Scaling On-Brand Marketing with Generative AI on Amazon Bedrock
A multi-tenant marketing platform rebuilt content production as a generative AI pipeline on Amazon Bedrock on Amazon Web Services (AWS), grounded in each customer's own material, checked by a guardrail and approved by a person before anything goes out.
The problem.
Our client's customers need marketing that is on-brand, based on facts and safe to publish, without a marketing team of their own. Written by hand, each approved post took a marketer about 2.6 hours and roughly US$5.20 of their time.
Generic AI copy is cheaper but brings a new risk: one post with an invented statistic or a promise of guaranteed returns can expose a business to legal and reputational harm. And the platform ran on a single server, so one failure stopped every customer at once.
What we built.
A pipeline of steps, not one prompt
We rebuilt content production as a generative AI pipeline on Amazon Bedrock. Claude Sonnet 4.5 drafts, critiques and gives the compliance verdict, Amazon Nova Lite writes briefs and image prompts, and Stable Image Core creates the images. Matching each task to the right model keeps quality high where it matters and cost low everywhere else.
Grounded in the customer's own words
Product documents are embedded with Amazon Titan Text Embeddings and stored in PostgreSQL with pgvector, with row-level security keeping every customer's data separate. Every draft is written from that material, not from the model's imagination.
Nothing goes out unchecked
An Amazon Bedrock guardrail checks every brief and draft for harmful content, unsupported claims and personal data. A block stops the post for human review, an error holds it, and nothing is published until a named approver signs it off.
Built to grow
The platform runs on Amazon ECS on AWS Fargate with auto scaling, behind Amazon CloudFront and AWS WAF, with Amazon RDS and Amazon ElastiCache. Network ACLs and security groups guard every subnet, and we run it as a managed service.
AWS services in this engagement
The results.
Measured in production over three weeks with three tenants, an approved post costs about US$0.91 instead of US$5.20, 83% less, and takes a marketer about 0.4 hours instead of 2.6. Each marketer's output rose from 9 to 24 approved posts a week. More than 15 businesses use the platform and have run more than 50 campaigns. Every post is approved by a person, and since 19 October every brief and draft is checked by the guardrail.
What we learned.
Our first production release did not enforce every guardrail policy on every path. Human approval still stopped anything unsafe from going out, but we now test guardrail enforcement as a blocking step in the release checklist, and the pipeline fails closed if the guardrail cannot be reached. We also learned to keep each model step narrow: a planner, a drafter and a checker are easier to test and cheaper to run than one model doing everything.
About Manthic.
Manthic Technologies is an AWS Advanced Tier Services Partner based in Lagos, Nigeria. We design, build and run generative AI products on AWS for our clients.
Tell us what you're trying to ship.
A 30-minute scoping call with the engineers who would do the work.