How We Built Our Entire Company Operating System Without Buying a Single SaaS Tool

When we started SCX.ai — a startup with a big idea and a small but amazing team — one of the biggest challenges right from the beginning was how to operationalise the business while moving fast. How do we find who is in the market, identify leads, get quotes out, do the quote-to-cash steps that make the auditors happy, run webinars, send newsletters, and then onboard our amazing customers who are putting their trust in us?

History says the answer is easy: just buy a lot of SaaS tools and all will be good.

The challenge with that thinking is both the cost of the licence and the fit between what the SaaS software does and what our business needs. We felt we couldn't build our business around an "85% fit" — we need a 100% fit if we are to be successful.

I want to say upfront that this is for sure a self-serving post. We used our own system, SCX.ai, to create our operating environment, and we believe this is the future for a significant part of the market. But that doesn't mean it won't be helpful to others.

So here is our journey and what we saw and learned.

We Never Evaluated SaaS — And That Was the First Win

Most startups spend weeks comparing CRMs, shortlisting engagement platforms, sitting through vendor demos, negotiating annual contracts. That's time and energy pointed at someone else's product instead of your own.

We decided from day one to build the operational layer ourselves, on our own sovereign AI infrastructure. Not because SaaS is bad — it's not — but because the evaluation cycle itself is a tax. Every week spent choosing between Salesforce and HubSpot is a week you're not closing customers. We wanted that time back.

So we connected to a range of models (that will become important as we go along) running on our own system and started building.

What We Built: One System, Not Seven Subscriptions

What emerged is a single platform that handles the entire revenue workflow. Not seven tools bolted together with Zapier, but one codebase, one database, one permission model, and one AI engine.

We used SCX.ai and our endpoint to get access to the right model for each part of the system — and we flexibly moved between them. For example, we use Project MAGPiE, our sovereign LLM, to ensure that when our system runs outreach it's talking in a way that works for the reader. When we're doing complex algorithmic approaches for code, we chose the DeepSeek models running on our endpoints so we could feel confident of security and control.

Here's what it does:

Signal-Based Lead Discovery

The system continuously scans public sources — news, web, social, ASX announcements, industry publications — looking for timing signals: funding rounds, AI initiatives, executive hires, infrastructure spend, partnerships. Each raw item passes through MAGPiE's classification pipeline, which determines relevance, extracts the company and people involved, and scores how well it fits our business. What arrives in front of the sales team is a prioritised, context-rich queue — not a static list bought from a data vendor.

Contact Discovery and Enrichment

Once a company surfaces, the system identifies the decision-makers: who they are, what role they hold, how confident we are in the data, and where it came from. Full audit trail, confidence scoring, duplicate detection. No manual spreadsheet reconciliation.

Outreach Tied to the Actual Trigger

Because every lead starts as a signal — a specific event that happened on a specific date — outreach drafts reference the real context. The system knows whether the trigger was a capital raise, a product launch, or an infrastructure investment, and it aligns the message accordingly. That precision is only possible when the intelligence layer and the outreach layer share the same data model.

Newsletter Campaigns From the Same Data

The signals that feed the lead queue also feed the newsletter engine. Curate, generate, personalise, send, track — all inside the same system. No CSV export, no "sync contacts" step, no third-party template builder.

Configure, Price, Quote (CPQ)

Select a company and contact, configure products from an admin-managed catalogue, calculate tax automatically (GST for Australian customers, GST-free for exports), generate a branded PDF, email it, and track the status through to acceptance. Pricing changes happen in the admin UI, not in code. Historical quotes preserve the pricing that was live at the time — important for compliance and audit.

Webinars

A presenter studio, a public viewer link, and a private Q&A channel visible only to admins. Built on WebRTC for low-latency streaming, with Supabase Realtime powering the Q&A. One less vendor, one less login, one less billing cycle.

Admin and Service Health Dashboard

A single screen that tests every service dependency — database, search APIs, enrichment, LLM — and reports health, latency, and data quality in real time. When something degrades, we know before it affects a customer.

Why 100% Fit Matters More Than Features

Any of the major SaaS tools in each category would give us most of what we need. The issue is never the 85% that works — it's the 15% that doesn't.

That 15% is where your business is different. It's the quote structure your customers expect, the compliance attachment your legal team requires, the scoring model that reflects your win/loss history, the webinar format that matches your sales motion.

When you buy SaaS, you adapt your workflow to the tool. When you build, you adapt the tool to your workflow. At scale, the second approach compounds: every internal improvement becomes a tighter fit, and every tighter fit becomes a faster close.

The Cost Picture (Real, Not Theoretical)

We're transparent about this because it matters.

A growing company with 30–60 employees and 10–25 revenue-facing seats would typically spend across these categories:

For 15 active seats, the total lands conservatively between $5,000 and $15,000 per month — and climbs quickly once you add premium data, higher-tier plans, and the human cost of keeping it all synchronised.

We only use our infrastructure — tokens from SCX.ai — as needed. We can scale up when we need it and scale down when there are fewer changes in the market that we need to adapt to. The savings are meaningful, but the real value is that every dollar we spend improves our system, not someone else's.

The Sovereign AI Layer Underneath

Everything described above runs on SCX.ai's own infrastructure — classification, entity extraction, scoring, summarisation, outreach drafting, newsletter generation. We use Project MAGPiE–based systems that handle all of the writing to keep it connected to the local market.

We also use specialist models where they're the right tool (particularly for coding tasks), but the strategic workloads — the ones that touch customer data, learn from outcomes, and define how we go to market — run on infrastructure we control.

That matters for three reasons:

1. The feedback loop is ours — Every win and every loss trains the model on our data, not a generic corpus.
2. The iteration speed is uncapped — When the market shifts, we update prompts and scoring logic the same day. No feature request, no vendor roadmap.
3. Data sovereignty is real, not a checkbox — We know exactly where data lives, how it's processed, and what it trains.

Who This Is (and Isn't) For

If you're a very early-stage team still searching for product-market fit, buy whatever gets you in front of customers fastest. Speed matters more than architecture at that stage.

But if you have a repeatable go-to-market motion and a growing team that's starting to feel the friction of disconnected tools — the duplicate data, the broken syncs, the "ops tax" of keeping everything aligned — it's worth asking whether 85% fit is actually good enough.

For us, the answer was no. So we built the other 15% — and in doing so, built something we believe is genuinely useful to others facing the same question.

If any of this resonates, or if you want to talk about what sovereign AI capacity looks like when it's embedded directly into operations, I'm always happy to have the conversation.