How I Built DM AI Assist: Lessons From Building an AI Marketing SaaS

The problem: marketing work is scattered

A small business doing its own marketing usually juggles social posts, SEO, paid ads, email, a posting schedule and a set of brand details that live in someone's head. Each of those often means a different tool, a different login and a different way of describing the business.

The result is extra time and inconsistency. The Instagram caption, the ad and the email are written on different days, in different moods, from different descriptions of the same offer.

What DM AI Assist includes

• Social Media Generator — platform-specific posts with variations, hashtags and calls to action.
• SEO Generator — titles, meta descriptions, outlines and content, with optional live search research.
• Google Ads Generator — headlines and descriptions for search campaigns.
• Email Marketing Generator — campaign emails and subject lines.
• Content Calendar — a dated posting plan across platforms.
• Brand Kit — business details stored once and reused by the generators.
• Projects / Client Workspace — organise work by business or client.
• DM Nova — a built-in helper assistant for questions inside the app.
• AI Marketing Assistant — a Pro feature that drafts marketing strategy from a brief.

Why I chose structured inputs instead of only a chatbot

A blank chat box assumes the user knows how to write a good prompt. Most business owners don't want to learn that — they want to say who they are, who they're selling to, what they want to achieve and which platform it's for.

So each generator asks for those things as form fields: business, audience, goal, platform, campaign details. The form does the prompt engineering in the background. It also makes outputs more predictable, because every request includes the context that matters instead of whatever the user remembered to type.

Lessons from building the product

1. AI output needs context

The same model gives very different quality depending on what it's told. A request with a clear audience and goal beats a vague one every time. That's why Brand Kit exists, and why the forms ask for more than a single line.

2. Marketing tools need consistency across channels

Individually good drafts can still make a confusing campaign if they don't agree with each other. Shared business context is the simplest fix I've found. I describe the workflow in how I use DM AI Assist to build a complete campaign.

3. API costs need safeguards

Every generation and every research lookup costs money. A product that calls paid APIs without limits can become expensive quickly, through heavy use or abuse. DM AI Assist uses monthly generation quotas, per-hour limits on research and blocks on repeated identical requests, and expensive actions only run when a user clicks a button.

4. User data isolation matters in a multi-user SaaS

Brand details, generations and client projects are private business information. The account is always taken from the signed-in session — never from something the browser sends — and the database only returns rows that belong to that account. It's invisible when it works, and it's the part I'm least willing to cut corners on.

5. External data helps when it's kept separate from AI interpretation

Adding live search results to the SEO Generator made its suggestions more grounded. But mixing real data with AI guesses can mislead people, so the interface labels what was observed and what the AI inferred. I show this in the SEO Generator example.

6. A feature isn't finished until the whole workflow is tested

More than once, a feature that "worked" in isolation broke somewhere in the real path: a button that only appeared on one result type, an image that generated but didn't download. Now I test the full journey as a signed-in user before calling something done — and when a feature isn't ready, it's marked as coming soon rather than shipped half-working.

7. Human review remains important

AI writes fast and confidently, including when it's wrong. Every output in DM AI Assist is a draft for a person to check. I explain more about that balance in how AI can help digital marketers and the AI marketing assistant guide.

What I'm improving next

I'm exploring deeper connections between the generators, so a campaign can move from one step to the next with less retyping. A future direction is turning generated content into visuals using Brand Kit details, and adding more search metrics to SEO Research. These are directions, not promises or dates.

You can read more about me and DigiMedia on the About page.