Processing Modes

COMPUTED

AI CDP calculates the value from raw data.

-- Home location from cell tower usage
SELECT dtx_id, cell_id AS home_cell_id
FROM cell_events
WHERE HOUR(timestamp) BETWEEN 20 AND 8
GROUP BY dtx_id
HAVING COUNT(*) = MAX(COUNT(*)) OVER (PARTITION BY dtx_id)

PASSTHROUGH

Source value mapped directly — no transformation.

attribute: account.arpu
mode: PASSTHROUGH
source: billing.customer_arpu
mapping: direct

tip

BSS is source of truth — don't replicate billing calculations.

CUSTOM

Tenant provides own logic in SQL/Python, version-controlled with approval.

def compute_engagement(profile):
    weights = {'app_sessions': 0.35, 'web_visits': 0.25,
               'support_contacts': 0.20, 'social_engagement': 0.20}
    score = sum(profile.get(k, 0) * v for k, v in weights.items())
    return min(100, max(0, score))

Choosing the Right Mode

Scenario	Mode	Rationale
Behavioral trait from CDRs	COMPUTED	CDP has raw events
ARPU from billing	PASSTHROUGH	Billing is authoritative
Churn score from ML	COMPUTED	CDP runs the model
Account status from CRM	PASSTHROUGH	CRM is authoritative
Custom loyalty tier	CUSTOM	Tenant-specific rules