replace or explicitly license prototype FPL, ESPN and news use. An accessible endpoint is not a commercial licence. The Premier League terms, FPL terms, and BBC terms are starting references for the required permissions.

Activate registered Google/Apple clients, account recovery and deletion, moderation ownership, live-feed monitoring and source correction handling before inviting a public audience. Connect an approved AI service with usage caps only once source-processing permission is settled. The private preview currently permits only its owner; trading and company leagues need additional authorized testers before a real social trial. Production rollout also requires season-scoped identifiers, settlement jobs, leaderboard materialization, pagination and load tests rather than the small-preview request-time scans.

For monthly vouchers, first agree a funded sponsor, eligibility, dates, scoring/correction policy, tie resolution and fulfilment. Publish the final competition terms before entries. Until then, show the prize idea as planned. Start by measuring completed legal squads, second-matchday return, useful trades, repeat window participation and conversations with other fans. Monetization can then be tested through optional premium analysis, ad removal and sponsored competitions; no revenue outcome is established by this prototype.

A richer product with a simple structure

The organizing principle is glance, understand, investigate. At a glance, show the score and the state that matters. One selection reveals the important changes. A further step opens native statistics, evidence, comparisons and scenarios. Richness comes from connected depth, not displaying every available number simultaneously.

Keep the same recognizable structure across sports: the day’s events, a personal watchlist, the live desk, exploration and news. Within that structure, every sport should speak its own language. Cricket needs innings and legal balls; tennis needs points and sets; racing needs a field of competitors and session states. A reusable visual system should not erase these differences.

The competitive standard is already high. ESPN offers favorites, personalized SportsCenter, viewing information and supported-device multiview. Sofascore sells AI match analysis. PGA TOUR combines player personalization, shot comparisons and AI commentary. ARENA’s opportunity is to make a coherent set of workflows especially useful; adding AI or increasing the number of tabs does not establish an advantage. ESPN app, Sofascore Analyst, PGA TOUR product updates.

ARENA v04: a real historical foundation for deeper analysis

ARENA now includes 760 recorded Premier League results across 2023/24 and 2024/25, with results browsing, historical standings, recent form, head-to-head comparisons, match forecasting and six player profiles. The two seasons contain 380 completed fixtures each. The implementation pins an OpenFootball source revision under CC0 and records its provenance; the results were cross-checked against Football-Data.co.uk. Coverage ends on 25 May 2025. This archive adds real historical depth to the wider thirteen-sport interface, whose other representative events remain demonstrations. 2023/24 snapshot, 2024/25 snapshot, source licence.

The historical table accounts for dated points decisions. Everton's adjustment changes from −10 on 17 November 2023 to −6 after the 26 February 2024 appeal, then −8 after a further deduction on 8 April. Nottingham Forest receives −4 from 18 March 2024. These decisions affect the overall table from their effective dates; home/away splits show points earned in those matches. This preserves the difference between sporting results and administrative adjustments. H2H and form expose the actual earlier meetings and their dates; two seasons of records are not an all-time archive. Everton original decision, appeal, further deduction, Forest decision.

Forecasts that can be inspected

The new goal model is a transparent, deterministic Poisson baseline. It calculates home/away scoring and conceding strength from earlier results, gives more weight to recent matches, and shrinks limited team samples toward league averages. Users can inspect projected goals, home/draw/away probabilities, possible scorelines and the supporting history. The projected-goal rates are model estimates; they are not observed shot-quality xG. This implementation is a simple baseline, not the full Dixon–Coles model or a validated guarantee of predictive advantage. Dixon and Coles' original football-model research.

Two modes serve different questions. A hypothetical comparison estimates an explicitly chosen matchup under the selected historical cutoff. A historical pre-match replay reconstructs a forecast for an actual archived fixture, excluding that result and every result on or after its date. Users can then reveal the recorded outcome. H2H is supporting context with no extra model weighting. The current inputs contain no injuries, lineups, player availability or betting odds. Reconstructed estimates must never be described as forecasts issued before those matches actually happened.

The backtest targets the latest 80 fixtures in the selected season and reports the number meeting its minimum-history requirements. It compares the fixed model with an earlier-results frequency baseline using accuracy, multiclass Brier score and log loss, alongside probability bins. These measurements should be read directly from the implementation; no performance figures are assumed here. Changing scenario settings does not transfer the fixed model's validation to the changed configuration. A small retrospective sample needs more seasons, calibration checks and an independently reserved evaluation period before commercial accuracy claims. Chronological evaluation, probability calibration.

Player outlook adds six sourced 2024/25 profiles with appearances, minutes, goals, assists, shots and comparable per-90 rates. A playing-time control defaults to 75 minutes and adjusts conditional projections. These multiply the historical rate by chosen minutes; count probabilities and intervals add a Poisson assumption. They do not estimate whether the player starts, current injuries, changed roles or opponent effects, and their intervals are not empirically calibrated. Full-season aggregates through 25 May 2025 never enter an earlier match replay. Understat's minute definition is retained consistently, and commercial player-data rights remain unconfirmed. Understat season source.

What must happen before production

The explanations currently use structured templates and computed facts. There is no external LLM or live prediction feed. A later AI layer should receive a bounded packet containing the forecast, eligible evidence, model version and missing inputs; it should explain those facts, link its claims and fall back to deterministic text when validation fails. It must not invent probabilities or player availability. Cache explanations by data revision and language, set spending limits, and measure factual errors before expansion.

Connect licensed schedules and live events through a validated ingestion service with stable identifiers, revisions, freshness checks and historical snapshots. Buy player appearances, availability and deeper statistics only for the first competitions that users repeatedly follow. Add sport-specific models in stages: basketball needs possessions and scoring context, tennis needs serve and format, cricket needs innings and legal deliveries, and racing needs a field and session state. A football model cannot simply be relabeled for thirteen sports.

Competitors already offer prediction features: Sofascore documents probabilities, recent form, H2H and postmatch comparisons. ARENA's opportunity is useful evidence, coherent workflows and demonstrated repeat use. Existing development capacity and local computation make this affordable to test; dependable all-sport coverage still requires phased data spending and operation. Forecast accuracy cannot guarantee superiority or profitability. Sofascore Analyst.

What the current build establishes

The current ARENA v05 interface opens on scores and demonstrates thirteen sport families, native match centres, a multi-event desk, scenario controls and user preferences. Three connected tools make the proposed advantage tangible: Rewind, Your Rules, and Check the Story. Rewind and the story audit use one synthetic Arsenal example; the rule builder provides native predicates for all thirteen sport families. These are local demonstrations, not records of real matches or reporting.

The original multi-sport score demonstration has no live feed. The v06 football/account connections are described above; a language model, notification delivery and payments remain unconnected. Replay, rule evaluation and correction cascades run against prepared data. A chart reacting to a selection proves an interaction; a scenario responding to a slider illustrates assumptions. The separate v04 prediction workspace now implements and evaluates a statistical baseline against real archived results; the original live examples remain synthetic. The interface uses recognizable sporting identities and public image URLs; their commercial rights have not been cleared. These boundaries remain visible through demo labeling and the asset register.

Use the build to test complete tasks: find an event, understand its state, follow several sports, inspect the evidence behind an explanation, compare compatible participants, and adjust preferences. Evaluate the smaller screen as seriously as the desktop. Every prominent control should produce a meaningful result, including empty and unavailable states.

Thirteen families, native depth

This matrix describes the intended product architecture. Advanced fields depend on contracted competition coverage; unavailable statistics should be absent or explicitly unknown.

Formats need separate rules: indoor and beach volleyball, rugby union and league, boxing and MMA, different racing series and esports titles. Detailed data is also uneven within a sport. Sportradar documents tiered tennis coverage and distinct global/league products; PandaScore separates fixtures, historical and live entitlements. Design around those differences from the start. Sportradar coverage model, tennis coverage tiers, PandaScore plans.

The first production priorities

Establish trustworthy live state. Select an affordable, coherent group of competitions across the sports that reachable users actually follow. Validate schedules, identifiers, native scores, corrections, postponements and final results before asking people to rely on notifications. Display source freshness honestly. A live event can contain statistics that update less frequently.

Connect the personal experience. Following an athlete, team, competition or individual event must have a clear meaning. Preferences should carry across devices; timezone, quiet hours and spoiler settings should apply consistently. The live desk should preserve selected events and make switching effortless. Core scores should remain useful without mandatory registration.

Deliver grounded catch-up. “What changed since I checked?” is more useful when it knows the relevant interval and links each statement to events. Calculate scores, rates and standings in ordinary code. Supply the explanation system with only relevant, permitted evidence. Check names and numbers, and regenerate or withdraw explanations when facts change. The live score service must work when AI is unavailable.

Add reporting with a defined editorial scope. Begin with original analysis and permitted links to reliable reporting. Connect availability and roster developments to affected events. Distinguish official announcements, attributed reports, denials and outdated claims. “Transfers” should adapt into trades, free agency, driver changes or esports roster moves. Repetition across aggregators is not independent confirmation.

AI that earns its place

Use AI to reduce effort: explain an unfamiliar rule in context, summarize a missed interval, find evidence within licensed records, translate a briefing accurately, or turn an alert request into a rule the user can inspect. Keep generated interpretation visibly distinct from observed data.

Comparisons must preserve context: surface in tennis, format in cricket, possessions in basketball, minutes in football, patch and map in esports. Show sample size and missing information. Predictions require their own historical evaluation and uncertainty; an LLM’s apparent confidence is not a calibrated probability. Scenarios should expose assumptions and remain separate from forecasts.

The longer-term opportunity is an explanation system that improves through reviewed corrections and user feedback, together with trustworthy language quality and useful personal context. Source relationships, verified event history, community distribution and accumulated product learning are harder to copy than a chat box. These are assets to build, not a moat already achieved.

Global breadth within a small external budget

The optional football pilot’s cash allowance cannot be reused as an all-sport production budget. Basic API access, detailed events, tracking, imagery, reporting, video and AI-processing rights can be separate purchases. API-Sports lists several sport products, but its catalogue does not encompass every family in this vision. One account is insufficient evidence of one unrestricted paid bundle. Obtain a written coverage and entitlement schedule. API-Sports catalogue, API-Sports terms.

Activate coverage in layers: schedules and results, verified live scores, events and statistics, then advanced analysis. Make additional sport hubs discoverable with truthful coverage information. A wave can include several sports without purchasing worldwide depth at once.

Before committing, price provider access, competitions, territories, request limits, historical storage, notifications, images and permitted AI use. Share ingestion and cached explanations across users; avoid a model request for every screen refresh. Reusing software lowers development expense, while licences, support and editorial work still require funding.

Launch gates and commercial model

The following are decision gates, not industry benchmarks:

1. Rights: written permission covers the actual public display, territories, notifications, retention and intended AI processing. Required costs fit the agreed spending cap.
2. Reliability: representative replay and live-trial cases handle revisions, missing fields and interrupted events; outages produce an honest stale state. An identified person owns incidents during supported coverage windows.
3. Comprehension: fans from different sport segments complete the core tasks accurately. Compare ARENA with their existing workflow and measure time, errors and preference.
4. Repeat use: activated testers return on relevant event days, use saved follows and revisit catch-up or the desk. Measure by sport and fixture opportunity rather than total registrations.
5. Economics: actual subscription or sponsor evidence supports expansion; measured delivery, data, AI and support costs remain sustainable as usage grows.

Keep scores, basic details and essential alerts free. Test paid value around persistent advanced workspaces, deeper saved comparisons, configurable alerts and useful briefing delivery. Charge for repeated benefit rather than an “AI” badge. Direct sponsorship can fund a relevant audience; restrained advertising can supplement it. LiveScore’s published offerings demonstrate these media revenue categories, but disclose no guaranteed achievable rates. LiveScore advertising and sponsorship.

Model contribution as subscription receipts plus publisher-net advertising and signed sponsorship, minus data, delivery, AI and operating costs. Include your team’s time when assessing profitability, even when payroll already exists. Treat uncontracted sponsorship, betting referrals and future B2B revenue as zero. Publisher tools or community integrations can become later businesses where licensing and support economics allow.

Expansion should follow retained use and dependable operation. The full vision remains broad: one carefully designed sporting home, with deep native detail and a dependable memory of what matters to each fan. Superiority will be earned in those repeated experiences and demonstrated through evidence.

1. Is it worth it?

Your ability to build apps and websites is a meaningful advantage: you can test several ideas without paying an outside agency. The harder constraints are acquiring regular users, buying appropriate data rights, delivering reliable updates during matches, and making enough money per user to sustain operations.

A worldwide audience is a reasonable long-term ambition. “Everyone who follows football” is a poor first customer definition. Fans of different competitions, languages and viewing habits have different needs, and expanding coverage multiplies testing, content and support demands. An interface available worldwide does not automatically have worldwide data, content or payment rights.

Choose the first audience through access: a supporter community you can recruit, a creator who can introduce the product, or fans in a language you can serve exceptionally well. Select the first licensed coverage wave through reachable demand. One sport and one to three competitions is a low-cost pilot option; a coherent small wave may span several sports if its rights and operating costs fit. A Champions League fan also cares about domestic matches; choose a coherent cluster rather than three unrelated cheap feeds.

The immediate goal is to win a recurring occasion: catching up after missing part of a match, following several clubs while working, or checking what is confirmed about a transfer. Fans may use you alongside an incumbent before making you their primary app. That is useful progress.

Development capacity should buy learning, not a giant feature list. First prove that a defined audience returns next matchday and values the core experience enough to pay or attract a sponsor. A beautiful launch followed by weak repeat use is not product-market fit.

2. What the competitors already do

Verified facts: scores, notifications, lineups, tables, news, transfers, ratings and advanced statistics are established expectations. AI is already part of this category.

These descriptions come from official company pages, FAQs and developer listings, not an independent performance benchmark. LiveScore's September 2025 AI announcement describes a partnership and intentions; it does not prove every proposed feature shipped. [1] [2] [3] [4] [5] [6] [30]

FotMob already documents delayed alerts and favorite-league customization; its listing includes Persian and Arabic. These are useful capabilities, not untouched markets. [7] [5]

Avoid comparing company audience headlines as if they were the same metric: annual reach, group users, app downloads and monthly active users are different. Public evidence reviewed here does not establish current standalone LiveScore.com revenue or profit.

Fresh workflow review for v03

The 9 September 2026 review covered the public product pages, official help articles and developer-maintained app listings of LiveScore, Sofascore, FotMob, Flashscore and ESPN. It was not authenticated, hands-on testing of every device, country or paid tier. Public descriptions are incomplete; the exact combined workflows below were not advertised in the pages reviewed, which does not prove global uniqueness or absence from those apps.

The stronger hypothesis concerns complete actions: reconstruct and compare two checkpoints, test a compound alert rule against a replay, or trace a claim and its later correction through saved material. Fans must find these actions useful enough to repeat; feature count does not answer that question.

3. The advantage to test

Recommendation: make scores effortless, then let fans inspect changes, control attention and check claims. Place the relevant tools beside the event and make their results explainable.

None is a proven exclusive feature. The edge would come from executing the combination better for your chosen audience. The assets that may compound are trust, a well-maintained correction system, useful preference history with consent, community relationships, strong language quality and eventually distinctive licensed data. An LLM subscription or attractive visual style can be copied quickly.

The prototype proves only that these interactions can be made concrete. Rewind and Check the Story currently share one synthetic Arsenal event; they do not implement unrestricted historical reconstruction or automatic auditing of news from the internet. Your Rules covers all thirteen sport families in local simulation, but reliable background delivery requires production data and notification services. Running deterministic logic locally avoids a per-use model charge; it does not make a live sports service cost-free.

4. How money works

Verified facts: LiveScore Group describes an ecosystem combining LiveScore with LiveScore Bet and Virgin Bet. Its historical £10 million monthly revenue milestone refers to the group in 2022; it cannot be used as current score-app revenue, profit or a forecast for your company. [8]

LiveScore sells display inventory, sponsorships, video placements, audio commentary sponsorship and branded social content. These are real media revenue streams, but public sales material does not disclose achieved rates or margins. [9]

Recommended launch sequence:

1. Free utility: reliable scores, basic match details, favorites and core alerts build habit.
2. Optional Pro: test $2.99–$3.99 monthly for repeated value. Ad removal alone faces low-price competitors: FotMob's US listing includes $1.99 monthly and $15.99 annual purchases, subject to market/current-offer differences. [5]
3. Direct sponsorship: a clearly labeled league or briefing sponsor may cover early bills with a small relevant audience. Count nothing until a buyer agrees to deliverables and price.
4. Restrained advertising: monetize free usage after approval and adequate traffic. Revenue depends on actual monetized impressions and audience, not installs.
5. Later referrals or B2B: approved merchandise/ticket partners or publisher tools may help, but require contracts, support and appropriate rights.

Betting affiliations may pay for qualifying customer acquisitions or a contractual revenue share. They are country-dependent and should contribute zero to the launch model. Operating a sportsbook introduces a separate regulated business, payments, trading/risk and responsible-gambling operations. It is unnecessary for this concept.

For ads, use publisher-net eCPM: net receipts per thousand monetized impressions. Google's AdSense description distinguishes the publisher share after buyer-platform fees; do not deduct an ad-network fee twice. [10]

5. Where live information comes from

Verified fact: Enetpulse publishes a LiveScore client case describing updates pushed as XML to a LiveScore server endpoint when match events occur, followed by publication and later checks. Enetpulse is therefore a publicly documented LiveScore data partner. The undated case does not establish today's exclusive supplier mix. [11]

Specialist suppliers collect and normalize information before distributing feeds. Opta describes expert analysts, optical tracking and proprietary AI among its methods. Your company buys the relevant feed and builds the fan product; creating a global collection network is a different business. [12]

Public prices are not binding quotes or complete commercial licenses. Currency, VAT, add-ons, term and rights can change the bill. [13] [14] [15] [16] [17]

Sportmonks' broad feature wording and its documentation are not fully aligned: documentation identifies xG, pressure and predictions as add-ons. Buy against a written field-level offer, not a marketing headline. [18]

“Live” is a chain of delays

A goal happens; a collector detects and validates it; the provider updates; your service receives it; devices render or receive notifications. Website speed improves only part of this chain.

API-Football advertises 15-second updates. Sportmonks describes typical event-to-API latency around 15 seconds and differentiates measured uptime from a contractual promise. Neither means every fan sees every event 15 seconds after it occurs. [19] [20]

Poll centrally and reuse responses. One request every 15 seconds for 24 hours is 5,760 requests/day. That leaves 1,740 on a 7,500-request plan for other requests; details, retries and pagination can consume them. The free 100-request allowance cannot support continuous useful live polling. This arithmetic is an example, not a production sizing guarantee. [13]

Measure provider receipt and device delivery separately; display actual freshness and outages. Do not advertise “fastest” from a comparison with one delayed television stream.

6. Rights and procurement before public launch

An API key is not a universal content license. API-Football permits app/web projects but restricts raw resale, disclaims competition commercial rights and flags visual-asset permissions. Sportmonks permits commercial products subject to its terms, while resale and branded assets need separate attention. [21] [22]

Recommendation: send a short procurement brief listing your entity, website/app, countries, competitions, languages and monetization. Obtain written answers covering:

• Every required field, historical depth, lineup timing and correction behavior by competition.
• Public display, notifications, caches, archives and post-termination retention.
• AI summaries, external model processors, translations, embeddings, training and derived analytics as separate uses.
• Crests, player photographs, article text, thumbnails, video and viewing guides individually.
• Web/iOS/Android/staging coverage, request/burst quotas, support during matches, annual cost and cancellation.

Reviewed generic pages do not settle all AI-processing rights. Neither assume permission nor conclude blanket prohibition. Media, betting and streaming rights are different scopes; Stats Perform's Football DataCo agreement illustrates competition-specific official media rights. [23]

Do not build the business on scraping competitors or republishing full articles. Use licensed feeds, original reporting, and permission-appropriate attributed links/excerpts. Legal review should address the actual uses and launch countries, not an abstract promise of worldwide coverage.

7. An optional football pilot cash budget

Planning ranges, not quotes: these exclude your existing team's salaries, founder time, company setup, tax/legal advice and any additional rights purchase. Start with one provider, one backend/worker and a responsive web product.

Set an operating target of $150–$300/month, with a three-month reserve of $450–$900, only if the chosen rights and services fit. Separate one-time costs and taxes from this reserve. A $500/month ceiling is a reasonable later experiment limit, not approval to spend or a promise of production capacity.

Prototype with sample data or a trial first. Share generated briefings across readers; avoid a new model call per page refresh. Do not buy a global newsroom, video rights, open-ended chat usage or paid installs before retention is visible. If required rights cannot fit the cap, narrow the launch or stop.

8. Transparent pilot financial scenarios

These are illustrations, not forecasts or industry benchmarks. MAU means unique monthly active users. Paying users below are active subscribers counted within MAU, and do not see ads. Subscription figures are receipts after the modeled fee, not accounting profit.

Base scenario assumptions

The 70% factor combines modeled fill, consent, blocking and viewability effects; do not deduct them again. The 15% fee is a scenario input: Apple's eligible Small Business Program offers that rate, but eligibility and associated accounts matter and other routes differ. [24]

Formulas

Ads = MAU × (1 − paid share) × sessions × opportunities × monetization factor × net eCPM ÷ 1,000

Subscription receipts = MAU × paid share × monthly price × (1 − fee rate)

Cash contribution = ads + subscription receipts − fixed operations − MAU × variable operations

Rounded components may differ by $1. The last two rows illustrate scaling, not a launch plan. Replace allowances with measured peak infrastructure, support and contracted data costs as usage grows.

Excluded: existing-team and founder compensation; acquisition spending; editorial labor; additional licensing and media rights; tax, refunds and currency effects beyond stated allowances; legal/entity expenses. Sponsors and affiliates contribute zero. The operations formula is a budgeting assumption, not a supplier offer.

Weak scenario

Assume 8 sessions/month, one ad opportunity, 50% monetization, $0.50 net eCPM, 0.3% paying at $3.99, and 21% subscription fee/refund allowance. Keep the same operating-cost assumptions.

Base revenue/MAU is $0.075495; contribution after variable cost is $0.050495. Therefore:

Break-even MAU = fixed costs ÷ (revenue/MAU − variable cost/MAU)

Base cash break-even is approximately 4,951 MAU before exclusions. With ads alone and everyone free under base ad assumptions, it becomes 14,706 MAU. In the weak case, revenue/MAU is below variable cost; there is no finite break-even without changing the economics.

An illustrative additional $5,000/month allocation to team time raises base break-even to about 104,000 MAU. That is not an estimate of your payroll. It demonstrates why cash-positive is not the same as profitable.

Do not buy installs from this model. Measure acquisition cost per retained user and actual cohort contribution. Annual subscriptions improve cash collection but must be analyzed as monthly equivalent revenue with refund obligations. Account for seasonality, churn and renewal behavior before estimating lifetime value.

9. What to build, give away and charge for

Make the core score experience useful without signup. Ask for an account when synchronization or saved history matters. Production team-following means future matches and settings persist; favoriting a single sample match in the prototype does not implement that system.

Keep goals and basic alerts free. A premium offer must explain an outcome, such as a saved weekly club briefing, rather than charging vaguely for “AI.” Test price and renewal behavior before filling a premium catalogue.

10. News and transfers

News can create off-matchday return, but a general feed is expensive to differentiate. Start with a small editor-curated briefing attached to selected clubs: original analysis, official announcements and attributed links. Deduplicate the same story across outlets. An AI paraphrase does not remove publication-rights obligations.

Sportmonks markets a €99/month news add-on with specified preview and post-match coverage; it is not a universal newsroom. Confirm languages, coverage and reuse rights before adding it. [25]

For transfers, store claims with sources, not one supposedly omniscient answer. Use statuses such as official announcement, reported talks, reported interest, denied, completed and outdated. Show original publication time, last verification, and what changed. Separate repeated aggregation from independent confirmation. Completed transfer records and live rumors are different products. Sportmonks lists a separate transfer-rumour add-on from €99/month, so adding both news and rumours can cost materially more than the base scores feed. [14]

Do not invent a transfer probability or “99% reliable” source score. Readers should see the evidence and uncertainty. Corrections must update the card, briefing and any associated notification. Add a human owner for sensitive or disputed claims.

11. How AI should work

Recommendation: use AI as a constrained explanation service downstream of verified match state.

1. Ingest facts: canonical match/player IDs, provider timestamps, event IDs, revisions and coverage flags.
2. Compute changes: a deterministic service calculates scores, standings and differences since a checkpoint. Routine goal alerts can use templates without AI.
3. Build a fact packet: only relevant licensed facts for that match, time window and language, each with a reference ID.
4. Generate a brief: require structured output and fact references. Separate observations from interpretation. Possession or xG alone does not prove tactical causation.
5. Validate: check every number, name and timestamp against inputs; reject unsupported claims, fall back to templates and state missing evidence.
6. Cache: generate once per match/revision/language, serve many users, personalize relevance cheaply. Invalidate outputs after corrections.
7. Evaluate: measure factual error, unsupported claims, source coverage, usefulness, latency and cost on a fixed set of real match cases.

Useful questions include “What changed since halftime?”, “Why did my team move in the table?” and “What is confirmed about this transfer?” A table explanation should use computed standings, including tie-break assumptions. Statistical explanations should distinguish definitions from claims about today's match.

Do not call an LLM for every refresh. Generate at meaningful changes, halftime and full time, with strict question quotas and spending alarms. If you later add forecasts, use a separately validated predictive model with held-out periods and calibration evaluation. The language model explains model output; it should not invent probabilities or promise betting returns.

12. Production architecture and quality

Start with one deployable backend and a worker:

Provider adapter → validated event store → current-state cache → your API → live client updates. A background summarization queue reads verified state. Favorites, accounts and notification preferences live in your own database. Keep provider keys server-side.

Poll live matches centrally, cache schedules/reference data longer, and reconcile final results again after the whistle. API-Football's integration guide describes server-side reuse and different refresh needs for different data types. [26]

Use stable internal IDs and adapters to allow provider migration. Preserve source and metric definitions; xG and ratings from different suppliers are not automatically comparable. Handle duplicates and revisions idempotently, retain an audit trail, and provide a simple operations console.

Push is not automatically self-recovering: Sportradar documents Soccer Push as an add-on and requires REST recovery after disconnects because prior messages are not replayed. Sportmonks' latest-updated endpoint has a limited lookback/field trigger scope, so it cannot be the only source for all changing analytics. [27] [28]

Measure platform delay from received event to device separately from supplier delay. Set internal performance targets only after a realistic provider trial. A stale banner and correct last-known score are better than fabricated freshness. Budget on-call ownership during selected matches before promising dependable global coverage.

13. ARENA v05 design and prototype boundary

Scores come first. The default view presents a populated score board with native sport state, compact competition groups, filters and a selected-event centre. Neutral surfaces, strong ink typography, restrained coral emphasis and compact top navigation establish the hierarchy. Mobile bottom navigation keeps the primary tools within reach. The user-requested Emil Kowalski design-engineering and Apple-design skills informed the layout and interaction revision. Matchup selectors sit inside the forecast, and replay reveals the real outcome beside its prediction. Recognizable team crests, competition marks and appropriate athlete images aid identification. Tabular scores and restrained live indicators give results priority. The opening experience is a useful sports tool, with deeper analysis available alongside the event. The ARENA name still requires trademark and domain clearance.

The demonstration retains thirteen sport families, event search, date/status filters, locally saved event follows, compact rows, spoiler masking, native match centres, a multi-event desk, comparisons and illustrative scenarios. Cricket uses runs/wickets and legal-ball arithmetic; tennis exposes sets, games and points; racing and golf show classification; baseball, American football and hockey display their native situation. This is representative depth, not complete competition coverage. Production tournaments and multi-entrant events still need robust event models, format rules and provider-specific entitlements.

Three working tools express the new direction:

The original Lab examples have no live feed; their scores, statistics, lineups, analytical text, scenarios and claim packets are synthetic. The v06 football/account features use the separate source and persistence paths described above. A connected LLM, notification delivery and payments remain unavailable. Recognizable real sporting identities do not make their accompanying example results or claims real. Prepared explanations and deterministic calculations demonstrate behavior; they do not validate a forecast or a model's factual accuracy. Existing calendar and preference controls operate on illustrative examples; event follows do not subscribe to every future match of a team or athlete.

Team marks, competition identities and athlete imagery are referenced through public URLs. Public accessibility is not commercial permission, and asset rights have not been cleared. The ARENA asset credits and source register records provenance; it is not a licence. Before public commercial use, secure the required permissions or replace assets with authorized alternatives. External availability can also change, so production needs stable, permitted delivery and appropriate fallbacks.

Evaluate desktop and mobile around complete tasks: identify the event, switch sports, follow it, inspect native detail, compare the selected checkpoints, tune and replay a rule, and understand a correction. A successful interaction review establishes usability of the demonstration. Reliability, comprehension against competitors, retained use and willingness to pay require trials with actual fans and appropriately licensed data.

14. Distribution, ownership and international readiness

Start distribution before building broadly. Recruit through fan creators, supporter groups, podcasts and communities where you can contribute something useful. Offer a relevant sourced match card or briefing with a deep link; track which partners bring returning fans. Use authorized outreach, not unsolicited bulk promotion.

Search pages may help over time, but generic AI previews are unlikely to be a quick acquisition advantage against established publishers. Paid acquisition should wait for measured retention and contribution economics. Choose a creator partnership because it reaches your segment, not because follower count looks large.

Own the operating work explicitly:

One person can cover several roles initially; none should be ownerless. Restrict coverage to the service window you can actually support.

Before selling internationally, confirm entity and payment eligibility, provider territories, taxes/refunds, privacy/consent requirements, account deletion, subscription disclosures and age-appropriate product choices. Translate names, formats, timezones and source context, not only navigation labels. Verify app-store and advertising availability for the actual business and audience. For example, AdSense lists restricted publisher territories; no location or eligibility conclusion should be inferred from your device timezone. [29]

15. A 90-day plan with decisions

The following are founder-set test gates, not industry benchmarks. Adjust them to competition calendars and cohort size before running the test.

Track favorite activation, next-matchday return, four-week return, useful brief consumption, factual errors, source opens, notification opt-outs, receipts and direct costs by cohort. Count fans with a relevant match opportunity; a fixture-free week should not be misread as product failure.

In comparative task tests, aim for a meaningful improvement such as 25% less time to understand a match with no worse factual comprehension. Treat this as an experiment target, not a claim to put in marketing before measurement.

Expand only after the first segment returns, the source/operations path is dependable and revenue evidence is credible. After two focused iterations, stop or reshape the concept if fans describe it as a novelty, the briefing does not earn preference, required rights break the budget, or retained users cost more to acquire than they contribute. Your team advantage makes this disciplined experiment possible; it does not remove the need for those results.

Sources and evidence notes

All sources below were accessed 9 September 2026. Dates are publication/update dates where verified; “undated” means no reliable publication date was established. Product and pricing pages can change. Company statements were used as evidence of offerings, not independent proof of quality or profitability. No vendor purchase, hands-on competitor usability testing or comparative latency benchmark was performed for this report.

1. LiveScore Group — Partnership with X and xAI, 17 September 2025.
2. Flashscore / Apple App Store — Flashscore Live Scores & News, current listing/release notes, accessed as above.
3. Flashscore — Advertising FAQ, undated.
4. Sofascore — What is Sofascore Analyst (AI Insights)?, updated 12 February 2026.
5. FotMob / Apple App Store — FotMob Soccer Live Scores, US listing, current listing, accessed as above.
6. OneFootball — Brand Solutions, undated page with June 2026 references.
7. FotMob — Frequently Asked Questions, undated.
8. LiveScore Group — Corporate overview and milestones, undated; historical milestones identified in text.
9. LiveScore Group — Digital Solutions, undated.
10. Google AdSense Help — AdSense revenue share, undated.
11. Enetpulse — Client case: LiveScore, undated.
12. Stats Perform — Opta brand and data products FAQ, undated.
13. API-Football — Pricing plans, undated current pricing page.
14. Sportmonks — Football API plans and pricing, undated current pricing page.
15. football-data.org — Pricing, undated current pricing page.
16. Sportradar — Sports Data API, undated.
17. Stats Perform — Power your products with Opta data, AI and APIs, undated.
18. Sportmonks — Football API v3 FAQ, undated documentation.
19. API-Football — Football API overview, undated.
20. Sportmonks — Data integrity and support, reviewed August 2026 according to page.
21. API-Football — Terms of service, page displays 21 May 2025 update in reviewed version.
22. Sportmonks — Terms of service, undated.
23. Stats Perform — Expanded partnership with Football DataCo, 12 December 2024.
24. Apple Developer — App Store Small Business Program, undated current program page.
25. Sportmonks — Football News API, undated.
26. API-Football — Complete beginner's guide, 13 March 2026.
27. Sportradar Developer — Soccer Push integration, documentation marked updated approximately four months before access.
28. Sportmonks — Get latest updated livescores, documentation marked updated approximately one month before access.
29. Google AdSense Help — Understanding AdSense country restrictions, undated current policy page.
30. OneFootball — Embed partnership terms and conditions, undated.