ABC is an agentic bookkeeping service for Singapore SMEs, built in partnership with Wenhao Dong. The thesis: AI automates 80% of routine bookkeeping, enabling one accountant to serve 70–100 clients at 70%+ gross margins — SaaS-like economics on services revenue. Per-ledger-line pricing ($1 clean / $2 manual, ~$1.33 blended) with a floor of S$150–200/month. The wedge is bank statement reconciliation; the moat is shared vendor memory across clients; the expansion play is Stripe-like ownership of the entire SMB financial stack.2
| Layer | What | Size | Source |
|---|---|---|---|
| Global TAM | Bookkeeping services worldwide | US$12.67B (2026), 9.4% CAGR → US$20B by 2030 | 2 |
| Segment TAM | Accounting outsourcing / BPO | US$56B (2022), 9.1% CAGR → US$110B by 2030 | 3 |
| Regional TAM | APAC accounting services | ~US$180B (27% of US$660B global accounting services) | 4 |
| SG Local TAM | SG registered entities needing bookkeeping | 471,321 companies + 136,971 sole props = 608K entities (ACRA Mar 2026). Est. 40% need outsourced bookkeeping = ~243K. At avg S$3,000/yr = S$729M | 1 |
| Addressable TAM | What ABC can realistically reach (Year 1–3) | 1,000–5,000 SMBs via accountant partner network. At S$200/mo avg = S$2.4M–12M ARR | Founder assessment |
| Company | Model | Pricing | Funding | Got Right | Got Wrong |
|---|---|---|---|---|---|
| Osome | Platform + human accountants | From S$75/mo | $30M+ | Bundled incorporation + accounting; strong startup/ecomm positioning | Still human-heavy operations; not truly AI-automated9 |
| Sleek | Platform + human accountants | From S$65/mo | $10M+ | Fast turnaround; modular packages; foreign founder focus | Limited automation; traditional delivery behind the app10 |
| Grof | Traditional firm + digital | From S$120/mo | Bootstrapped | PSG grant expertise; established client base | Minimal AI; traditional service model18 |
| Harvest Accounting | Boutique accounting firm | S$150–800/mo | Bootstrapped | Local expertise; personalized service | No automation; limited scale |
| Company | Model | Scale | Playbook | Why It Doesn't Apply to ABC |
|---|---|---|---|---|
| Basis.ai | Agentic AI for accounting firms | $1.15B valuation, $100M raised Feb 2026, 30% of Top 25 firms5 | Multi-agent architecture on OpenAI; autonomous end-to-end workflows (tax, reconciliation, audit)17 | Serves LARGE accounting firms in US. Different market segment. But if they expand to APAC/SMB, ABC is in their blast radius. |
| Pennylane | All-in-one accounting platform | $4.25B valuation, $204M raised Jan 2026, ~€100M ARR, 6,000 firms, 800K companies6 | Platform play — replaced fragmented tools. 1,000 employees. | Europe-focused (France → DACH expansion). Not in APAC yet. But proves the market values integrated AI accounting at massive multiples. |
| Botkeeper | AI bookkeeping for accounting firms | DEAD — $90M raised, shut down Feb 20267 | Sold to accounting firms; claimed AI but used offshore labor (Philippines). Customer concentration (30–40% revenue from 10 clients).8 | The CAUTIONARY TALE. Fake automation + customer concentration = death. ABC must prove genuine automation. |
| QuickBooks (Intuit) | Accounting SaaS + AI features | US$16B revenue (Intuit FY2025) | Incumbent adding AI (transaction categorization, receipt matching). 7M+ subscribers. | Horizontal platform. Won't build Singapore-specific vendor memory or WhatsApp intake. But sets the automation floor — anything ABC does must beat QBO's built-in AI. |
| Xero | Accounting SaaS + AI features | NZ$2B revenue, 4.2M subscribers19 | Cloud-first, strong in APAC/UK. Adding AI features. PSG-approved in SG. | ABC builds ON TOP of Xero. Xero is the ledger; ABC is the intelligence layer. Partnership opportunity, not competitor — unless Xero builds its own agentic layer. |
| Dext | Receipt capture + expense automation | Acquired by Iris Software, 500K+ users | OCR + auto-categorization. 99% extraction accuracy. | Focused on receipt capture, not full bookkeeping. Narrow tool, not a service. |
Botkeeper (2015–2026, $90M raised, DEAD)
Bench Accounting (2012–2024, $110M raised, sold for parts)
| Metric | SG Market Rate | Osome / Sleek | ABC Target | Source |
|---|---|---|---|---|
| Monthly fee | S$150–800/mo | S$75–227/mo | S$150–200/mo | 10 |
| ARPU (monthly) | ~S$400 | ~S$100 | S$200 | Industry average |
| Churn (annual) | 10–15% | Unknown | 10% (target) | Industry benchmark |
| CAC | S$500–2,000 | Unknown | S$200–500 (partner referral) | Estimated |
| LTV (3-yr) | S$10,800 | ~S$2,700 | S$6,480 | Calculated |
| Gross margin | 25–35% | ~40–50% | 65–75% (target) | Industry vs thesis |
| Component | Per-unit Cost | Assumption | Source |
|---|---|---|---|
| AI inference (categorization) | S$0.50–5.00/mo | GPT-4o-mini at $0.15/M tokens; ~200 txns × 3 calls × 2.5K tokens each | OpenAI pricing Mar 2026 |
| OCR / receipt processing | S$1–3/mo | ~50 receipts/mo at $0.02–0.05 each | Dext / AssemblyAI pricing |
| Xero API / integration | S$0 | Xero Partner API is free for partners | Xero partner program |
| Human exception handling | S$20–50/mo | 20% of transactions need human review; bookkeeper at S$4,000/mo across 80 clients | ABC thesis |
| Infrastructure (hosting, storage) | S$2/mo | AWS / Cloudflare, amortized | Estimated |
| Total COGS per client | S$24–60/mo |
The core equation for ABC's viability is the Leverage Ratio (L):
L = Revenue / Cost to deliver
Traditional accounting: L = 1.3x — 15 clients/BK × $400 / ($4K + $500 overhead). Margins are thin because every dollar of revenue requires almost a dollar of human labor.
AI-leveraged (ABC thesis): L = 2.8–3.3x — 80–100 clients/BK × $200 / ($4K + tokens + overhead). The gap is pure automation leverage: AI handles the routine, the bookkeeper handles the exceptions.
| Clients/BK | Revenue/mo | Cost/mo | L | Gross Margin |
|---|---|---|---|---|
| 15 traditional | $6,000 | $4,500 | 1.3x | 23% |
| 40 | $8,000 | $5,100 | 1.6x | 36% |
| 70 | $14,000 | $5,550 | 2.5x | 60% |
| 80 | $16,000 | $5,700 | 2.8x | 64% |
| 100 thesis | $20,000 | $6,000 | 3.3x | 70% |
| Signal | Source | Implication |
|---|---|---|
| Basis.ai raises $100M at $1.15B, Feb 2026 | 5 | Validates market — but also means well-funded global competition |
| Botkeeper shuts down after $90M, Feb 2026 | 7 | Fake automation + customer concentration = death. Warning, not refutation |
| Pennylane at $4.25B, 800K companies, Jan 2026 | 6 | Platform play works at scale. European focus — APAC still open |
| 80% of bookkeeping tasks automatable (McKinsey 2025) | 13 | The capability exists — the question is implementation quality |
| 46% of accountants use AI daily, up from 18% in 2023 | 14 | Adoption accelerating — the market is moving |
| SG fintech adoption 38% → 57% in one year | 11 | SG SMEs are ready for digital solutions |
| 73% of SG SMEs cite rising costs as top threat | 11 | Cost pressure creates demand for cheaper bookkeeping |
| Transaction categorization: 90–97% accuracy (2026) | 20 | Technical feasibility confirmed for core use case |
| PSG grant: up to 50% funding for SME accounting software | 15 | Government co-pay could halve effective price for customers |
Eric San (HK) — AI/software infrastructure, 5 AI ops clients currently, multi-project operator. Brings the automation stack (mufu, Cursor, eval pipelines) and architectural discipline.
Wenhao Dong (SG) — Client pipeline, SMB network in Singapore, runs Smilie (gifting), hands-on operator. Brings warm intros, real client data, and on-the-ground execution.
| # | Advantage | Why It Matters |
|---|---|---|
| 1 | Wenhao's SMB network in Singapore | Warm intros to prospects — no cold start |
| 2 | Accounting partner as co-builder + first customer | Real use cases, real data, domain credibility |
| 3 | Eric's AI infrastructure stack | mufu, Cursor, eval pipelines — speed to build |
| 4 | Real Xero data from Smilie | Hardest test case — 4,500 txns/month. Proves or kills thesis. |
Phase 0: Ship bank statement reconciliation for Smilie (partner's real data). Measure automation rate and token cost. 2 weeks. If automation > 75%, proceed. If < 60%, reconsider.
| Phase | Timeline | Scope | Target Metric |
|---|---|---|---|
| Phase 1 | Apr–May 2026 | 5–10 pilot clients from partner + Wenhao network. Free/discounted. | Measure real automation rates, collect feedback, build vendor memory |
| Phase 2 | Jun–Aug 2026 | 20–50 paying clients. S$150–200/mo. Hire first bookkeeper. | Prove L > 2x on real data |
| Phase 3 | Sep–Dec 2026 | 100+ clients. Decide: fund/scale or lifestyle business. | Prove L > 2.5x. Net retention > 100% |
VERDICT: Conditional — proceed to pilot, but with hard gates.
The unit economics are sound IF automation exceeds 75% on real SMB data. Token costs are negligible (S$0.50–5/client/month), and the leverage math shows L > 2.5x is achievable at 70+ clients per bookkeeper. The SG market is underserved by truly AI-native accounting services, and Wenhao's network + accounting partner provide a warm-start advantage that most AI accounting startups lack.
But the window is narrow. $350M+ has been deployed into AI accounting in the past 6 months. Basis.ai and Pennylane are global machines that could reach APAC within 12–18 months. ABC's only defensible position is Singapore-specific vendor memory + speed. The Smilie trial is THE decision point — if automation exceeds 75% on 4,500 txns/month, the thesis holds and it's worth committing. If it doesn't, clean exit before May 15.
The one thing that would change the answer: proven automation rate on real, messy SG SMB data. Everything else is theory until then.
What ABC needs to become a very valuable company:
1. L > 3x sustained (clients-per-bookkeeper > 70, automation > 75%)
2. 1,000+ clients within 24 months (2% SG market share)
3. Net revenue retention > 110% (expand from bookkeeping → AR/AP/tax)
4. Vendor memory provably creating switching costs (measurable)
5. PSG pre-approval to halve effective customer price