An AI tool I designed for IDFC FIRST Bank’s marketing team. Fill in a brief, get on-brand copy for every channel, get it approved.
The team wrote copy by hand for every channel, using old, scattered reference docs.
They needed much more copy, much faster. But in a bank, nothing ships without sign-off.
The user was a marketer on a deadline, not a customer. The goal: fewer decisions, less rework.
Templates were fast but generic. Manual review kept copy on-brand but was slow.
A tool that fixed only one of these wouldn’t get used.
Take a marketer from a brief to approved, on-brand copy for every channel, with as little rework as possible.
That split the work in two: get the input right, so the copy comes out on-brand, and make the output easy to approve.
Writing the first draft is the easy part.
Editing, approving and tracking it is the real work.
Most AI tools stopped at the first draft.
I designed for everything after it: editing, feedback, approval and version history.
A blank prompt box leads to off-brand copy, and marketers shouldn’t have to learn prompting.
A structured brief: product, goal, tone, theme, audience, channel, key message, number of variants.
Copy comes out on-brand by default, not after corrections.
WhatsApp, email, ads and SMS each have their own rules. The same copy pasted everywhere reads wrong.
Pick a channel and the copy changes shape: short with emoji for WhatsApp, subject and body for email, headline and description for ads.
Brief once, get copy that fits every channel. No style guide to memorise.
Marketers always tweak the first draft. A separate editor for every change slows them down.
Each variant is a card you can edit directly. Clear colours show which card is selected or being edited.
From first draft to final copy, marketers never leave the screen.
Every card needed copy, edit, favourite, feedback and delete. Putting them all on each card buried the copy.
One toolbar above the cards. Nothing selected: sort, select all, generate more. Card selected: actions for that card.
Cleaner cards, no extra screens. The decision I’m proudest of.
“Try again” tells the AI nothing, and marketers shouldn’t have to rewrite prompts.
Pick what went wrong (quality, relevance, consistency or format), add a note if needed, and regenerate.
Faster, more useful revisions. The bank also learns what its team means by “on-brand”.
In a bank, everything needs sign-off. Approvals over email and spreadsheets would undo all the speed.
The marketer sends a variant for approval. An approver accepts or rejects it in the tool, and every decision is logged.
This is what made AI usable for a bank at all, without slowing anyone down.
After several rounds of changes, good drafts got lost and versions were hard to compare.
A tab for each version, showing when it was made and how many variants it has. One click to switch.
Nothing gets lost, so marketers feel free to try more ideas.
Built on Material 3 using IDFC’s red and its tints. Set in Inter, which stays readable in a tool that’s mostly text.