LIVE|CORTEX|v.1

Digital Marketing AI

Optimization Engine

Cortex is your decision engine, accelerating conversions and lowering acquisition costs across $500M+ in managed marketing spend. Real-time monitoring across every channel surfaces optimization opportunities the moment they emerge. Recommendations grounded in every platform’s primary documentation.

Learn
What Cortex is

AI Powered, Expertise Refined

AI alone moves at the speed of computation but misses what only experience teaches. Expertise alone teaches what computation cannot but cannot work at the speed the data demands. Cortex is the pairing - every recommendation analyzed at machine speed and signed off by an expert before it ships.

01 / Platform Knowledge

Unified Data Retrieval

For every decision, Cortex queries every relevant platform in parallel - Google Ads, GA4, Search Console, Amazon, Meta, Merchant Center, AI engines. Aggregated and reconciled before any action ships.

02 / Real Expertise

Collective Learnings

Pattern recognition across hundreds of client accounts and thousands of historical tests. Every winning headline, audience split, and bidding strategy proven elsewhere becomes a starting point for yours.

03 / A.R.I. Framework

Structured Decision Protocol

28 documented decision classes, each with its own statistical thresholds, confidence requirements, and escalation rules. Auditable. Repeatable. Every action ships with its reason and expected impact attached.

What Cortex runs

Eight disciplines, one engine.

Cortex operates every paid, owned, and earned surface that drives qualified search traffic — feeding a single learning loop instead of eight disconnected playbooks.

Google Ads

Search, PMax, Demand Gen

Amazon Ads

Sponsored, DSP, Marketplace

Bing Ads

Microsoft Search & Audience

Meta Ads

Facebook & Instagram

Email Marketing

Lifecycle, broadcast, automation

SEO

Technical, content, authority

GEO

Generative engine visibility

AEO

Answer engine optimization

Primary Sources

Every platform. 1 unified corpus.

Before Cortex answers anything about an ad platform, it runs a hybrid keyword + semantic search across 846,351 passages of primary source documentation. The model only writes claims it can cite back to a URL.

Hybrid retrieval, end to end

Every query is embedded with text-embedding-3-large (3072d), fused with BM25 lexical ranking via reciprocal rank fusion, then reranked with Cohere rerank-v3.5 to surface the most relevant primary sources.

Documents: 100,767 across 124 sources
Passages: 846,351 embedded
Index: halfvec(3072) HNSW
Rerank: Cohere rerank-v3.5
Latency: 38 ms p50
Refresh: Monthly diff-tracked
cortex_search
> corpus_search("Performance Max minimum spend", k=3)
// hybrid BM25 + halfvec(3072) cosine + RRF + dedup
[
{
"platform": "google",
"title": "Create a Performance Max Budget",
"url": "developers.google.com/google-ads/api/performance-max/create-budget",
"snippet": "average daily budget of at least 3x your CPA..."
},
{
"platform": "google",
"title": "Step 5: Set your budget",
"url": "support.google.com/google-ads/answer/15864534"
},
{
"platform": "google",
"title": "PMax for store goals - optimization tips",
"url": "support.google.com/google-ads/answer/13775728"
}
]
> 3 results · 41 ms · grounded answer drafted
The rules layer

A synthesized layer above the corpus.

Cortex doesn’t just retrieve docs. It extracts the rules that actually run paid search and SEO. Each one tagged, scored for confidence, and cited back to its primary source.

28 / Distilled rules
28

Each rule is grounded in a specific paragraph of primary source documentation. Confidence is scored against the strength of that grounding.

11Conditional rules
12Patterns
4Benchmarks
1Antipattern
BenchmarkConfidence 0.81

LCP above 2.5s degrades Core Web Vitals ranking eligibility and correlates with measurable organic traffic decline.

Source: developer.chrome.com/.../core-web-vitals/lcp
ConditionalConfidence 0.78

Target ROAS bidding requires at least 30 conversions in 30 days. Below threshold, switch to Maximize Conversion Value to unlock delivery.

Source: developers.google.com/.../smart-bidding
AntipatternConfidence 0.74

Broad match keywords paired with manual CPC bidding produce irrelevant traffic at scale. Broad match should only be used with smart bidding.

Source: support.google.com/.../broad-match-best-practices
Grounded reasoning

Every recommendation cites its sources.

Cortex does not guess. Every claim it makes traces back to a specific paragraph in a specific canonical source. Google’s own docs. Meta’s. Shopify’s. Amazon’s. The Search Quality Rater Guidelines. Recommendations that survive a 3 AM audit.

#1247A.R.I. - Acquisition | PMax
Action: switch bid strategy from Target ROAS to Max Conversion Value
Reason: Failing Data Sufficiency. 18 conversions L30 against Google’s 30-in-30 rulecite for sustained Target ROAS bidding.
Impact: unlock delivery so the model can accumulate signal toward re-applying a target.
Citations: developers.google.com/.../smart-bidding · support.google.com/google-ads/answer/13954221
100,767
documents
846K
passages indexed
124
sources
< 100ms
retrieval latency
The whole canon

Every channel that matters, distilled.

Canonical platform documentation across paid ads, SEO, ecommerce, analytics, and AI search. Chunked, embedded, and tier-weighted so retrieval surfaces official sources first.

tier-weighted official · partner · curated
Google Ads
Meta Ads
TikTok Ads
Amazon Ads
Microsoft Ads
ChatGPT Ads
LinkedIn Ads
Pinterest Ads
Snapchat Ads
Reddit Ads
Apple Search
DV360
The Trade Desk
Shopify
WooCommerce
BigCommerce
Webflow
Klaviyo
Mailchimp
Postscript
HubSpot
GA4
Search Console
Merchant Center
GTM
Conversion API
Triple Whale
Northbeam
Schema.org
Search Central
Quality Rater Gd.
web.dev
Lighthouse
Cloudflare
llms.txt spec
OpenAI bots
Anthropic bots
Perplexity
Wikipedia
Wikidata
W3C / MDN
+ 1 more
How teams query Cortex

Real questions. Cited answers. One tap to act.

Three sample conversations - each pulling the client’s live account data, applying the corpus, and ending in a one-tap action rather than just an answer.

Paid spend leaking into organic territoryLive
Structured data audit with sourced citationLive
AI search citation auditLive
How Cortex works

A continuous Ingest > Analyze > Optimize loop.

Retrieve, analyze, decide, act, measure. Monitoring performance across Google Ads, Amazon Ads, Meta Ads, Microsoft Ads, ChatGPT Ads, GA4, Search Console, and Merchant Center.

Step 01Retrieve

Pull live data from every relevant platform.

For a Google Ads bid adjustment, that means campaign performance, search terms, auction insights, GA4 conversion data, Search Console query data, and Merchant Center feed status.

Step 02Analyze

Distinguish signal from noise.

Cortex evaluates the retrieved data against the corpus of platform documentation and historical baselines. Statistical significance is checked. Confounding variables are isolated.

Step 03Decide

Match data against the protocol.

Each decision class has defined criteria: thresholds, lookback windows, confidence requirements, guardrails. If the data clears the criteria, the action is approved.

Step 04Act

Execute on the platform.

Approved actions run directly via API. Budget shifts, bid changes, negative keyword additions, audience adjustments, creative pauses. Higher-stakes actions route to the team for sign-off.

Step 05Measure

Track every outcome against intent.

Every action is measured against the outcome it was intended to produce. Decisions that produced lift get reinforced. Decisions that did not get reviewed and refined.

The system, live

Every signal. Every decision. Every action.

Sentries surface every signal in your account. Cortex evaluates each one against the documented playbook. Only the ones that clear the criteria become Optimizer actions. Below is what that looks like, in real time.

cortex_system · live traceLIVE
Sentry · Signals- / sec
Cortex · Evaluation- / sec
Optimizer · Actions- / sec
Signals surfaced
0
Sentry events
Evaluated
0
Cortex passes
Actions taken
0
Optimizer runs
Why Cortex

Every test. Every pattern. Yours.

An intelligence layer running across every account in our portfolio. The wins, the warnings, and the patterns all flow into yours.

01

Inherit Every Test We've Already Run

Every winning headline, audience, and bidding strategy proven in another account becomes a starting point for yours. Every emerging pattern (CPC spikes, seasonal shifts, creative fatigue) Cortex spots across the portfolio gets surfaced to your account before it costs you.

02

Decisions in Hours, Not Weeks

Cortex runs the Retrieve, Analyze, Decide, Act, Measure loop 24/7. Opportunities get acted on the same day they appear, not at the next account review.

03

Smarter Every Quarter You're Here

Every test, season, and account expands Cortex's corpus. Year two in your account is materially sharper than year one, and year three sharper still.

04

Your Account's Memory Doesn't Churn

If your strategist moves on, every test, hypothesis, and learning Cortex has gathered about your account stays in place and keeps compounding for the next one.

05

See Competitor Moves The Day They Ship

Cortex watches competitor SERP movement, ad creative changes, landing page tests, and offer shifts. You hear about a new play the day it ships, not next quarter.

The loop runs continuously

Every account is monitored every day.
Every decision is documented.
Every outcome is measured.

This is Cortex. Your decision engine behind every Capconvert optimization.