← All case studies

CRM for marketing agencies: from the client to the prompt

CRM for marketing agencies: meetings, chats and briefs in 1 timeline per client, and the prompt comes out built for Higgsfield or Magnific. A sample concept.

Year2026
IndustryMarketing agencies and studios
DurationConcept: we built it for ourselves first
What we did
  • CRM for agencies
  • Per-client knowledge base
  • Creative workflows via API
  • Prompts with brand context
Vermú Lindero Meeting · Oct 14

“Less studio, more sidewalk”

Goes into the prompt neighbourhood sidewalk bar
HiggsfieldMagnific
Sample brand and data

The case on video, explained.

Go to the questions ↓

Open the client and you already know everything

This CRM for marketing agencies keeps each client's complete timeline: meetings, chats, emails, briefs, pieces and what they liked or didn't. All of it lives in a base that feeds the creative workflow, in Higgsfield or Magnific via API. When someone asks for a piece, the prompt comes out built with that brand's context, with no need to explain it again.

What the client said never reaches the piece

The client explains it once, in detail, in a meeting. Then the account lead sums it up over WhatsApp and whoever produces it types it into the tool from memory. Something falls away at each step.

In the meeting

“Less studio, more sidewalk, at 7 pm. People aged 25 to 45, no little umbrellas, and let amber lead.”

  • Sidewalk
  • 7 pm
  • No umbrellas
  • Aged 25 to 45
  • Amber
In the account lead's WhatsApp

They want something summery, on the street, young people. I'll send you the logo.

  • Sidewalk
  • 7 pm
  • No umbrellas
  • Aged 25 to 45
  • Amber
In the prompt, from memory
Vermouth bottle, summer vibes, fresh, professional photo
  • Sidewalk
  • 7 pm
  • No umbrellas
  • Aged 25 to 45
  • Amber
And the piece comes back “This isn't it. I already told you sidewalk.”
Vermú Lindero · sample data

The CRM: the whole story in one column

This is the screen for whoever handles the client. Before a meeting, or when the client writes on a Saturday, they open the record and see what was promised, what was approved and what they didn't like. If the account changes hands, the story stays. Tap each item to see what it says.

Estudio Bruma · CRM
Reason for rejection

“Too nighttime. We agreed on the 7 pm sun.”

Goes to the base: afternoon light, never night
  • Before a meetingYou read what's pending in a minute.
  • When the account changes handsThe new person doesn't ask the client anything again.
  • When something goes wrongYou can see what was asked, who approved it and when.
Estudio Bruma and Vermú Lindero · sample data

The base: transcripts and briefs, turned into data

A transcript alone is no use to anyone: it's forty minutes of talk. Each meeting and each brief yields the facts that matter for production, and each one points to the sentence it came from. If the brand changes its mind, the fact is updated and you can see what it said before. Tap a field to see its sentence.

Oct 14 meeting

We're tired of the studio thing. Less studio, more sidewalk, like when you bring the little table outside. And at 7 pm, with that sun, not at night. The person who drinks this is 25 to 45, it's not a nightclub. And for colour, let the bottle's amber lead.

  • Also in the base
  • Chats
  • Emails
  • Brand manual
  • Approved and rejected pieces
Vermú Lindero base
Sample data

The prompt comes out with the brand built in

Same request, with and without the client's base. Without it, you get a bottle in summer like any brand's. With it, the prompt already knows the scene, the light, what's out and what got rejected last week. And each part says where it came from, so it can be corrected in a moment.

Request: “summer feed for Lindero”
Vermouth bottle, summer vibes, fresh and refreshing, professional product photo, high quality.
  1. SUBJECT Amber vermouth bottle, glass on ice with lemon peel, metal bar table.Oct 21 brief
  2. SCENE Neighbourhood sidewalk bar, table set outside.Oct 14 meeting
  3. LIGHT Warm golden hour, 7 pm. Not night.Rejected Nov 3
  4. PEOPLE Adults 25 to 45, natural, out of focus.Oct 14 meeting
  5. PALETTE Amber, bottle green, sun-bleached cream.Sep 2 manual
  6. AVOID Cocktail umbrellas, studio backdrop, neon, nightclub.Oct 3 WhatsApp
  7. FORMAT 4:5, clear space at the bottom for the legal line.Oct 29 email
What comes out · illustration

Any drinks brand could use this.

This only works for Lindero, and nobody had to explain the brand again.

Vermú Lindero · sample data

The workflow: Higgsfield or Magnific, via API

Nobody copies and pastes. The workflow takes the request, adds that client's base and builds the call for whichever tool the agency uses. Change the tool and the format: how it's asked changes, and the brand's context stays the same.

  1. Request “Summer feed”
  2. Client base Picks what's useful
  3. Built prompt With its source
  4. Higgsfield Via API
  5. Candidates To approve
The request sent via API 
POST higgsfield /generate
{
  "model": "gpt_image_2",
  "aspect_ratio": "4:5",
  "references": ["manual_lindero.pdf", "feed_v1_aprobado.jpg"],
  "prompt": "Amber vermouth bottle on a metal bar table, neighbourhood sidewalk, warm golden hour, not night. Avoid cocktail umbrellas, studio, neon. Clear space at the bottom for the legal line."
}

The request builds itself. A person reads it and decides to generate.

Sample request and models

What the person decides goes back to the base

Generating spends credits and a bad piece reaches the client. So what's generated stays a candidate until someone at the agency approves it. Their decision stays on the timeline, with name and date, and the next prompt already knows it. See both outcomes.

Rejected
Feed v2
What the decision-maker says

“Too nighttime. We agreed on the 7 pm sun.”

  1. Timeline Feed v2 rejected, with the reason, the date and the prompt that made it
  2. Base Light: never at night
  3. Next prompt It already comes with that line LIGHT Warm golden hour, 7 pm. Not night.
Vermú Lindero · sample illustrations and data

You know why each piece comes back

Since each rejection keeps its reason, the agency stops correcting blindly. You can see what bothers each client, and that gets checked before generating and before sending them anything.

Why their pieces came back
  • Light and time of day45 %
  • Scene27 %
  • Legal line18 %
  • Logo10 %

With Lindero, nearly half the rejections are about light. That now goes in every prompt.

Estudio Bruma · sample clients and data

We built it for ourselves first

Mithandir works this way with its own clients and brand: each company's timeline is in our CRM, meetings get transcribed, and our back office requests images from Magnific via API with a prompt built from the brand's method. Each image keeps the model and prompt that made it, and goes through approval.

  1. 01The timelineEach company, with its story in the CRM.
  2. 02The meetingsTranscribed and stored by client.
  3. 03The methodBuilds the prompt from what the brand knows.
  4. 04The imageComes out via API, with its origin, to approve.

Answer and we’ll tell you if this works for your company.

We review it quickly. If it’s a fit, we set up a meeting.

Assessment

5 questions, one minute. With this we come to the call already knowing how your team works with its clients.

What best describes your team?
How many people work on accounts and production?
Where is the most time lost between the client and the piece?

Pick up to 3.

Pick up to 3.

Where does what each client says end up today?
When would you like this solved?
Last step: who should we write to?

We only use it to reply about this assessment.

Other systems in production