How to Build a Travel Agency SaaS in 2026: SAFAR Case Study.

01Why we built SAFAR

I founded DreamIT in 2011 in Dhaka. Over fifteen years we've shipped more than 240 projects from Doha, Dhaka and Madrid — apps, websites, AI systems for clients in Qatar, Bangladesh, the UAE and the GCC. Along the way we also built our own products, including 4UAI, our multi-modal AI workspace.

The seed for SAFAR was a Dream Group internal problem. Dream Tourism, our group's travel arm, was running 14 agents on a stitched-together stack: a Saudi-built ticketing system, Excel for visa files, three WhatsApp numbers, a paper folder for Hajj contracts. Every operations manager I spoke to in Doha and Dhaka had the same story. We kept being asked: "do you guys do software for travel agencies?" After the fifth identical conversation I stopped saying "no, not specifically" and started saying "we will."

That decision became SAFAR by DreamIT — the AI operating system for travel and visa agencies. This is the honest story of how we built it. If you're thinking about building a vertical SaaS in 2026, I hope it saves you some of the mistakes we made.

02The customer research process

Before writing a single line of code, we spent eight weeks on customer research. Twenty-seven structured interviews with travel and visa agency owners and operations leads across Doha, Dhaka, Riyadh, Dubai and Kuala Lumpur. Each interview was 60–90 minutes, with three questions I returned to relentlessly:

1. Walk me through the worst day you had last month, hour by hour.
2. If you could fire one tool from your stack tomorrow, which one and why?
3. If a software vendor knocked on your door today, what would you want them to demo?

The patterns were brutally consistent. WhatsApp chaos, document re-typing, payment reconciliation hell, ticket changes lost in chat history. Nobody asked for blockchain. Nobody asked for a "metaverse experience". They asked for the same five boring things, over and over. That's a market.

03Tech stack decisions

I'm going to be specific here because tech-stack posts are one of the few times engineers really do read carefully. SAFAR runs on:

• Next.js 15 with the App Router for both marketing site and product UI. Server components for everything we can. We compared Next.js against Laravel and Remix — see our piece on the 2026 AI trends for Qatar businesses for the broader context.
• Postgres on Supabase for the database. Row-level security from day one for multi-tenant isolation. We considered managed Postgres on Neon and Vercel Postgres; Supabase won because of auth, storage and edge functions in one platform.
• Drizzle ORM over Prisma. Faster cold starts on serverless, schema-as-TypeScript that we actually like reading.
• Tailwind + a small in-house component library rather than shadcn out-of-box, because we wanted SAFAR to feel like a distinct product, not a Vercel template.
• Inngest for background jobs — document OCR, WhatsApp delivery, payment reconciliation. The retry semantics saved us during the launch week.
• Multi-model AI layer — Claude Sonnet for long-context document understanding, GPT-4o for fast triage, Gemini 2.5 for low-cost batch tasks. We route per-task, not per-vendor.
• WhatsApp Business API direct via Meta, not through a BSP, once volume justified it. See our WhatsApp guide for the why.
• Vercel for hosting. Single command deploys, edge functions in the Frankfurt region for low latency to GCC users.

If I were starting today I'd probably revisit the ORM choice and look harder at edge-runtime for AI calls. Everything else still holds up.

04MVP scope

The hardest part of building a vertical SaaS is saying no to features your first customers are asking for. Our MVP scope was deliberately narrow:

• Visa file management with OCR
• Multi-applicant grouping
• WhatsApp Business API integration
• One AI co-pilot, drafting replies and summarising files
• Stripe + bank transfer payments

That's it. No tickets at MVP. No Hajj at MVP. No agreements module. We shipped those over the following six months as paying customers asked for them — and we knew the customers were real because they were already paying.

05The first 10 paying agencies

Vertical SaaS founders ask me about this constantly. How do you get the first ten? Honestly: don't be cute. Sell directly to the people you already know.

Agencies 1–3 came from Dream Tourism (our own group), Dream Workforce's referral network, and one Doha-based partner who'd worked with us on a custom build years ago. They paid 50% off list price in exchange for weekly feedback calls during the beta.

Agencies 4–7 came from one warm intro at a Doha travel agents' association event. I spent a full day at that event, ran 22 conversations, demoed SAFAR on my laptop to anyone who'd watch. Four of those became paying customers within 60 days.

Agencies 8–10 came from SEO. Specifically, the search "visa agency software qatar". We'd written one good guide (an earlier version of the visa software buyer's guide) and it started ranking around month 5. Inbound demos converted at roughly 1 in 3.

By month 8 we had 10 paying agencies. By month 11 — public launch — we had 27. That's where SAFAR is today as I write this in May 2026.

06AI co-pilot architecture

The co-pilot is the most interesting technical piece, so I'll go deeper here. SAFAR's AI is structured as a multi-agent system:

1. A router classifies every inbound request — customer message, document upload, internal command — and decides which sub-agent handles it.
2. The document agent handles OCR, field extraction, validation, and triggers human review at confidence thresholds.
3. The reply agent drafts customer responses in the right language, grounded in the agency's policies and the customer's file history.
4. The compliance agent runs sanctions screening, validates Mahram relationships for Hajj groups, flags PII exposure.
5. The orchestrator sits above all of them, manages context, and decides when to escalate to a human agent.

Each agent gets a different model — Claude Sonnet for the orchestrator and document agent (long context wins), GPT-4o for the reply agent (speed and Arabic fluency), Gemini 2.5 Flash for the compliance agent (cost per call matters when you're screening every transaction). We re-evaluate the routing quarterly. The architecture maps directly to what we wrote about in 2026 AI trends: multi-agent goes mainstream, vertical AI beats horizontal.

07Pricing model evolution

We've changed our pricing three times in 11 months. Each change was educational.

V1 — per file (months 1–4). Pay-as-you-go, USD 1.20 per visa file. Customers loved it for the predictability — until they hit volume. A 60-file/day agency was suddenly looking at USD 2,160/month with no caps. We were also leaving money on the table from light users who saw the per-file pricing as expensive on first glance.

V2 — per seat with file caps (months 5–7). USD 49 per agent per month, capped at 100 files per seat. Customers hated the caps. Agencies have spiky workloads (visa season vs off-season) and caps felt like punishment.

V3 — per seat, unlimited (months 8+). USD 39 per agent per month, everything included. Simpler to sell, predictable for customers, predictable for us. Revenue went up ~30% within 60 days of the change. Lesson: pricing simplicity sells better than pricing fairness.

08Lessons learned

The honest list, in no particular order:

1. Ship narrow, charge early. We charged from day one. Free betas attract the wrong customers.
2. Pick a market you can physically visit. Doha and Dhaka are 4 hours apart. I've been in agency back-offices through both cities. That's not optional for vertical SaaS.
3. Hire one ops person who came from the industry. Our first non-engineering hire was a former visa agent. Her instincts beat our PMs' instincts every time.
4. Write before you code. Every feature spec started as a 1-page customer story. If we couldn't write it clearly, we shouldn't build it.
5. Resist the "platform" temptation. Customers asked for accounting, HR, payroll integrations. We said no. SAFAR does travel and visa, very deeply.
6. SEO is a 6-month investment. Most of our inbound now comes from guides we wrote 4–8 months ago. Start writing before you're ready.
7. Multi-model AI from day one. Single-vendor lock-in on AI is a 2024 mistake. Build the routing layer.

09Building vertical SaaS in 2026

The single biggest tailwind for vertical SaaS in 2026 is that AI has lowered the cost of building deeply specialised software. The kind of features that used to require a 30-engineer team — natural-language document understanding, multilingual customer comms, dynamic itinerary generation — are now small AI-orchestration projects.

That means a 6-person team like ours can ship a product that competes with platforms built by 50-person engineering orgs. We're not the only ones noticing — see our broader take in building a SaaS product in the Middle East.

If you're thinking about a vertical SaaS:

• Pick an industry where you already know one or two operators well.
• Spend two months in customer interviews before writing code.
• Charge from week one of beta.
• Ship the smallest possible MVP that solves one real workflow end-to-end.
• Use multi-model AI routing — don't lock yourself to one vendor.
• Start an SEO content engine on day one. You'll thank yourself in month six.

If you'd rather have our team build it with you, that's literally what DreamIT's app development team in Qatar and our AI engineering team do for clients.

10Frequently asked questions

How long did it take to build SAFAR?

From the first customer-research interview to public launch, SAFAR took 11 months. The first paying customer onboarded at month 6 on a private beta. We were live with 10 paying agencies at month 8 and opened public signups at month 11.

What is the tech stack behind SAFAR?

SAFAR runs on Next.js 15 with the App Router, Postgres on Supabase, Drizzle ORM, Tailwind, and a multi-model AI layer using Claude Sonnet, GPT-4o and Gemini 2.5 depending on the task. Background jobs run on Inngest. WhatsApp uses the official Meta Business API. Hosted on Vercel with Frankfurt edge for GCC latency.

Why did DreamIT build a vertical SaaS instead of an agency service?

We will keep doing agency work — it's where we learn what customers actually need. But vertical SaaS like SAFAR compounds in ways agency revenue never does. One product, sold many times, improving every month for everyone. SAFAR is our bet that the best 2026 software businesses go very deep on one industry.

How much did it cost to build SAFAR?

SAFAR cost roughly USD 380,000 to build to public launch, including 11 months of a 6-person team, design, infrastructure, and the first round of AI training. We self-funded from DreamIT's services revenue rather than raising outside capital. For context on what this kind of project costs in Qatar, see our app development cost guide.

Can DreamIT build a similar vertical SaaS for my industry?

Yes. DreamIT builds custom vertical SaaS products for clients in property, manpower, retail, and F&B alongside our own products. If you have a clear industry pain point and want to build a SaaS around it, our app development and AI teams can take you from blank page to live product in 6–9 months.

Want to see SAFAR in action — or talk about building your own vertical SaaS? Start a free SAFAR trial or book a founder-led call. I take the first call with most new partners myself.