I want to start with the honest version of where we were before ENZO OS existed — because I think the sanitised founder story (where everything is a neat journey from problem to solution) does a disservice to anyone trying to build something real.
In early 2025, ENZO Digital was operationally functional but operationally fragile. We had clients across India, the UAE, and the UK. We had a small team. And we had a system that was held together with late-night WhatsApp messages, a Google Sheet with 47 tabs that only Shubham fully understood, and a Notion database where half the pages hadn't been updated in two months.
We weren't failing. But we were spending 30–40% of our operational capacity on administration — chasing approvals, recreating monthly reports from scratch, manually generating invoices, and having the same conversation about project status three times across different channels before anyone actually had an answer.
That's when I told Rhythm: we're building our own system. Not buying one. Building one.
1. The Problem We Were Solving
The tools available to digital marketing agencies in India fall into two categories: generic project management tools (Asana, Monday, ClickUp) that aren't built for agency-specific workflows like client reporting and quotation generation, or expensive Western agency management software (Agency Analytics, Workamajig) that costs $200–500/month and still doesn't handle Indian-specific requirements like GST calculations, Rajasthan vs other-state IGST/CGST splitting, and INR-denominated billing.
More fundamentally, none of them had AI embedded in the actual workflow — not as a chatbot feature bolted on, but as a native part of how performance data is analysed, how reports are summarised, and how client insights are surfaced.
We had three specific problems we needed to solve:
- Client reporting was entirely manual. Every month, Saksham or Shubham would open Google Analytics, Google Ads, Meta Ads Manager, and a Google Doc — and spend 2–4 hours per client compiling data, writing commentary, and formatting a report. For 4 clients, that's a full working day of non-billable time every month.
- Quotation generation was inconsistent. We were generating quotations in Google Docs, recalculating GST manually, and versioning them in Drive folders. Errors happened. Time was wasted. New clients occasionally received different pricing for the same service because there was no single source of truth.
- No visibility into agency financials in real time. I couldn't tell you at any given moment what our total active contract value was, what our collection rate was this month versus last, or which services were most profitable. I had to compile this from multiple sources every time I needed to think about the business.
We decided to build rather than buy for two reasons. One: no existing tool solved all three problems in one system, and we weren't willing to manage three separate tools with no integration. Two: building our own system would force us to deeply understand the architecture of AI-powered business software — and that understanding would become a service we could offer clients.
2. What ENZO OS Actually Is
ENZO OS is a web application hosted at os.enzodigital.in — accessible from any device, browser-based, no installation required. It's the central operating system for everything that happens inside ENZO Digital that isn't client-facing work itself.
Three people currently use it: Saksham (Founder & CEO), Shubham (Co-Founder & CGO), and Rhythm (Lead Developer). Role-based access means Rhythm sees projects, tasks, SOPs, and his work assignments — but not quotations, financial data, or billing. Saksham and Shubham have full admin access to everything.
3. Every Module — What It Does
4. How AI Is Embedded in the Workflow
The AI layer in ENZO OS is not a chatbot. It's not a "ask me anything" interface. It's embedded in specific, high-value workflow moments where AI changes the quality or speed of output in a measurable way.
Report Analysis and Metric Extraction
When a performance report is uploaded — a PDF or data export from Google Analytics, Google Ads, or Meta Ads — Claude reads the document and automatically extracts the key metrics: sessions, conversions, CPA, ROAS, CTR, revenue. It populates the relevant fields in the performance tracker without manual data entry. What previously took 15–20 minutes of copy-pasting across documents now takes under a minute.
AI Executive Summaries
Once metrics are extracted, Claude generates an executive summary of the reporting period — highlighting what improved, what declined, what anomalies exist, and what the data suggests for the next month's strategy. This is not a generic summary — it uses the client's specific data, their historical benchmarks, and our agency context. The result is a draft that Saksham reviews and refines rather than writes from scratch.
Month-on-Month Comparison
ENZO OS automatically generates comparison charts showing current month vs previous month vs same month last year for every tracked metric. Claude adds written commentary on the trends — whether a dip is seasonal, whether an improvement is attributable to a specific campaign change, whether a metric is concerning or within normal variance.
"The goal was never to replace the thinking — it was to eliminate the non-thinking work that was consuming 30–40% of our time. AI handles the extraction and the first draft. I handle the judgment and the client relationship."— Saksham Mehra, Founder & CEO, ENZO Digital
5. The Tech Stack and Why We Chose It
| Layer | Technology | Why We Chose It |
|---|---|---|
| Frontend | React 18 + Vite + Tailwind CSS | Same stack as enzodigital.in — Rhythm's strongest environment, fast build times, component reusability |
| Routing | React Router DOM v6 | Client-side routing for SPA feel — no page reloads between modules |
| Backend & DB | Supabase (PostgreSQL) | Managed database + authentication + real-time + storage in one service. Free tier sufficient for our current scale. Row-level security for role-based access. |
| Authentication | Supabase Auth | Built-in, handles session management, JWT tokens, email verification |
| AI Layer | Anthropic API (Claude Sonnet) | Best-in-class for document analysis and structured output extraction. Consistent JSON response quality for metric extraction. |
| Charts | Recharts | React-native, lightweight, clean defaults, easy to customise to ENZO brand colours |
| PDF Generation | jsPDF + html2canvas | Client-side PDF generation without server dependency — quotations and invoices generated instantly in the browser |
| Hosting | Vercel + GitHub | Automatic deployments on push, edge network, zero DevOps overhead for a 3-person team |
The infrastructure cost of ENZO OS at our current scale: approximately ₹0/month. Supabase free tier, Vercel free tier, Anthropic API at usage-based pricing that amounts to a few hundred rupees per month for our current report volume. The entire system costs us less per month than a single tool subscription we replaced.
6. How We Built It — The Development Approach
ENZO OS was built using an AI-accelerated development workflow. Rhythm led the architecture and engineering — every decision about data model design, component structure, authentication flow, and API integration was made by him. The AI-assisted development approach meant that what would have taken a 3-person engineering team 4–6 months to build, a single developer shipped in weeks.
This is the same development approach we now bring to client projects. AI doesn't replace the engineer — it eliminates the mechanical parts of engineering (boilerplate, repetitive patterns, documentation lookups) so the engineer spends their time on decisions that actually require judgment: architecture, edge cases, performance, security.
The result was a production system with genuine complexity: role-based access control, real-time database subscriptions, PDF generation, AI API integration, GST calculation logic for multiple Indian tax contexts, and Recharts visualisations — all shipped and in daily use within the agency.
When we quote a client on building an AI-powered internal tool, we're not estimating from theory. We've built one. We know what takes time (data modelling, access control, edge cases in PDF generation), what's fast (UI components, basic CRUD, chart integration), and where AI integration actually adds value vs where it's complexity for its own sake. That experience is in every estimate we give.
7. Five Things We Learned Building AI Products
The highest-ROI place to use AI in a business system isn't answering general questions — it's the moment when raw data needs to become an actionable insight. Report analysis, anomaly detection, performance commentary: these are the points where AI eliminates hours of skilled human time per month.
Getting Claude to reliably extract 12 specific metrics from a PDF into a clean JSON object, every time, regardless of how the report is formatted — this is genuinely difficult to get right. Prompt engineering for structured extraction is a skill in itself. We spent more time on this than any other single feature. The payoff is that every downstream feature (charts, summaries, comparisons) depends on clean data extraction working perfectly.
We underestimated this. Adding role-based restrictions to a system that was initially built without them is 3× harder than designing them in from the start. Supabase's row-level security policies are powerful but require careful design. If you're building any multi-user internal tool, model your permissions before you model anything else.
Building ENZO OS was the easy part. Getting everyone to actually use it — consistently, for everything — took deliberate effort. Old habits (WhatsApp for quick client updates, Google Sheets for one-off calculations) die hard. The system only has value when it's the system of record. Incomplete adoption is worse than no system at all because you end up maintaining both.
The first version of ENZO OS had features we thought we'd use every day and didn't touch for two months. The features we use constantly — quotation generation, AI report summaries, performance tracking — are the ones we designed based on actual pain points rather than theoretical completeness. Build the 20% that solves 80% of the friction. Add complexity when the pain of not having it becomes clear.
8. How Clients Are Responding
Clients don't see ENZO OS directly — it's our internal system. But they experience its effects in every interaction.
Monthly reports that previously took 2–4 hours to produce are now generated in under 30 minutes — more consistent in format, richer in AI-generated insight, and delivered earlier in the month. Quotations that previously arrived 24–48 hours after a client meeting now arrive within the hour. Invoice queries ("I haven't received my invoice for March") get resolved in seconds rather than requiring a search through email threads.
Several clients have asked directly: "How do you produce reports this fast? Our previous agency took a week." When we explain that we've built custom tooling with AI-assisted analysis, the response is consistently positive — it signals investment in infrastructure and operations that clients associate with a serious, long-term agency relationship rather than a freelancer-scale operation.
Two clients have asked if we can build something similar for their businesses. That's the origin of ENZO OS as a service.
9. What's Next — MCP Integration Roadmap
The current version of ENZO OS requires manual data upload — performance reports are exported from Google Analytics, Google Ads, and Meta Ads and uploaded as PDFs or CSV files. The next major evolution is direct data integration via MCP (Model Context Protocol) servers.
The planned MCP integration stack:
- Google Analytics MCP: Pull traffic data, conversion data, and goal completions directly into ENZO OS without manual export.
- Google Ads MCP: Pull campaign performance, spend, ROAS, and impression share in real time.
- Meta Ads MCP: Pull ad account performance, CPL, creative performance, and audience data.
- Automated monthly reporting: On the 1st of each month, ENZO OS automatically pulls last month's data for all active clients, generates AI summaries, and posts draft reports to a review queue — ready for Saksham's review and dispatch within hours of month-end.
This MCP integration layer — connecting Claude to live business data sources through standardised server interfaces — is what transforms ENZO OS from a reporting tool into a genuine AI-powered command centre for agency operations. We expect to ship this in Q3 2026. You can read our full deep-dive on MCP architecture in our MCP servers guide.
10. Why We're Making It a Service
Building ENZO OS taught us something we didn't fully appreciate when we started: the operational problems we were solving are not unique to ENZO Digital. Every growing agency, every founder-led service business, every professional services firm faces the same friction — fragmented tools, manual reporting, inconsistent processes, no financial visibility.
The difference is that most of them are solving it with off-the-shelf software that doesn't fit their workflow — or not solving it at all and accepting the operational drag as a cost of doing business.
We're now offering ENZO OS as a custom-built service: we take the architecture we've built, adapt it to a client's specific workflow, integrate their data sources, and deploy it as their internal operating system. The core modules (client management, quotations, billing, P&L, AI reporting) are transferable to any service business. The AI layer is configurable to whatever data sources and reporting requirements the client has.
This isn't a SaaS product with a monthly subscription. It's a custom implementation — we build it, we deploy it, we maintain it. For businesses serious about operational efficiency and AI integration, it's the highest-leverage system investment available.
Want Something Like ENZO OS for Your Business?
We build custom AI-powered operating systems for agencies, professional services firms, and growing businesses. Built to your workflow, integrated with your data, deployed and maintained by ENZO Digital.
Talk to Us About ENZO OS →