The Context Graph Every Workflow Reads From.
ACE, the Adeqo Context Engine, joins the records from each connected system into one graph per account. Workflows and agents read from that graph.
- Paid platformsAds · Meta · TikTok
- Analytics & searchGA4 · GSC · GTM
- CRM & commerceHubSpot · Shopify
- WarehouseBigQuery · Looker
- AI answer surfacesChatGPT · Perplexity
- Validations
- Recommendations
- Reports
- AI agents
A launch gate, an anomaly check, and a weekly report all resolve the same campaign to the same node.
How the Graph Connects.
The chips above are the node types. These are the edges between them.
01A Campaign Contains Ad Sets, an Ad Set Contains Creatives
That containment is the platforms’ own, and a keyword sits at the same depth as a creative.
02A Landing Page Emits a Conversion Event
The event belongs to the page that fired it, which is why a /quote thank-you firing GA4 and not Google Ads reads as one edge present on one side only.
03A Conversion Event Joins a CRM Record
The join runs through the source field the form wrote, so when HubSpot dropped paid social UTMs on /book and /quote, an estimated 11% of paid social leads were mis-attributed.
04A Tracking Check Sits on the Join It Scores
Page to event, event to CRM record, platform to conversion import, domain to domain. The gate reads each as pass, warning or critical.
05A Business Priority Orders the Work
It ranks what the graph produces, so a list of open issues arrives as a recommendation queue and an entry in the AM brief.
When Two Systems Disagree.
Google Ads, GA4, HubSpot, and Shopify each report the same week differently. These rules decide which count a recommendation is written against.
01The Taxonomy Decides Which Campaign
A record attaches to a campaign on the account’s naming and UTM schema. That schema is the join key, whatever each platform files the channel under.
02Commercial Outcome Outranks the Platform’s Count
When Meta’s reported conversions and the pipeline in Salesforce point different ways, the recommendation is written against pipeline, qualified leads and revenue.
03Measurement Is Questioned Before Performance
The anomaly workflow asks whether a movement is downstream of tracking before it asks whether it is downstream of the platform, which is how the 22% CAC rise on the sample account was traced to a dedup fault.
04A Failed Threshold Produces a Recommendation
CAPI dedup at 14% fails an 8% check, and the result is a spend-hold recommendation with a named owner, who decides whether to take it.
05Market Overrides Are Signed
Per-market and per-vertical exceptions are approved and logged, and an override that opens a held launch gate requires AM sign-off.
Deterministic Before Generative.
A decision belongs to the rule engine when it can be written as a predicate over stored fields. Everything else is reasoned, and arrives as a proposal a person can accept or reject.
- Naming and taxonomy
- UTM schema
- Launch QA gates
- Tracking thresholds
- Approvals
- Cross-channel interpretation
- Anomaly explanation
- Recommendations
- Report narrative
- Workflow coordination
See What ACE Would Hold for Your Site.
The sample audit runs a subset of these checks and returns the result.