WHY I'M BUILDING QUIVER

ASHWIN VISWATMULA · APRIL 23, 2026 · 2 min read

I’ve spent the last few years inside the marketing analytics stack at a large financial services company. The work was fulfilling, the impact was real, and the tools were — with no exaggeration — held together with tape.

This is the thesis behind Quiver. It’s the short version; I’ll expand each piece later.

The gap I kept running into

Enterprise marketing at a financial institution looks nothing like what the marketing-automation category was designed for. A single campaign involves:

Sequential waterfall targeting against millions of customer records, using propensity-model scores from multiple models and sub-models, not demographic filters.
Multi-incentive allocation where different segments get different offers, driven by model thresholds, with strict mutual exclusivity.
Multi-week scheduling with parent drops, trigger-based sends, catch-all weeks, and per-week resting rules.
Two layers of compliance review — creative review of copy and subject lines, model review of the targeting logic — both of which are currently emailed as documents.
Cross-functional handoffs between marketing managers and analytics teams with explicit “mark complete” gates between steps.

Salesforce Marketing Cloud, Braze, HubSpot — none of them support this natively. Teams end up with spreadsheets for audiences, email threads for approvals, SQL runbooks for analytics, and vendor UIs for deployment. The state of a campaign lives in six places at once.

The bet

My bet is that the data warehouse is the right anchor point for this workflow, not a marketing cloud.

The audience already lives there. The model scores already live there. The account data, the tenure data, the spend data, the prior-campaign resting history — all there. The moment you anchor orchestration on the warehouse, everything gets simpler: segmentation is SQL, allocation is a join, resting is a window function, and the compliance artifact is just the query.

The longer-term version of this is even more interesting: orchestration as signal emission. Instead of firing channel-specific actions — “send this email via SendGrid” — journey nodes write structured signal rows into BigQuery tables. The email sender is a table consumer. So is the site personalization engine. So is the direct-mail vendor. Adding a new channel is adding a new consumer, not modifying the engine.

This makes the orchestration platform channel-agnostic by construction. It also makes the warehouse the system of record for what the company said to whom and when, which turns out to be the most important table you can have when you’re trying to do attribution, measurement, or auditing later.

Why now

Three things are converging:

Warehouse-native everything. The rest of the stack (dbt, Hightouch, reverse-ETL in general) already moved here. Orchestration is one of the last holdouts.
The compliance wall is rising. Model-risk management, CFPB, OCC, GDPR — the documentation burden is not getting lighter. Anything that makes the audit trail automatic is worth a lot.
Agents are coming for marketing ops. The campaign setup work is exactly the kind of multi-step configuration an agent should do, but only if the substrate is structured enough to reason about. A warehouse-native, signal-emitting platform is that substrate.