Custom Email
Marketing Software

CASE STUDY // EMAIL INFRASTRUCTURE

How Veritly built a fully custom email delivery pipeline with a fine-tuned LLM trained specifically on the brand's tone, catalog, and historical campaign data.

Monthly Send Volume350K

Emails dispatched monthly via a self-hosted SMTP stack, covering promotional, behavioral trigger, and transactional flows.

Inbox Delivery Rate96.8%

Achieved via dedicated IP warm-up, DKIM/DMARC alignment, and active bounce suppression — up from ~79% on shared ESP infrastructure.

Open Rate Uplift+19%

Average open rate improvement across campaigns after switching to fine-tuned subject lines and send-time optimisation.

The Problem

The client was a growing DTC brand paying escalating fees across Klaviyo and a secondary ESP. All automations, customer list segments, and three years of campaign history lived inside vendor databases. Additionally, off-the-shelf AI copy tools produced generic, robotic output that diluted the brand's established tone.

SaaS Dependency Lock-in: Escalating list size and email volume led to exponential platform fees and vendor lock-in risk.

Generic Output: Off-the-shelf AI assistants generated broad marketing copy lacking brand voice specificity.

The Solution

We engineered a fully custom sending infrastructure and fine-tuned a language model on the brand's historical campaigns, customer support chats, tone guidelines, and product catalogs to automate high-performance personalized email campaigns.

Fine-Tuning Loop: The platform continuously retrains the model by feeding campaign performance (CTR/opens) back into the dataset, ensuring the AI aligns closer with audience preferences.
Sending Pipeline

Core Infrastructure Pillars

SMTP sending layer

Direct server integration allowing full control over reputation, bounce retries, and deliverability.

Design Builder

Drag-and-drop template generator producing clean, pre-tested responsive HTML formats.

Native Analytics

Real-time open, click, conversion, and revenue tracking linked straight into the model training pipeline.

The Fine-Tuned Model

TRAINING DETAILS

Rather than using broad prompts on generic models, this LLM is fine-tuned specifically on the client's internal marketing history.

Campaign History // Learns language and layouts that historical data shows drove conversions
DTC Tone Guidelines // Capture tone nuances, prohibited words, and brand vocabulary preferences
Product Catalog // Ingests description texts, sizing charts, and category nomenclature

Adaptive Send Types

AUTOMATED ENGINES

The platform manages a variety of automated trigger sends utilizing the fine-tuned copy generator.

Newsletters // Batch copy creation based on seasonal topics and product inputs
Personalized Welcomes // Trigger emails written for individual sign-up sources
Behavioral Sequences // Dynamically reference browsed products and abandoned carts
Product Roadmap

Phase 2: Predictive Send Automation

Send Time Optimization

Predicting individual subscriber open profiles to automatically schedule delivery times.

Segment-level models

Finetuning specific LLM model variants optimized for high-value buyers versus new subscribers.

AI-assisted hygiene

Monitoring subscriber engagements and automatically triggering re-engagement or suppression paths.

Real-time stock integrations

Mapping ad-hoc inventory changes, price reductions, and restocks directly to email templates.

Get Started

Ready to own your email infrastructure?

Book a session with our technology team to discuss custom integrations, database triggers, and workflow automations.