CreatiSoul
Agency OS · Design
AI Integration Map

Where AI plugs into the journey

CreatiSoul Agency OS treats AI as a layer that runs across the whole customer journey, not a single feature. This map places every realistic AI use-case on the stage where it earns its keep — from capturing a lead to expanding a customer — and tags each one with how soon we can ship it (Near-term vs Later) and the capability that powers it (Claude, MCP tools, the existing platform creative engine, or CRM data). The point is to be honest about what's a quick win vs. what needs more data and trust first.

Most of this map is PROPOSED — confirm the priorities

One capability is already real: the platform's brief → AI video / image engine at app.csoul.cloud (campaigns, briefs, Higgsfield generation) — the CRM links to it, never duplicates it. Everything else here is a proposed plan: the Near-term / Later split, which use-cases come first, and how aggressive auto-actions should be are all starting points for the owner to rank and validate. Treat the horizon chips as a recommendation, not a commitment.

How to read this map

Horizon — how soon
Near-termBuildable on what exists now — Claude API + the data the CRM already holds. Low risk, ships in the first AI passes.
LaterNeeds accumulated CRM history, scoring models, or deeper integrations / human trust before it's safe to rely on.
Enabling capability
ClaudeReasoning / writing / summarizing. The default brain for drafting, analysis, NBA.
MCPTools / web / external reach — enrichment lookups, web research, channel actions via MCP servers.
EngineThe live platform creative engine (briefs → AI video / image). CRM hands off, doesn't rebuild.
CRM dataPatterns over the CRM's own records — deals, activities, invoices, engagement health.
Capture / Qualify
3 use-cases
Nurture / Close
3 use-cases
Deliver
4 use-cases
Bill / Collect
2 use-cases
Retain / Expand
3 use-cases
Cross-cutting
3 use-cases
Stage-by-stage AI use-cases
Capture / Qualify
A lead lands from a website form or manual entry. AI turns a raw name + email into a researched, scored, routed prospect — before a human has to read it.
2 Near-term1 Later
Lead enrichment
Near-term

From an inbound email / domain, fill in the Company (industry, size, location, likely GST status) and find the right service lines to pitch. Pre-populates fields a rep would otherwise google by hand.

Powered by Claude MCP web Company
Lead scoring
Near-term

A hot / warm / cold score + one-line rationale from form content, budget signals and fit. Starts rules + Claude judgement; sharpens into a learned model once we have won/lost history.

Powered by Claude CRM data Lead
Auto-triage & routing
Later

Dedup against existing Companies, classify intent (new business vs. support vs. spam), and suggest the owner + pipeline. Suggest-first; only flips to auto-assign once routing accuracy is trusted.

Powered by Claude CRM data Lead Pipeline
Nurture / Close
The rep is working an open Deal. AI does the writing, the call-notes and the "what should I do next" thinking — so the rep spends time talking, not typing.
2 Near-term1 Later
Draft outreach, replies & proposals
Near-term

First-touch emails, reply drafts and scoped proposals built from the Deal's service line, budget and the account's history — in CreatiSoul's voice. Human edits and sends; AI never sends blind.

Powered by Claude CRM data Deal Activity
Summarize calls & emails → Activities
Near-term

Paste a call note or email thread; AI writes a clean Activity on the timeline plus a next-step checklist. Turns messy notes into a structured, searchable account history.

Powered by Claude Activity
Deal-risk & next-best-action
Later

Flags deals that have gone quiet, stalled in a stage, or skipped a step — and suggests the next best action (call, re-quote, drop). Needs enough closed-deal patterns to be trustworthy.

Powered by Claude CRM data Deal Stage
Deliver
The account is a customer with running Engagements across Website, Social, Meta ads, Google ads and SEO. This is where AI does the most — and where the platform creative engine already lives.
1 Live (Platform)2 Near-term1 Later
Creative generation already exists — the CRM links, it doesn't rebuild

Video / image creative is produced on the live platform (app.csoul.cloud): campaigns → briefs → AI generation via the Higgsfield engine. A Meta-ads or Social Engagement in the CRM opens the platform for the actual creative work; the CRM stays the system of record for the client, the engagement and the deliverable status. No duplicate creative tooling in the CRM.

Creative / content generation
Live on Platform

AI video & image for Meta/Social campaigns via the platform's brief → generation engine. From the CRM, a creative Engagement deep-links to the matching campaign / brief on app.csoul.cloud rather than copying the workflow.

Powered by Platform engine Engagement → Platform
Content calendars
Near-term

Draft a monthly social calendar per Engagement — themes, post cadence, hooks — that becomes Deliverables & Tasks. The team approves and edits; AI removes the blank-page work each cycle.

Powered by Claude Engagement Deliverable
Ad-copy & SEO drafts
Near-term

Headlines & primary text for Meta / Google ads, plus SEO page titles, metas, briefs and keyword groupings — tied to the client's brand profile and service line. The text layer beside the platform's visual layer.

Powered by Claude MCP web Engagement
Auto status & client report summaries
Later

Roll up an Engagement's tasks + ad results into a plain-English monthly client report and an internal status digest. Later because it depends on ad-performance feeds being wired in first.

Powered by Claude MCP (ad feeds) Engagement Task
Bill / Collect
Invoices are GST-aware (CGST/SGST vs IGST, export under LUT) and multi-currency (Razorpay ₹ India, PayPal foreign). AI handles the chasing and watches the numbers — the maths stays deterministic.
1 Near-term1 Later
Draft reminders & dunning in the client's tone
Near-term

Generates the due / +3 / +7 / +14 reminder ladder, escalating politely — warmer for long-standing retainers, firmer for late one-offs. Pulls the right amount, currency and invoice number; a human approves before it goes.

Powered by Claude CRM data Invoice Payment
Payment-anomaly flags
Later

Watches for short-pays, wrong-currency receipts, duplicate or unusually late payments and a subscription that should have renewed but didn't — and raises a flag for review. Needs a payment history to learn "normal." Flag-only, never auto-adjusts the ledger.

Powered by CRM data Claude Payment Subscription
Retain / Expand
A retainer customer is at risk or ready for more. AI watches engagement health, suggests save-plays when a relationship cools, and spots the right moment to upsell another service line.
1 Near-term2 Later
Churn-risk prediction
Later

A health score per customer from signals like slipping deliverables, slowing replies, late payments and falling ad spend. Later — it needs real churn history before its warnings can be trusted.

Powered by CRM data Claude Company Engagement
Save-play suggestions
Later

When a customer trips a risk flag, suggest the retention play — a check-in call, a re-scope, a results recap, or a goodwill concession — drawing on what saved similar accounts before.

Powered by Claude CRM data Deal Activity
Upsell recommendations
Near-term

Spots customers buying one service who'd benefit from another — a Social client with no SEO, ads running with a weak site — and drafts the pitch + a Renewals & Expansion deal. The rules version is buildable now.

Powered by Claude CRM data Company Deal
Cross-cutting AI — available from anywhere in the app

Cross-cutting AI

Not tied to one stage — these follow the user everywhere. The assistant panel is the front door to every capability above, and client research feeds the brand profiles the whole journey relies on.
2 Near-term1 Later
In-CRM AI assistant panel
Near-term

A side panel on any record: "ask about this account," "draft anything," "summarize." It's scoped to the open Company / Deal / Engagement, so it answers with that record's real context — the single front door to every use-case on this page.

Powered by Claude CRM data MCP
Client research & brand-profiling
Near-term

Researches a prospect or client from their website & public presence and builds a brand profile — tone, audience, positioning, competitors. This profile feeds proposals, ad-copy, content calendars and the platform's briefs, so every output sounds on-brand.

Powered by Claude MCP web feeds Platform
Agency-wide "ask my book of business"
Later

Cross-account questions in plain English — "which retainers are at risk this quarter?", "total overdue in ₹ & USD?", "which clients have no SEO engagement?" Sits above the whole CRM; Later because it needs the full data model populated to be reliable.

Powered by Claude CRM data

Suggested build order PROPOSED — confirm

The same use-cases, sequenced. The principle: ship the assist-first writing & summarizing wins early (low risk, instant time-savings), wire scoring & risk models after data accumulates, and keep all auto-actions human-approved until trust is earned.

Wave 1 — Quick wins
FIRST AI PASSES · NEAR-TERM
  • AI assistant panel on every record
  • Draft outreach / replies / proposals
  • Summarize calls & emails → Activities
  • Client research & brand-profiling
  • Lead enrichment + first-pass scoring
  • Reminder / dunning drafting in tone
Wave 2 — Delivery scale
ONCE ENGAGEMENTS EXIST · NEAR-TERM
  • Content calendars → Deliverables / Tasks
  • Ad-copy & SEO drafts per engagement
  • Upsell recommendations (rules-based)
  • Deep-link to platform engine from creative engagements engine live
Wave 3 — Intelligence
AFTER DATA + TRUST · LATER
  • Lead auto-triage & auto-routing
  • Deal-risk & next-best-action
  • Churn-risk + save-plays
  • Auto status & client reports (ad feeds)
  • Payment-anomaly flags
  • Agency-wide "ask my book of business"

Operating principles for the AI layer

How AI behaves across all of the above — so it stays useful and trustworthy in a single-tenant, owner-run agency.

Human-in-the-loop by default

AI drafts and suggests; a person reviews before anything is sent to a client or changes the ledger. Auto-actions earn their autonomy.

Grounded in CRM context

Outputs are scoped to the real record — this Company, this Deal, this brand profile — not generic. No hallucinated facts about an account.

Maths stays deterministic

GST splits (CGST/SGST vs IGST), LUT exports, currency totals and balances are computed by code. AI writes about money, never calculates it.

Reuse the platform engine

Creative video / image generation is not rebuilt in the CRM — it links out to the live engine at app.csoul.cloud. One place for creative production.

How to use this page

Read the map as a menu, not a mandate. The only thing that exists today is the platform's creative engine (Deliver stage) — everything else is a proposal. The owner's job here is to rank what matters most and confirm the Near-term / Later split: which Wave-1 wins do we want first, and how aggressive should auto-actions ever get?

Real vs. proposed: the brief → AI video/image engine on app.csoul.cloud is live. Every other use-case, and the entire horizon ordering, is PROPOSED — confirm. The CRM objects each one writes to (Company, Contact, Deal, Activity) are live in CRM-1a; Lead, Engagement, Deliverable, Task, Invoice, Payment and Subscription are planned — so use-cases that depend on those land after the objects do.