From $5K to $25K/month: The Exact AI Stack a 14-Person Agency Runs On

I founded Super In Tech in 2011 as a one-person web development shop in Mohali, Punjab. Two years ago we were at $5,000/month in revenue. Today we're a 14-person AI automation agency doing $25,000/month — and the system that got us here is the same one we now sell to clients.

This is the honest breakdown of what we use, what it cost us, and the decisions that mattered. No marketing fluff. If you're an agency owner reading this, you'll see exactly where the leverage points were.

The starting state: $5K/month, mostly trading time for money

In early 2024 we were running on:

• 6 people
• Spreadsheets for project management
• Manual proposal writing (3-5 hours per proposal)
• WhatsApp for client communication
• Manual invoicing
• Sales follow-up done by me, late at night, on a phone

Revenue was $5K/month. Profit margin was thin because everyone — including me — was doing low-leverage work. Senior team members were writing proposals, answering tier-1 client questions, and chasing late invoices. The bottleneck wasn't demand. It was that we couldn't ship faster.

The first lever: an internal ops platform we built ourselves

In 2019 we built Command Center — our internal operations platform. It started as a Notion replacement and grew into a full project management system with:

• Client onboarding workflows
• Task assignment with WhatsApp notifications
• Time tracking (no separate Harvest/Toggl tool)
• Invoicing integrated with project completion
• KPI dashboards per team member
• An internal chat tool (replaced Slack for internal communication)

Why this mattered: we stopped paying for 8-9 separate SaaS tools. More importantly, the data was unified. We could see exactly where time was being lost — and 60% of it was on coordination, not actual delivery work.

Cost saved per month from killing SaaS subscriptions: ~$400 (Notion, ClickUp, Harvest, FreshBooks, Slack pro, Calendly pro, Loom pro, Toggl).

Time saved per week from unified workflows: roughly 12-15 hours across the team.

This wasn't AI yet. It was a data foundation. But it became the substrate AI could later run on.

The second lever: AVA (voice AI) — built on ourselves first

In 2024 we built AVA, our voice-first AI assistant. The bootstrap pattern was specific: we built it for ourselves first, then sold it to clients.

For me personally, AVA did three things:

1. Missed-call recovery on the agency phone line. Inbound calls in India don't happen in business hours predictably — Indian SMB founders message at 11pm, US clients call at 6am IST. Before AVA, we missed about 30% of inbound calls. AVA picks up, asks intent, books a callback if I'm not available. We went from ~40 missed inbound queries/month to ~3.

2. Outbound qualification on prospect lists. When we got a list of 200 leads from a partner, AVA would call them in batches with proper qualification logic. The qualified ones got booked on my calendar. The rest got tagged with notes. I went from "I need to call this list" being a dreaded weekend task to "my calendar fills up automatically."

3. Morning briefings. AVA reads my calendar, scans Slack/WhatsApp for urgent messages, summarizes overnight email, and tells me at 7am what matters today. 30 minutes saved per morning, 5 days a week, 22 working days a month = 11 hours/month of focused founder time.

Build cost (paid in engineering time, not cash): ~280 hours by our internal team over 4 months.

Operating cost in 2026: ~$200/month in raw infrastructure (Claude API, ElevenLabs, Deepgram, Twilio).

Revenue this unlocked: rough math, $30K/year in deals I would have missed without callback recovery + $20K/year of founder time freed for high-leverage work. AVA paid back its build cost in the first quarter we ran it.

Then we started selling AVA to clients. 2,400+ active users today. About 35% of current monthly revenue comes from AVA seats and AVA-built voice deployments.

The third lever: GoHighLevel for everything client-side

We've been a GoHighLevel partner since 2017 — before AI was even part of the product. Today GHL runs nearly every client-facing workflow:

• Lead capture forms (every client has their own GHL sub-account)
• Proposal automation (template-based, auto-personalized)
• Sales pipeline tracking
• Email + SMS sequences
• Calendar booking (replaces Calendly)
• Invoicing for retainer clients
• WhatsApp Business API integration

Why GHL specifically: it's the only platform that combines CRM + marketing + payments + scheduling + WhatsApp under one roof at SMB pricing. Alternatives (HubSpot + Calendly + Stripe + Twilio + Zapier) cost 3-5x more in subscriptions and require 5x more integration work.

Monthly cost: $497 (Pro Plan + agency sub-account allocation)

SaaS tools replaced: HubSpot Pro ($800/mo), Calendly Business ($16/mo), Stripe (per transaction), Mailchimp ($150/mo), Buffer ($30/mo), Zapier Pro ($73/mo), Loom Business ($16/mo). Total replaced: ~$1,100/month for $497.

The fourth lever: Claude API for everything text-based

This is the big one. Once we standardized on Claude (Anthropic) for all text-based AI work, the agency velocity changed dramatically.

Things we now use Claude for daily:

• Proposal generation — from a 10-minute discovery call recording → first-draft proposal in 3 minutes. Engineer reviews and finalizes in 30 minutes. (Was 3-5 hours.)
• Email drafting — every client communication starts as a Claude draft. Editor catches issues and personalizes. Saves ~6-8 hours/week across the team.
• Code review — internal engineering uses Claude to review pull requests before human review. Catches obvious bugs and style issues.
• Content production — every blog post, case study, social post starts as a Claude draft.
• Customer support drafts — tier-1 support replies are Claude-drafted, human-reviewed before send.
• Meeting summaries — every Zoom call gets transcribed and summarized by Claude. Action items pulled out automatically.

Monthly Claude API cost: ~$1,200 (Opus + Sonnet across all use cases)

Human hours saved per week: rough estimate 60-80 hours across the 14-person team.

At average team cost ($25/hr fully loaded), 70 hours/week × 4.3 weeks = ~$7,500/month of labor effectively replaced. We pay $1,200 for it. Net leverage: ~6x.

The fifth lever: model-agnostic stack discipline

The trap most agencies fall into is locking themselves to one AI vendor. We deliberately stay model-agnostic. The stack:

When a new model ships (Claude 4.7, GPT-5, Gemini 2.0), we benchmark it on our real workloads in a weekend and decide whether to migrate. That discipline alone has saved us from being locked into one vendor's pricing.

What scaling from $5K to $25K actually took

Four business decisions that mattered more than any individual tool:

Decision 1: Fixed-price builds, never hourly

We stopped billing hourly in mid-2024. Now every build is scoped, fixed-priced, milestone-paid. Reasons:

• Hourly billing punishes speed and rewards drag
• Fixed price forces us to scope rigorously upfront
• Clients prefer predictable cost
• It made our delivery velocity 2-3x faster because we stopped padding hours

Decision 2: Operate-and-iterate retainer model (not project-and-leave)

About 70% of our current revenue is recurring operate-and-iterate retainers, not one-off builds. The model:

• After we ship a system, we stay on a monthly retainer to operate it
• We tune prompts, swap models when better ones ship, integrate new data
• We measure the business metric weekly and report
• Clients pay $497-$7,500/month depending on system complexity

This flipped the business from "hunt for the next deal" to "compound existing clients." Way less stress.

Decision 3: 30-day results guarantee

Every contract has it. If we haven't moved the agreed metric within 30 days of launch, we keep working free until we do. Sounds risky. In practice: we're rigorous about scoping, so we've never had to invoke it on a real project. But putting it in the contract closes 30-40% more deals because it removes the buyer's fear of paying for nothing.

Decision 4: Founder-led, weekly

I personally review every active engagement once a week. Not as a bottleneck — as a quality check. The team handles delivery, I check that we're moving the metric and that the client is happy. Takes about 4 hours every Friday. Catches issues early. Keeps the company's reputation tight.

What I'd do differently if I were starting today

Reflecting on the journey:

1. Build the internal ops platform earlier. Command Center took us a year to build properly. We should have started in 2017. Manual coordination was eating 20+ hours/week of team time we didn't realize until we eliminated it.

2. Pick one CRM and commit. We tried HubSpot, Pipedrive, Zoho, custom builds before settling on GHL. The thrashing cost us 6+ months of velocity. Pick one early and commit.

3. Start with voice AI, not chatbots. Voice picks up where chat hands off. Chatbots are mostly toys. Voice agents are productivity multipliers.

4. Charge more, deliver more. We undercharged for 5+ years out of insecurity. Once we doubled prices and added the results guarantee, close rate went UP, not down. Premium pricing signals confidence.

The full monthly stack cost in 2026

For full transparency, here's what we pay per month to run a $25K/month, 14-person agency:

That's 8.6% of revenue going to tools. For comparison, the typical agency at our revenue level pays 15-20% to tools because they haven't consolidated onto unified platforms.

What's next on the roadmap

We're not done. The next 18 months for us:

• Hit Gold tier on the GoHighLevel affiliate program (cumulative $100K+ commissions)
• Reach 1,000+ sub-accounts on our GHL agency reseller setup
• Ship Clone Studio publicly (multi-tenant SaaS for founder-cloning + reel generation)
• Open a satellite office in either Bangalore or Austin (TBD based on team distribution)

The playbook isn't a secret. Most of it is in our pillar guides. The hard part isn't knowing what to do — it's having the discipline to ship the unsexy infrastructure work (internal ops, integration plumbing, eval suites) instead of chasing the next shiny client.

If you're an agency owner and any of this resonates, book a 30-minute call. I won't pitch you. We'll just compare notes on what's working and what isn't in your stack. Founder to founder.

VJ Founder, Super In Tech