Daily News Narrator

Schema v2.0 | Generated: 2026-03-31 | Domain: al-futtaim-daily-news | Client: Al-Futtaim Group

Use Case

The Context

Al-Futtaim Group's Communications team operates a Daily News Narrator agent that delivers a curated intelligence briefing to Vice Chairman Omar Al Futtaim and the wider leadership team every morning before 9:30 AM. The briefing covers 5-6 news items across geopolitics, markets, competitors, and sector developments, each tied back to specific Al-Futtaim divisions (Automotive, Retail, Real Estate, Financial Services, Healthcare & Education). The agent replaces a manual process currently handled by Sreeram Muralidharan (Strategic Intelligence), saving 15-20 minutes daily. The output is delivered as a WhatsApp text message and a WhatsApp voice note (audio script synthesized via ElevenLabs).

Real-World Inputs

The agent receives a batch of 20-50 raw news articles scraped or pulled via API from 6-7 approved sources: Reuters, Wall Street Journal, CNN, South China Morning Post, The National (UAE), and Khaleej Times. Articles arrive as plain text or HTML-extracted content. The agent also has access to a static corporate context document containing Al-Futtaim's 5 divisions, 200+ brands, key markets (UAE, Saudi Arabia, Egypt), strategic priorities, and competitor/partner watchlist (Toyota, BYD, IKEA, Marks & Spencer, Cenomi Retail, Orient Insurance).

Three Layers of Difficulty

Realistic (bread-and-butter)
Clean, well-sourced articles from Reuters or WSJ. Clear relevance to Al-Futtaim sectors: a Toyota production announcement, an IKEA expansion in the Gulf, a Saudi PIF investment shift. The agent selects, summarizes, and writes implications with obvious division mappings. 70% of production volume.
Challenging (messy but real)
Truncated/paywalled articles from WSJ or SCMP with only the first 2-3 paragraphs visible. Tangentially relevant articles (e.g., a China EV policy article that indirectly affects BYD pricing in the Middle East). Stale articles (>24h old) mixed with fresh ones. Regional articles with Arabic entity names requiring contextual understanding. Articles covering multiple Al-Futtaim divisions simultaneously (e.g., a Dubai real estate + retail story).
Deceptive (intentionally misleading)
Clickbait headlines that appear highly relevant but have unrelated article bodies (e.g., "Toyota Faces Major Disruption" about a factory in Japan unrelated to Al-Futtaim markets). Articles about a different "Al Futtaim" entity or similarly-named company. Well-written opinion pieces disguised as news that could introduce bias into the briefing. Sponsored content or press releases formatted as news articles.
Litmus Test: "Would Sreeram Muralidharan look at this daily briefing and find it indistinguishable from one he would have written himself, including the specificity of Al-Futtaim implications?"

Input Schema

2 entities
Entity 1: News Article (batch of 20-50 per run)

Content

FieldTypeDescriptionConstraintsRequired
headline string Article headline as published 10-20 words, may contain quotes or em-dashes yes
body string Full or partial article text 200-800 words (full), 50-200 words (truncated/paywalled). See Content Texture below. yes
source_publication enum Publishing outlet Reuters | WSJ | CNN | SCMP | The National | Khaleej Times yes
source_url string Original article URL Valid URL format; may be behind paywall yes
published_at datetime Publication timestamp ISO 8601; typically within last 24h yes
region_tags array<enum> Geographic relevance 1-3 tags from: UAE, Saudi, US, China, Egypt, Global yes
sector_tags array<enum> Industry/sector relevance 1-3 tags from: automotive, retail, real_estate, financial_services, healthcare, geopolitical, trade, energy, technology no
Content Texture

Wire-service style (60%): Short 1-2 sentence paragraphs, facts-first, attribution in opening sentence ("DUBAI (Reuters) —"), inline quotes from officials or analysts, specific numbers early. 200-400 words. Structure: lead → details → reaction → context.

Analytical style (25%): Longer 3-5 sentence paragraphs, market data references, multiple analyst quotes with firm names, comparative data ("up 3.2% from Q3"), forward-looking projections. 400-800 words. Structure: thesis → evidence → expert views → implications.

Regional style (15%): Government official quotes (often from Arabic-language sources, translated), local regulatory context, UAE/Saudi policy references, entity names that may appear in Arabic transliteration. 250-500 words. Structure: announcement → official quote → local context → regional impact.

Presentation

Format Plain text 70% HTML (scraped) 30%
Noise Clean 60% Ads/nav removed 25% Extraction artifacts 15%

Variation

Completeness Full article 70% Truncated/paywall 20% Summary only 10%
Relevance Clearly relevant 40% Tangentially relevant 35% Irrelevant noise 25%
Freshness <6h old 60% 6-24h old 30% >24h stale 10%
News bucket key_markets 25% economic_financial 20% competitors_partners 25% geopolitical 15% sector_specific 15%
Entity 2: Corporate Context (static per run)

Content

FieldTypeDescriptionConstraintsRequired
divisions array<object> 5 divisions: Automotive (Toyota, Lexus, BYD), Retail (IKEA, M&S, Cenomi), Real Estate (Dubai Festival City), Financial Services (Orient Insurance), Healthcare & Education Each division includes: name, president, key brands/entities, markets yes
key_markets array<string> Primary: UAE, Saudi Arabia, Egypt. Extended: Middle East, North Africa, CIS (via Cenomi) Ordered by priority yes
competitors_partners array<object> Watchlist: Toyota, BYD, IKEA, Marks & Spencer, Cenomi Retail, Ace Hardware, Toys R Us, Orient Insurance peers Each entry: name, relationship (partner/competitor/both), division yes
strategic_priorities array<string> Sustainability/decarbonisation, AI & digitalisation, Emiratisation From corporate profile Oct 2025 yes
recent_events array<object> Recent group developments (e.g., Cenomi acquisition Sep 2025, BYD KSA launch) Each: event, date, division, significance no

Presentation

Format JSON (structured) 100%

Variation

Completeness Full profile 90% Outdated recent_events 10%

Pairing Strategy

Generation order: News articles first — the article batch defines what content is available. Corporate context is static and paired with every batch. The key test dimension is article selection and implication quality, not context variation.
Pairing Type%DescriptionExpected Output
Rich batch (all buckets covered) 40% Batch contains clear articles for all 5 news buckets, easy selection 5-6 items covering all buckets, strong implications per division
Sparse batch (gaps in buckets) 25% Batch has no articles for 1-2 buckets; agent must still deliver 5-6 quality items 5-6 items with double-coverage on available buckets, noted gaps
Noisy batch (high irrelevance ratio) 20% 40%+ of articles are irrelevant noise; agent must filter aggressively 5-6 items selected from relevant subset only; no filler items
Mixed quality batch 15% Mix of full, truncated, and summary-only articles; agent must work with incomplete data 5-6 items; truncated articles used only if enough context for implication; flagged if uncertain

Output Schema & Derivation

Output 1: Daily News Digest (text)

FieldTypeDescriptionConstraintsRequired
header string Date header Exact format: Daily News | DD Mon YYYY yes
items[] array 5-6 news items Exactly 5-6 items; never fewer, never more yes
items[].item_number int Sequential number 1-6 yes
items[].headline string Bold, concise, action-oriented headline 8-15 words; active verb; includes key actors; enclosed in quotes in output yes
items[].para_news string What happened — summarizes the news from cited sources 60-100 words; 3-4 sentences; attributes to specific publications ("According to Reuters..."); includes specific numbers, names, facts yes
items[].para_context string Market/analyst reaction and broader implications 40-80 words; 2-3 sentences; expert commentary, market movements, global context yes
items[].para_implication string Al-Futtaim-specific implication 30-60 words; 1-3 sentences; MUST start with "Potential implication for the group:"; MUST name at least one specific AF division (automotive, retail, real estate, financial services, healthcare); MUST include actionable recommendation yes
items[].sources[] array<object> Source citations 1-3 sources; each: {name: string, url: string}; name from approved list only yes
items[].bucket enum News category key_markets | economic_financial | competitors_partners | geopolitical | sector_specific yes
footer string Classification footer Exact: Internal Use: Al-Futtaim Group yes

Output 2: Audio Script

FieldTypeDescriptionConstraintsRequired
script_text string Spoken news script for voice synthesis 800-1200 words; news anchor tone; transitions between items ("Moving on to..."); no URLs; no markdown formatting; pronounceable numbers ("three point two percent" not "3.2%") yes
estimated_duration_seconds int Estimated audio duration 120-240 seconds; ~30-40s per item + 10s intro + 10s outro yes
voice_style enum Voice synthesis style directive news_anchor yes

Output Quality Constraints

ConstraintRule
Bucket diversityNo two items from the same bucket unless the batch has 6 items and one bucket dominates the news cycle
Division coverageAt least 3 different AF divisions mentioned across all implications
Source diversityAt least 3 different publications cited across all items
Implication specificityEvery implication names a specific division AND gives an actionable recommendation: a concrete, owner-clear directive (e.g., "negotiate pricing protections with BYD" not "explore opportunities"). Recommendations must map to a named division or business function.
ToneNeutral, analytical, authoritative. Facts-first (no hedging like "may," "could," "might" without evidence); direct source attribution; expert quotes where available; market moves with specificity ("Tadawul rose 1.8%" not "markets reacted positively"). Never sensational, never opinion-driven.
Output formatText digest: Markdown (.md) delivered as WhatsApp message body (demo: email body). Audio script: plain text (.txt) passed to ElevenLabs for voice synthesis. Filenames when attached: daily_news_YYYY-MM-DD.md, daily_news_audio_YYYY-MM-DD.txt
NumbersInclude specific figures (%, $, market share) wherever available in source articles
Audio-text consistencyAudio script must cover all items from the text digest in the same order

Derivation Procedure

1
Filter — From the batch of 20-50 articles, discard irrelevant articles (no connection to Al-Futtaim markets, sectors, competitors, or partners). Discard stale articles (>24h) unless no fresh alternative exists for a bucket. Discard duplicates (same story from multiple sources — keep the most detailed version).
2
Classify — Tag each remaining article with a news bucket (key_markets, economic_financial, competitors_partners, geopolitical, sector_specific) and region tags. An article may fit multiple buckets — assign the primary one.
3
Select — Pick 5-6 articles maximizing bucket diversity (ideally one per bucket + one flex). Prioritize: (a) direct Al-Futtaim brand/partner mentions, (b) key market developments, (c) sector moves with clear group implications. For truncated articles, only select if enough context exists to write a meaningful implication.
4
Assemble constants — Set header to Daily News | DD Mon YYYY (today's date). Set footer to Internal Use: Al-Futtaim Group. These are hardcoded constants, not derived from input.
5
Write digest items — For each selected article, write the 3-paragraph structure: (1) para_news: summarize facts with source attribution and specific numbers; (2) para_context: market reaction, expert commentary, global significance; (3) para_implication: start with "Potential implication for the group:", name specific AF division(s), include actionable recommendation. Generate a punchy headline with active verb.
6
Generate audio script — Convert the 5-6 digest items into a spoken news script. Add intro ("Good morning. Here are today's key developments relevant to the group."), transitions between items ("Moving to regional markets..."), and outro ("That concludes today's briefing."). Convert numbers to spoken form. Remove URLs and formatting. Estimate duration at ~150 words per minute.
7
Validate — Check: (a) 5-6 items present, (b) bucket diversity met, (c) at least 3 AF divisions mentioned, (d) every implication has an actionable recommendation, (e) all sources from approved list, (f) audio script covers all items in same order, (g) total text word count 1000-1500, (h) no opinion or sensational language.

Complexity & Sizing

Medium
Agent Complexity
Low
Surface Area
Flow Check
Purpose
12
Total Samples

Complexity Signals

SignalEvidenceRating
Logic DepthFilter 20-50 articles down to 5-6, rank by relevance, balance across 5 bucketsMed
Output Complexity2 outputs (structured text digest + spoken audio script), strict 3-paragraph template per itemMed
Jobs-to-be-done4 jobs: filter, write, generate implications, generate audioHigh
Domain JudgmentMust understand Al-Futtaim business divisions and map news to specific division implicationsMed

Coverage Distribution

Happy Path 7 (58%)
Edge Case 3 (25%)
Error Case 1 (8%)
Adversarial 1 (8%)

Per-Category Composition

Happy Path (7)
Edge Case (3)
Error Case (1)
Adversarial (1)
Article completeness Full article 100%
Article style Wire-service 57% Analytical 29% Regional 14%
Relevance Clearly relevant 70% Tangentially relevant 30%
Freshness <6h old 85% 6-24h old 15%
Noise Clean 85% Light noise 15%
Batch type Rich (all buckets) 70% Sparse (1 gap) 30%
Bucket coverage key_markets 20% economic_financial 20% competitors_partners 20% geopolitical 20% sector_specific 20%
Article completeness Truncated/paywall 50% Summary only 20% Full 30%
Relevance Tangentially relevant 50% Clearly relevant 30% Irrelevant noise 20%
Freshness 6-24h old 50% >24h stale 30% <6h old 20%
Noise Extraction artifacts 40% Ads/nav removed 40% Clean 20%
Batch type Sparse (2 gaps) 40% Noisy (high irrelevance) 40% Mixed quality 20%
Article completeness Empty/garbled body 50% Duplicate articles 50%
Noise Garbled/broken 100%
Batch type Mostly broken articles 100%
Article completeness Full (polished) 100%
Relevance Appears relevant but misleading 100%
Deception type Clickbait headline 50% Wrong entity / sponsored content 50%
Noise Clean (deception requires polish) 100%

Few-Shot Examples

3 examples
Example 1: Happy Path — Rich Batch, Wire-Service Article (Key Markets)

Input (article batch — showing 1 of 28 articles)

{
  "headline": "Saudi Arabia's PIF reshuffles $925 billion portfolio, pauses NEOM gigaprojects",
  "body": "RIYADH (Reuters) — Saudi Arabia's Public Investment Fund announced a major portfolio restructuring on Monday, pausing several high-profile gigaprojects including key phases of NEOM while redirecting $40 billion toward consumer-facing ventures and tourism infrastructure.\n\nThe sovereign wealth fund, which manages an estimated $925 billion in assets, said it would prioritize investments with shorter payback periods following a strategic review commissioned by Crown Prince Mohammed bin Salman. \"We are recalibrating our timeline to ensure sustainable returns,\" PIF Governor Yasir Al-Rumayyan told reporters at a press conference in Riyadh.\n\nMarket analysts reacted positively to the pivot. Goldman Sachs analyst Farouk Soussa noted that \"PIF's shift toward consumer ventures could unlock significant partnership opportunities for retail and hospitality operators in the Kingdom.\" Saudi Arabia's Tadawul index rose 1.8% on the news, led by gains in consumer and real estate sectors.\n\n...\n\n[Article continues with details on specific projects paused, timeline for consumer venture fund deployment, and quotes from regional economists on impact to GCC construction sector — ~450 words total. Style: wire-service.]",
  "source_publication": "Reuters",
  "source_url": "https://www.reuters.com/business/finance/saudi-pif-reshuffles-portfolio-2026-03-30/",
  "published_at": "2026-03-30T06:15:00Z",
  "region_tags": ["Saudi", "Global"],
  "sector_tags": ["real_estate", "retail", "financial_services"]
}

Expected Output (item 4 of 6 in digest)

{
  "item_number": 4,
  "headline": "Saudi Arabia refocuses its $925 billion Public Investment Fund after gigaproject delays",
  "para_news": "According to Reuters, Saudi Arabia's Public Investment Fund has announced a major portfolio restructuring, pausing several NEOM gigaproject phases while redirecting $40 billion toward consumer-facing ventures and tourism infrastructure. PIF Governor Yasir Al-Rumayyan confirmed the pivot follows a strategic review commissioned by Crown Prince Mohammed bin Salman, emphasizing \"sustainable returns\" over megaproject timelines.",
  "para_context": "Goldman Sachs analyst Farouk Soussa described the shift as unlocking \"significant partnership opportunities for retail and hospitality operators in the Kingdom.\" The Tadawul index rose 1.8% on the news, with consumer and real estate sectors leading gains.",
  "para_implication": "Potential implication for the group: This pivot creates direct opportunities for Al-Futtaim's retail division (IKEA, Cenomi Retail's 500+ KSA stores) to partner on PIF-backed consumer venues. Real estate division should recalibrate timelines for any KSA construction projects tied to NEOM-adjacent developments and explore new tourism infrastructure opportunities.",
  "sources": [
    {"name": "Reuters", "url": "https://www.reuters.com/business/finance/saudi-pif-reshuffles-portfolio-2026-03-30/"}
  ],
  "bucket": "key_markets"
}
Derivation: Happy path — clean wire-service article, clearly relevant to AF's Saudi operations (Cenomi acquisition, real estate division). Filter: passes all checks (fresh, relevant, full text). Classify: key_markets (Saudi PIF). Select: high priority due to direct brand mentions (Cenomi's KSA presence). Write: para_news extracts core facts + attribution; para_context adds analyst reaction + market data; para_implication names two specific divisions (retail, real estate) with actionable recommendations.
Example 2: Edge Case — Truncated Analytical Article (Competitors/Partners)

Input (article — paywalled, truncated)

{
  "headline": "China's EV export strategy shifts as domestic subsidies wind down, threatening margin compression for BYD and rivals",
  "body": "China's State Council is preparing to phase out its remaining electric vehicle purchase subsidies in the next five-year plan, according to three people briefed on the deliberations, a move that would fundamentally alter the competitive dynamics of the world's largest EV market.\n\nThe policy shift, expected to be formalized by mid-2026, would remove the last layer of direct consumer incentives that have helped Chinese EV makers like BYD, NIO, and Xpeng achieve rapid domestic growth. Industry analysts warn this could accelerate an already fierce price war and push manufacturers to rely even more heavily on export markets for margin relief.\n\n[Paywall: Subscribe to WSJ for full article]",
  "source_publication": "WSJ",
  "source_url": "https://www.wsj.com/business/autos/china-ev-subsidy-phaseout-2026-03-30",
  "published_at": "2026-03-30T02:30:00Z",
  "region_tags": ["China", "Global"],
  "sector_tags": ["automotive"]
}

Expected Output (item 5 of 6 in digest)

{
  "item_number": 5,
  "headline": "China plans to phase out EV subsidies, pressuring BYD margins and export strategy",
  "para_news": "The Wall Street Journal reports that China's State Council is preparing to phase out remaining electric vehicle purchase subsidies in the next five-year plan, expected to be formalized by mid-2026. The move would remove direct consumer incentives that have fueled rapid growth for BYD, NIO, and Xpeng in the world's largest EV market.",
  "para_context": "Industry analysts warn the subsidy removal could intensify the domestic price war and push Chinese EV manufacturers toward export markets for margin relief. The shift signals Beijing's confidence in the sector's maturity but creates near-term uncertainty for manufacturers reliant on domestic incentives.",
  "para_implication": "Potential implication for the group: Al-Futtaim's automotive division, which launched BYD in KSA through Al-Futtaim Electric Mobility Company, should proactively negotiate pricing protections. Margin compression in China may lead BYD to push for higher export volumes to the Middle East — an opportunity to secure competitive pricing for the region.",
  "sources": [
    {"name": "Wall Street Journal", "url": "https://www.wsj.com/business/autos/china-ev-subsidy-phaseout-2026-03-30"}
  ],
  "bucket": "competitors_partners"
}
Derivation: Edge case — truncated/paywalled article (only 2 paragraphs visible). Agent has enough context to write a meaningful implication because: (a) headline + first 2 paras establish the core news, (b) BYD is a named AF partner, (c) automotive division implications are clear even without full article. The agent does NOT fabricate details from the paywalled portion — context paragraph uses available information + reasonable inference ("signals Beijing's confidence"). Covers variation dimensions: truncated completeness, analytical style, competitors_partners bucket.
Example 3: Adversarial — Clickbait Headline with Irrelevant Body

Input (deceptive article)

{
  "headline": "Toyota Faces Major Supply Chain Crisis as Key Supplier Declares Bankruptcy",
  "body": "NAGOYA — A tier-3 supplier of specialty fasteners to Toyota Motor Corporation's Tsutsumi plant in Aichi Prefecture has filed for bankruptcy protection, citing declining orders from non-automotive clients. The supplier, Matsuda Metal Works, produced approximately 0.3% of the fastener volume used in Toyota's Japanese domestic production line.\n\nToyota spokesperson Kenji Yamamoto confirmed that the bankruptcy would have \"no material impact\" on production schedules. \"We have multiple qualified suppliers for these components, and our procurement team activated secondary sourcing within hours of the notification,\" Yamamoto said.\n\nAnalysts at Nomura Securities dismissed the incident as routine supply chain management. \"This is a non-event for Toyota's operations,\" said analyst Takeshi Ono. \"Tier-3 supplier rotations happen regularly in the Japanese automotive ecosystem.\"\n\n...\n\n[Article continues with background on Japanese SME bankruptcy trends and government support programs — ~400 words total. Style: wire-service.]",
  "source_publication": "Reuters",
  "source_url": "https://www.reuters.com/business/autos/toyota-supplier-bankruptcy-japan-2026-03-30/",
  "published_at": "2026-03-30T08:45:00Z",
  "region_tags": ["Global"],
  "sector_tags": ["automotive"]
}

Expected Output

EXCLUDED from digest.

Reason: Despite the alarming headline ("Major Supply Chain Crisis"), the article body reveals this is a tier-3 supplier bankruptcy affecting 0.3% of fastener volume at a single Japanese plant, with Toyota confirming "no material impact" and Nomura calling it a "non-event." This has zero operational relevance to Al-Futtaim's Toyota franchise in the Middle East. Including it would waste a slot and undermine briefing credibility.
Derivation: Adversarial — clickbait headline designed to trigger Toyota keyword matching. The agent MUST read the article body, not just the headline. Filter step catches this: (a) Toyota confirms no material impact, (b) affects Japan-only domestic production, (c) 0.3% volume = negligible. The correct action is exclusion, not inclusion with a "minimal impact" implication. This tests the agent's ability to resist headline-driven false relevance.

Anti-Patterns

Generic implicationsImplications that say "this may affect the group's operations" without naming a specific division or giving an actionable recommendation. Why it's wrong: the entire value of this briefing is the AF-specific mapping. Sreeram's manual briefings always name divisions.
Headline-only filteringSelecting or rejecting articles based on headline keywords alone without reading the body. Why it's wrong: adversarial examples have misleading headlines (clickbait Toyota, wrong "Al Futtaim" entity). Real relevance requires body analysis.
Same-bucket clusteringAll 5-6 items from the same bucket (e.g., 5 geopolitical stories). Why it's wrong: the briefing is designed to give Omar a 360-degree view across markets, competitors, and sectors. Bucket diversity is a hard constraint.
Fabricated details from paywalled articlesInventing analyst quotes, specific numbers, or conclusions not present in the truncated article text. Why it's wrong: executive briefings must be factually grounded. If the paywall cuts off key context, the agent should work with what's available or exclude the article.
Audio-text divergenceAudio script that covers different items, reorders items, or adds commentary not in the text digest. Why it's wrong: the text and audio are two formats of the same briefing. Recipients may consume both; inconsistency erodes trust.
Under-target sizingGenerating fewer than 5 items or a digest under 1000 words. Why it's wrong: the briefing must deliver exactly 5-6 items every day. Omar expects a consistent format; a 3-item digest signals a broken pipeline.
Hallucinated sourcesFabricating publication names, URLs, analyst quotes, or statistics not present in the input articles. Why it's wrong: executive briefings must be factually grounded. A single fabricated source destroys credibility with the VC.
Placeholder division namesReferencing "the retail division" or "the group's automotive arm" instead of specific AF brands (IKEA, Cenomi Retail, BYD, Orient Insurance). Why it's wrong: Sreeram's manual briefings name specific entities. Generic labels indicate the agent doesn't understand the corporate structure.
Repetitive recommendationsEvery implication ending with "monitor developments" or "explore opportunities." Why it's wrong: each implication must have a distinct, actionable recommendation tied to the specific news item. Repetition signals the agent is templating rather than reasoning.

Open Questions (2)

Question 1
The sample shows 5 news items. Sreeram mentioned 5-6 in the call. The agent spec says "5-6."
Should the agent always target exactly 5, or flex up to 6 when the news cycle is particularly dense? Current ruleset allows 5-6.
Question 2
The sample shows Sreeram selects news items based on editorial judgment. The confirmed source list has 6 publications + "1-2 others TBD."
Are there additional approved news sources beyond the 6 confirmed ones? The ruleset currently restricts to the 6 confirmed publications only.