How a Small Tax Firm Doubled Revenue With a No-Code AI Assistant (The TaxWorld Ezylia Case Study)

How a Small Tax Firm Doubled Revenue With a No-Code AI Assistant (The TaxWorld Ezylia Case Study)

The usual disclaimer up front: the core numbers in this story come from a vendor-published case study on CustomGPT.ai's website. TaxWorld (Ireland/UK) is a CustomGPT.ai customer, and the metrics — 2,000+ queries per day at 98% accuracy, 97.5% resolution rate across 189,351 documented queries, 500+ hours saved per week, 200% year-over-year revenue growth — were published by the vendor.

That noted: the mechanics are specific, the query count is documented, and the setup is replicable with zero engineers. What follows is an honest analysis of what TaxWorld actually did, whether the numbers hold up to scrutiny, and what a 10-person CPA firm in the US or Canada can realistically take from it.

---

What TaxWorld Actually Did

TaxWorld is a small technology company based in Ireland that builds AI-powered tools for small and mid-sized accounting practices. Their primary clients are the kind of firms you likely run — 5 to 50 accountants handling business and personal tax, VAT, trust structures, and compliance work.

The problem TaxWorld's clients faced was familiar: senior staff time disappearing into answering repetitive tax questions. A client calls, asks about a tribunal ruling. A junior accountant spends 45 minutes researching an answer that a more experienced colleague could have answered in five. Multiply that across a 20-person practice and 40+ client interactions per week, and you have hundreds of hours of capacity evaporating into queries that AI can handle.

TaxWorld founder Alan Moore built Ezylia to solve that problem — not for his own firm, but as a product his accounting firm clients could use. He used CustomGPT.ai, a no-code platform, and connected it to a corpus of thousands of legislative documents, tribunal decisions, and case law records. No internal engineers. No six-figure development budget. According to Moore: "CustomGPT.ai let us punch far above our weight. With almost no engineering budget, we built an assistant that now answers tens of thousands of complex tax questions and fuels our revenue growth every month."

The result was Ezylia — an AI assistant that handles tax research and client Q&A, deployed across TaxWorld's accounting firm clients.

---

The Numbers: What They Claimed and What They Mean

Let's go through each metric honestly.

2,000+ queries per day at 98% accuracy. This is aggregate volume across TaxWorld's entire client base, not one firm. If TaxWorld serves 50 accounting firm clients each handling 40 queries per day, you get to 2,000. For a single 10-person CPA practice, the comparable figure is more like 50–100 client and internal queries per week. That's still meaningful volume for an AI assistant to handle — it's not nothing.

97.5% resolution rate across 189,351 queries. This is the most credible number in the dataset. It's specific, it's documented, and 97.5% across nearly 190,000 queries is not a launch-week vanity metric. It suggests a system that's been tested at real scale. The caveat: resolution rate means the AI returned an answer the user accepted. It does not mean every answer was legally defensible. For high-stakes compliance questions, AI-assisted answers should still be reviewed by a qualified professional.

500+ hours saved per week. This is aggregate across TaxWorld's client base. In headcount terms: 500 hours per week is roughly the equivalent of 12–13 full-time employees doing nothing but answering research and Q&A queries. At the individual firm level, if Ezylia deflects 30–40% of incoming tax questions, a 20-person practice might realistically recover 8–15 hours of senior professional time per week.

200% year-over-year revenue growth, ARR ~€1 million. This is the metric that gets the headline — and the one that needs the most context. TaxWorld is a software company. Their 200% revenue growth reflects their AI product gaining accounting firm clients, not a CPA practice increasing its advisory revenue. The growth came from selling Ezylia to firms, not from a firm's own client base expanding because they used AI.

That distinction matters. If you're running a 10-person CPA firm, the 500 hours per week freed is your metric. The 200% revenue growth belongs to TaxWorld's product business.

---

Is This Replicable for a 10-Person CPA Firm?

Here's the honest breakdown.

What it requires:

• A document library. You probably have one — it's just not organized for AI consumption. Tax code PDFs, IRS publications, your firm's procedures manual, a common Q&A document you've built over the years. That's your training corpus.
• CustomGPT.ai access. The platform starts at approximately $99/month for the Starter tier, with $399/month and $599/month tiers for higher query volumes.
• 10–20 hours of setup time. Connecting document sources, training the assistant, reviewing initial responses, and iterating on accuracy. This is a week-long build, not a weekend project — but it doesn't require a developer.

What it doesn't require:

• Engineers or an IT department
• Custom software development
• A large budget — $99–$399/month is the likely range for a small firm's first deployment

The realistic outcome for a 10-person firm:

If your team handles 50+ client or internal tax questions per week, an AI assistant trained on your firm's common Q&A and procedures can deflect 30–40% of those to self-service. That frees 10–15 hours of senior professional time per week — roughly the equivalent of half a part-time staff position.

At a billing rate of $200/hour, 10 recovered hours per week equals $2,000/week in recovered capacity. Against a $99–$399/month platform cost, the math favors acting.

The honest caveat:

AI accuracy is bounded by document quality. If your process documentation is inconsistent — or doesn't exist in organized form — the AI assistant's answers will reflect that inconsistency. TaxWorld achieved 98% accuracy because they trained Ezylia on a curated, specific corpus of legislative documents and tribunal decisions, not generic content. A CPA firm building on disorganized PDFs and ad hoc files will get disorganized answers. Start by auditing and organizing your firm's existing documentation before building on top of it.

---

CustomGPT.ai: What It Is and How to Evaluate It

CustomGPT.ai is a no-code platform that lets you build AI assistants trained on your own documents. You connect sources — PDF files, website URLs, Google Drive folders — and the platform trains an AI model on that content. You deploy the assistant via a chat widget on your website, an internal staff portal, or via API.

For accounting firm use cases, the platform handles:

• Tax code PDFs and IRS/HMRC publications
• Firm procedures and process documentation
• Client FAQ libraries
• Tribunal decisions and case law records
• State or jurisdiction-specific tax guidance

Pricing tiers (approximate — verify at CustomGPT.ai before purchasing):

• Starter: ~$99/month — internal Q&A with moderate query volume
• Basic: ~$399/month — higher query volume, more document sources
• Standard: ~$599/month — enterprise-scale query volumes

One alternative worth considering: Notion AI works well for simpler internal Q&A within a team workspace. For external-facing client queries or for training on structured legal and regulatory documents, CustomGPT.ai is the more capable option. The tradeoff is setup time and cost — Notion AI is already in your workflow if you use Notion; CustomGPT.ai requires a dedicated build.

---

The Knowledge Base Angle: Why This Matters Beyond Client Q&A

The default mental model for an AI assistant is client-facing: a chat widget that answers inbound questions. That's a valid use case.

The more immediately valuable application for most small firms is internal. When every junior staff member can query "what's our procedure for a client with multi-state nexus in four jurisdictions" and get a consistent, documented answer, training time drops and work quality improves. The institutional knowledge that currently exists only in the heads of your two most experienced people becomes queryable by the entire team.

That's not client automation — it's firm-level knowledge infrastructure. It's more defensible (you control the document corpus entirely), more likely to achieve high accuracy early, and more immediately useful to a firm that's focused on capacity and consistency rather than client self-service.

For more context on AI tools across tax, contract review, and consulting practices, see our overview of AI tools by practice area for professional services firms.

---

The Crossing Report Verdict

Verdict: The setup is real. The 200% revenue claim is not your number — but the 500 hours per week is.

TaxWorld's Ezylia case study is one of the few published AI examples for small accounting firms that includes specific, documented metrics — not projections, not pilot program estimates, but a running system with a 97.5% resolution rate across 189,351 queries.

The revenue growth headline belongs to TaxWorld's software business, not to a CPA firm's client revenue. But the capacity math is yours to claim. If you handle 50+ tax queries per week, a no-code AI assistant trained on your firm's documents can deflect a third of them — recovering meaningful senior professional time at under $400/month in platform costs.

The one thing to do this week: Audit three documents from your firm that answer the questions your team gets asked most often — a procedures guide, a common Q&A file, a jurisdiction-specific reference. If those three documents are current and organized, you have the foundation for a CustomGPT.ai build. If they're not organized, start there. The tool is only as accurate as what you feed it.

---

FAQ

What is TaxWorld Ezylia and what does it do?

TaxWorld Ezylia is an AI assistant built by TaxWorld (Ireland/UK) using the no-code platform CustomGPT.ai. It handles tax research queries and client Q&A for accounting firms. The system was trained on thousands of legislative documents, tribunal decisions, and case law records. As of 2026, Ezylia handles 2,000+ queries per day at 98% accuracy, with a 97.5% resolution rate across 189,351 documented queries. It was built by TaxWorld founder Alan Moore with zero internal engineering staff.

Can a small CPA firm build an AI assistant without a developer?

Yes. TaxWorld built Ezylia using CustomGPT.ai, a no-code platform that connects to existing document libraries — PDFs, URLs, Google Drive files — and trains an AI assistant on that content. The setup required no internal engineers and approximately 10–20 hours of configuration time. For a CPA firm with organized process documentation, client FAQ libraries, or firm procedures, a comparable build is possible without any coding.

How much does CustomGPT.ai cost for an accounting firm?

CustomGPT.ai pricing starts at approximately $99/month (Starter tier), with higher tiers at approximately $399/month and $599/month depending on query volume and features. For a small accounting firm handling internal Q&A and a limited volume of client-facing queries, the $99–$399/month range covers most early-stage deployments. Verify current pricing at CustomGPT.ai before purchasing — tiers and rates are subject to change.

How many hours per week can AI save a small accounting firm?

TaxWorld's Ezylia freed 500+ hours per week across their client base of small and mid-sized accounting practices. For a single 10-person CPA firm, the realistic target is narrower: if your team handles 50+ client or internal tax questions per week, an AI assistant trained on your firm's processes and common Q&A can deflect 30–40% of those queries to self-service — freeing an estimated 10–15 hours of senior professional time per week. At a $200/hour rate, that's $2,000 per week in recovered capacity.

Is a no-code AI assistant accurate enough for tax and compliance questions?

Accuracy depends on document quality, not platform sophistication. TaxWorld achieved 98% accuracy and a 97.5% resolution rate by training Ezylia on a curated, specific corpus — tax legislation, tribunal decisions, and case law records — not generic internet content. For a CPA firm, the same principle applies. An AI assistant trained on your actual procedures, IRS guidance PDFs, and documented firm processes will significantly outperform one trained on poorly organized or inconsistent documentation. The accuracy floor is your document library, not the tool.

---

The Crossing Report covers AI adoption for professional services firms. Subscribe free to get the analysis, without the vendor spin.