Client case study generator (n8n + LLM)

Grow2.ai builds an AI automation that closes the case study production cycle — from collecting raw material on a completed project to a ready publication draft. One account manager starts the process, receives a draft in six hours, and hands it to an editor for final review.

What it does in the production cycle:

1. Collects project artifacts: client call transcripts, email correspondence, briefs, result metrics, links to final materials in the CMS and file storage.
2. Summarizes long sources — hours of recorded conversations, dozens of emails, report drafts — into compact structured summaries with a focus on figures, decisions, and direct client quotes.
3. Extracts key facts for the case: the client's initial pain point, exactly what the team did, quantitative results (conversions, revenue, time), client quotes with attribution.
4. Generates a structured draft following the problem → approach → results → client quote template, 600–1200 words depending on the preset.
5. Publishes the draft to the CMS with the status "on review" and attached source links for every fact used — the editor checks context in one click.
6. Notifies the editor in the work channel with a link to the draft, a list of sources, and a checklist of what requires manual validation before publication.

What remains for the marketer:

The editor checks the factual accuracy, adjusts the tone of voice for the brand, selects visuals, aligns the final version and quotes with the client, and publishes. The LLM provides the structure and draft text — not the final material ready for release.

What the solution does not do:

• Does not publish a case study without human review. Client quotes, figures, and project names always go through manual verification — responsibility for factual accuracy remains with the editor.
• Does not collect material automatically from the CRM if project artifacts are not structured. Input completeness is the account manager's responsibility: no transcript — no case.
• Does not adapt the style to a specific brand without prompt configuration. Generation follows a general template; fine brand-level work is the editor's task at the final stage.

The architecture is built on a workflow engine as the orchestrator and an external LLM for generation and summarization. The workflow engine collects data from the CMS and file storage, splits it into chunks, runs it through several LLM steps, and consolidates the result into a draft.

Technical flow:

1. Trigger. The account manager fills out a form in the CMS or sends a command in the work chat. Specifies the project ID, client name, links to key artifacts — or relies on auto-collection by tags in storage.
2. Artifact collection. The workflow engine pulls via API: meeting transcripts from the transcription service, email threads from the email system or CRM, documents from the file storage, metrics from analytics systems.
3. Summarization long → short. Each long source — a 60–90 minute transcript, an email thread of 40 messages — goes through the LLM with an extraction prompt: what happened, what decisions were made, what numbers came up, which client quotes are worth keeping. The output is a structured summary of 200–400 words.
4. Fact extraction. A separate LLM step extracts canonical facts: before/after metrics, project timeline, team composition, technologies used, direct client quotes with attribution.
5. Draft generation. A combined prompt receives all summaries and structured facts, generates a case study using the template: client context → problem → approach → results → quote → next steps.
6. Publishing the draft to the CMS. The workflow engine creates a record in the CMS via API, attaches source links as internal references, sets the status to "on review", and tags the editor.
7. Notification. A webhook sends a message to the work channel with a direct link to the draft and a list of what requires manual review: quotes, figures, names.

Key components:

Implementation stages:

1. Week 1 — inventory. Grow2.ai inventories the sources of case artifacts: where transcripts are stored, how they are linked to the project in the CRM or PM system, which metrics are available via API. Aligns the case study template with marketing.
2. Week 2 — workflow build. A pipeline is deployed in the workflow engine: trigger → collection → summarization → generation → publication. Prompts are written to match the agency's style and structure.
3. Week 3 — calibration. Generation is run on 3–5 completed projects, marketing compares the draft against the manual version, prompts are refined for tone, wording, and length.

The solution runs as a self-hosted workflow engine on the agency's own infrastructure or as a managed instance. For agencies with sensitive client data, Grow2.ai recommends the self-hosted option with LLM via an API provider with an enterprise NDA.

Before launch, Grow2.ai collects a checklist of access credentials and data without which automation fails in production.

Data and sources:

• Client call transcripts in a structured format — via a transcription service or a built-in recorder in the conferencing system, with access via API or export.
• A structured repository of project artifacts: file storage with a consistent folder structure or project tagging in a PM system.
• A CMS with an API for creating draft entries and managing publication statuses.
• Access to outcome metrics via analytics, CRM, dashboards — digital confirmation of the project's impact.
• An archive of 5–10 previous case studies in final form — a reference for the prompt and tone calibration.

Access and subscriptions:

• API keys: workflow engine (self-hosted or cloud), LLM provider, CMS, file storage, transcription service.
• Sandbox CMS for testing publication without risk of breaking production.

Team readiness:

• The account manager understands which project artifacts go into the input — without this, the collection fails.
• The editor takes on a new role: not writing from scratch, but validating and polishing the AI draft.
• The marketing lead approves the case study template and quality criteria for the draft.

Timeline:

The base implementation takes 1–3 weeks. Week 1 — source inventory and template alignment. Week 2 — building the workflow in the workflow engine, writing prompts. Week 3 — calibration on 3–5 historical cases, handoff to the team with brief documentation and monitoring of the first two weeks in production.