The insight (from 3 months operating Donna): Full-context CRM enables instant quality replies for servicing. Codified expert workflows make delivery automatic via LLMs. Together these collapse the cost of both service and delivery in any field sales organization.
The product: A purpose-built audio capture device as the data wedge — capturing in-person conversations that don't exist digitally. Donna/mufu as the intelligence engine — contextualizing, servicing, and delivering from that data.
The architecture:
| Layer | Function | Already Proven |
|---|---|---|
| CAPTURE | Device records field conversations | bob-1 WA: 1,000+ msgs captured. Farmio SG: 5 WA-connected devices agreed for sales team. Audio recorder piloted in Bob’s client meetings. |
| CONTEXTUALIZE | Full relationship history per client | Donna tracks 100+ relationships with warmth, cadence, action items. Daily. |
| SERVICE | Instant quality replies, proactive follow-ups | Bob servicing pivot (Mar 18): upsell/renew = 10-100萬/client. |
| DELIVER | Codified workflows produce actual work products via LLM | Malaysia franchise data pack, CV screening, brand books — all auto-generated from briefs. |
Virtual call intelligence (Gong) can go global because a Zoom call in English is a Zoom call anywhere. Field sales AI cannot, for five structural reasons:
Field sales in HK is Cantonese, Cantonese-English code-switching, or Mandarin with Canto particles. Bob's team meetings are mixed code. Siro's models are trained on English-language home services pitches — door-to-door roofing in Ohio. Transcription accuracy for a Cantonese insurance agent at a client lunch in Wan Chai would be unusable. Whisper handles Cantonese, but downstream analysis (objection detection, coaching, sentiment) requires language-specific fine-tuning that US players haven't built and won't prioritize.
US field sales (Siro/Rilla's market): appointment-set, pitch-driven, script-adherence coaching. HK field sales: relationship-driven. 飲茶 meetings. WhatsApp voice notes. Introductions through mutual connections. The coaching model isn't "did you follow the script" — it's "are you maintaining the relationship cadence?" That's exactly what Donna already does.
Siro integrates with ServiceTitan. VoiceLine integrates with SAP/Salesforce/Dynamics. Bob's CRM is 18 Google Sheets and a WhatsApp group. HK SMEs don't run Salesforce — they run WhatsApp + Excel + maybe a POS. The integration layer for Asia field sales has to be WhatsApp-first, not Salesforce-first. Donna is already WhatsApp-native.
Siro sells through US industry platforms (ServiceTitan partnerships). They have zero channel into HK insurance agencies or F&B marketing firms. Even if Siro wanted Asia, they'd rebuild distribution from scratch — a 2-3 year effort minimum.
Rilla: $4K+/user/year, $20K minimum. Siro: enterprise pricing. HK SMEs think "cheaper than one junior hire" (HK$15-20K/month for the whole team). The value bar and price sensitivity are fundamentally different markets.
| Company | Funding | Geography | What They Do | Asia Threat |
|---|---|---|---|---|
| Siro $50M SERIES B |
$50M+ (SignalFire, CRV, Index) | US only | In-person sales recording, coaching, analytics. 36% close rate improvement. Hundreds of customers.1 | Low today. All customers US home services/wholesale. No Asia plans disclosed. Language/culture moat protects HK. |
| Rilla Voice | $3.7M seed | US only | Field sales speech analytics. $4K+/user/yr. Home services.3 | None. US roofing/HVAC only. |
| VoiceLine | €12.4M total | Europe only | Voice → CRM for field sales. ABB, Knauf clients.4 | None. Explicitly scaling "across Europe." |
| Catch AI | Unknown | US only | $250/rep/mo. Home services recording.5 | None. |
| Listel | Unknown | US (likely) | $49 wearable for field sales + CRM sync.2 | Low. Generic hardware. No Asia localization. |
Plaud launched a Developer Platform with CRM JSON output. Their public roadmap includes Team Collaboration, API Integration, and Zapier/n8n connectors.6 Early developer customers report 3x enterprise deal closures. This is real and happening now — not "eventually."
But: Plaud's B2B layer will be horizontal (generic team dashboards, generic CRM sync). It won't include relationship intelligence, servicing automation, or delivery generation. It's the same gap as Siro vs. Donna — they capture and maybe analyze, they don't service and deliver.
| Player | What | Threat Level |
|---|---|---|
| Gong (SG office) | Virtual call intelligence. $100-160/user/mo. Enterprise only. | Low — different product (virtual, not field). Wrong price for SMEs. |
| Swiftlet.ai | Conversation intelligence for in-person. Unclear traction. | Watch — closest to what we're building. Need deeper investigation. |
| Donna (activatedonna.com) | EU field sales AI. 40+ languages. Stanley/Stella, DPG Media.7 | Low in Asia. Name conflict for branding. |
| The real incumbent | Excel + WhatsApp + memory | This is what we're actually replacing. |
| Market | Size | Source |
|---|---|---|
| Conversation Intelligence | US$1.77B (2025) → $3.47B (2032), 10.3% CAGR | QY Research8 |
| Sales Enablement Platforms | US$5.23B (2024) → $11.3B (2030), 16.3% CAGR | Grand View Research9 |
| Plaud (consumer hardware proof) | $250M ARR, 1.5M devices, profitable | Sacra10 |
| Segment | Size | Source |
|---|---|---|
| HK insurance agents | 78,786 licensed (Sep 2024) | Insurance Authority11 |
| HK real estate agents | ~42,000 licensed | EAA estimates |
| HK F&B/marketing agencies | ~5,000+ firms | Census & Statistics Dept |
| SG field sales (Wenhao market) | ~200,000 sales professionals | MOM estimates |
The number that matters: You don't need market share. You need 50 users. 10 teams of 5 reps at HK$3K/user/mo = HK$150K/mo = HK$1.8M/year. That's already meaningful against Eric's HK$1M recurring target. 100 users doubles it.
| Component | Cost | Assumption |
|---|---|---|
| Audio transcription (Deepgram) | HK$12 | 600 min/mo @ $0.0043/min12 |
| LLM — summaries (GPT-4o-mini) | HK$14 | 20 meetings × 3K in + 1K out tokens13 |
| LLM — coaching + delivery (Sonnet) | HK$94 | Coaching analysis + deliverable generation |
| Cloud/storage | HK$15 | ~2GB audio/mo per user |
| Donna engine (amortized) | HK$200 | Context engine, warmth scoring, briefings |
| Hardware amortization | HK$6 | B2B-sourced device ~$15 over 24 months (vs. Plaud $179 retail) |
| Support/ops (at 50+ users) | HK$150 | Amortized customer success |
| Total COGS | HK$491 | ~US$63/user/mo |
67% of sales professionals feel AI coaching tools are surveillance, not support15. Reps quietly stop using recording tools for important calls. Field reps resist more than inside sales because they're more autonomous.
Mitigation: The wearable form factor (vs. a visible meeting bot) eliminates "bot embarrassment." But the pilot MUST prove value to the rep first — time saved, deals closed — before proving value to the manager. If Veaky/Crystal don't use the device voluntarily after week one, the thesis breaks.
Donna doesn't eliminate work — the Malaysia franchise pack took hours of Eric's research + prompting + verification. At personal scale, Eric absorbs that labor. At team scale, someone has to do it for every client. The "collapse" may just move cost from rep salary to AI ops.
Mitigation: True for bespoke research. False for repeatable workflows (client reports, follow-up emails, meeting summaries). Bob's Veaky Report Step .xlsx is a repeatable workflow — automate that first, not bespoke deliverables.
Plaud's Developer Platform is live. Team Collaboration is on their public roadmap6. Window may be 6-12 months.
Mitigation: Plaud will build horizontal B2B features (generic dashboards). They won't build Cantonese-specific coaching, WhatsApp-native CRM, or industry-specific delivery templates. The intelligence layer is the moat, not the hardware.
HK PDPO requires informing the other party before recording — but does NOT require explicit opt-in consent16. Simple disclosure ("this meeting is being recorded") is legally sufficient. Insurance/finance sectors may have additional compliance requirements.
Mitigation: Manageable. Standard practice in financial services already. Build disclosure into the meeting flow.
Five people named, zero committed. Mitchell is a renewed contact. Wenhao has a SG kill date Apr 15.
Mitigation: Don't recruit before validation. Run Bob pilot first (2 weeks, 1 device). Results → Wenhao pitch → Bosco product input → Mitchell outbound. Sequential, not parallel.
| Capability | Donna (Running Today) | Siro/Rilla/VoiceLine |
|---|---|---|
| Full conversation history across channels | WA + in-person + email + calendar | In-meeting only. No between-meeting tracking. |
| Relationship warmth scoring | 75+ contacts scored with cadence | Deal scoring (Gong). Nobody does relationship health. |
| Auto-generated deliverables | Franchise data packs, CV screening, brand books | Nobody does this. They coach. They don't deliver. |
| Daily manager briefings | PCRM loop: what happened, what's next, what's at risk | Dashboards exist. Nobody does relationship-aware daily briefings. |
| WhatsApp-native | Built on WA from day one | Zero WA integration across all competitors. |
| Cantonese/Mandarin context | Eric operates in mixed code daily | English-trained models. |
The thesis is no longer theoretical. Two pilots are active with paying (or payment-committed) stakeholders across different verticals and geographies. Both validated independently within the same week (Mar 25–27, 2026).
| Dimension | Detail |
|---|---|
| Who | Bob — founder of Hopeman Group, R&B marketing agency. 216+ F&B clients across HK. Deep SMB network. |
| Deal | 3-month trial (~$21K HKD/mo + token reimbursement). Kickoff completed Mar 25 at Hopeman office. |
| What he sees | Bob can command his sales team to use it — adoption is top-down. Bob himself tracks everything already. He sees this as a force multiplier, not a behavioral change. |
| Value signal | “Servicing development >>> personalized SOP.” Bob frames the value as catching emotional moments — data spikes, going-cold signals, upsell triggers — and routing them to the right person. Revenue impact: upsell/renew = 10–100萬 per client. This is revenue acceleration, not cost savings. |
| New product potential | Bob sees meeting prep + account servicing intelligence as new product lines within his agency offering. Every client meeting generates structured deliverables → faster turnaround → higher perceived service quality → grounds for premium pricing. |
| Channel potential | HK F&B CHANNEL — 216 existing clients. If the tool works for Hopeman's internal team, Bob becomes a distribution channel into HK F&B SMEs. He has the relationships; we have the product. |
| Status | Two pilots agreed: (1) meeting prep/followup for 3 sales staff, (2) account servicing intelligence (data in hand, building now). |
| Dimension | Detail |
|---|---|
| Who | Paco Chan — founder of Farmio, food supply chain SaaS. 1,000+ B2B customers, ~$6M revenue across Asia. Eric’s past cofounder. |
| Deal | Pilot scoped Mar 27. 5 WA-connected devices agreed for deployment to Singapore sales team. |
| What he sees | Farmio already has WA AI (automated ordering, action locking, context-from-chat). The pilot is augmenting their existing tech stack, not replacing it. Integration path is clear because they’re already WhatsApp-native. |
| Value signal | Field sales teams hate logging to CRM. Farmio's reps (4-5 SG, 2 HK) manage ~250 customers per office. Current CRM logging is manual and resisted. Capturing activity from WA data means reps get value (follow-up automation) without behavior change (no manual logging). The attendance tracking alone — knowing who visited which client, when — is immediate value. |
| Channel potential | SG SUPPLY CHAIN + HK CORPORATE — Paco operates across SG, HK, and expanding. Farmio’s customer base is B2B supply chain — different vertical, same field sales pain. If the pilot generalizes, Paco becomes distribution into SG supply chain and HK corporate sectors. |
| Status | Three-pillar pilot agreed: (1) sales attendance from WA data, (2) per-conversation follow-up automation, (3) market insights from cross-conversation analysis. Sequenced by complexity. |
| Week | Action | Success Metric |
|---|---|---|
| Week 1 | 1 Plaud NotePin to Bob or Veaky. Record one real client meeting. Auto-generate meeting summary + follow-up email + first-draft deliverable from transcript. | Was the auto-generated output usable? Did the rep find it valuable (not just Bob the manager)? |
| Week 2 | Automate Veaky's Report Step .xlsx — pull multi-platform data, generate client report from captured context. | Time saved vs. manual process. Measurable before/after. |
| Week 3-4 | Expand to 3 devices. Capture → service → deliver loop running for multiple clients. | Rep voluntary usage (not forced by Bob). Client feedback on deliverable quality. |
| Month | Action | Revenue Target |
|---|---|---|
| Month 1 | Bob pilot (case study) | Free or HK$10K/mo |
| Month 2 | Pitch Wenhao conversion (same arch, better market). Show Bob results. | — |
| Month 3 | Chaky as 2nd customer (insurance). Bob as case study. TVP application. | HK$15-30K/mo |
| Month 4-6 | Bosco product input. Mitchell outbound. 3-5 teams. | HK$50-100K/mo |
| Month 6+ | SG via Wenhao. HCMC via Vincent (if raise). Productize repeatable workflows. | HK$150K+/mo |
| Person | Role | Why | When |
|---|---|---|---|
| Wenhao KEY | Co-builder / SG | SG SMB access. Sales background. Kill date Apr 15 = natural decision point. | Month 2 (after Bob results) |
| Bosco | Product / Sales Ops | K-Pay field sales ops experience. Knows manager pain firsthand. Interested in mufu. | Month 3 (product advisory first) |
| Mitchell | Outbound / Lead Gen | Sold qualified leads. His own lead-to-conversion gap IS the problem. In transition. | Month 4 (needs case study to sell) |
| Bob | Pilot + HK SME Channel | 216 clients = expansion pipeline. Already champion. | Now (pilot) |
| Vincent | SEA / Fundraise | HCMC connections. Incorp Asia (corporate services = field sales intensive). | Month 6+ (if raise) |
Strong — thesis validated across two verticals. Device economics dramatically improved. Channel partners emerging from pilot customers.
Since v2, two things changed: (1) early validation from two independent pilots (Bob/Hopeman in HK F&B, Paco/Farmio in SG supply chain) confirms the architecture generalizes, and (2) device sourcing at 1/10th Plaud retail collapses the hardware barrier for B2B deployment.
The competitive landscape remains the same — Siro ($50M), Rilla, VoiceLine are US/EU only. Asia has zero funded players. The "deliver" layer (auto-generating work products from captured conversations) remains genuinely differentiated. Both Bob and Paco independently validated that servicing quality and CRM-bypass are the primary value drivers — not coaching.
The path:
What would change the verdict: