THE DECISION
Who provides the SME?
This single choice determines
pricing, margin, compound speed,
and what kind of company you become.
The transition: copilot → autopilot
Each phase = less SME, more system. Each SME correction pushes the threshold.
Phase 1
40% autonomous
Drafts, executes routine tasks, handles data pipeline
60% copiloting
Reviews, corrects, applies domain judgment. Every correction = training signal.
High touch
Phase 2
75% autonomous
Handles most workflows. Knows the domain patterns. Flags only edge cases.
25% oversight
SME handles exceptions and judgment calls only.
Low touch
Phase 3
95% autonomous
Full autopilot. Today's judgment is now intelligence in the system.
5% audit
Spot-check. SME now free to onboard new verticals.
Autopilot
The mechanism: Each SME correction is a labeled training signal. The system doesn't add rules — it subtracts wrong defaults.
The SME's domain expertise is progressively encoded into the system. The copilot becomes the autopilot through reps, not through better models.
Three models for who provides the SME
A
Client's SME
PLATFORM MODEL
How it works
- Client already has the domain expert internally
- Deploy the AI system as an instance inside the client
- Client's own SME does the copiloting day-to-day
- SME generates the reps that train the system
- Charge: platform fee + setup + ongoing support
- SME cost = zero on your P&L
Trade-offs
- ✅ Margin starts higher than 30% (no SME cost)
- ✅ Scales faster — no hiring bottleneck
- ✅ Lower barrier to entry for first clients
- ❌ No control over rep quality
- ❌ Weak SME = weak training = slow Phase 2
- ❌ System quality depends on the client's people
- ❌ Compound speed varies wildly per client
B
Your SME (outsourced, per period)
SERVICES MODEL
How it works
- Hire SME via outsourcing (LatAm, nearshore) per domain
- Allocate to client during Phase 1
- SME copilots the AI, generates high-quality reps
- When system transitions to Phase 2, reallocate SME to next client
- Same SME serving multiple clients in the same domain = compound²
- SME cost: ~$14K/month (nearshore rate)
Trade-offs
- ✅ Full control over rep quality
- ✅ Guaranteed compound — training is consistent
- ✅ The SME IS the asset — compounds alongside the system
- ✅ Client 5 starts near Phase 2 because SME already trained the domain
- ❌ Capital required upfront to fund SMEs before revenue covers cost
- ❌ Cash flow pressure on first 2-3 clients
- ❌ Hiring = bottleneck on new verticals
C
Hybrid — Your SME seeds, client's SME scales
HYBRID MODEL
How it works
- Phase 1 (weeks 1-4): Your SME does setup + initial reps
- Controls quality of the most valuable training signals (first reps)
- Phase 1→2 transition: Client's SME takes over daily copiloting
- Your SME does periodic QA on rep quality
- Phase 2→3: Your SME audits only. Reallocated to new client
- SME cost is temporary — customer acquisition cost, not operating cost
Trade-offs
- ✅ Controls training quality where it matters most (initial reps)
- ✅ SME cost exits P&L after setup period
- ✅ Client 5+ gets faster onboarding (domain already seeded)
- ✅ Best margin trajectory of all three models
- ❌ More complex to operate (handoff protocol needed)
- ❌ Risk: client's SME degrades rep quality after handoff
- ❌ Needs QA layer to catch quality drift
Compound speed per model
Client onboarding → Phase 2
Client 1
Client 3
Client 5+
Model A
Depends on client's SME quality. Unpredictable.
6-12 mo
4-8 mo
3-6 mo
Model B
Controlled. SME compounds across clients in same domain.
4-6 mo
2-3 mo
~1 mo
Model C
Seeded start + client continuation. Best of both after handoff.
3-5 mo
2-3 mo
~1 mo
What each model makes you
MODEL A
Platform company
You sell the system. Client operates it. You're Salesforce — high margin, low control, dependent on client capability. Revenue: SaaS-like recurring.
MODEL B
Services company
You do the work. You own the reps. You're the next Accenture — but with AI margins. Revenue: outcome-based, expanding margin per client.
MODEL C
Hybrid: seed & scale
You control the critical training, then transfer operations. You're the company that makes autopilots — not the company that flies planes. Revenue: setup + platform + gain-share.
This is a founder decision. It determines capital requirements, hiring strategy, margin structure, compound speed, and the type of clients you can serve. Every downstream choice — pricing, team size, vertical selection, fundraising — flows from this one call.