YBA Edge Foothold

Phased AI onboarding for SMB customers who want to start small, prove value, then ladder up.

3-specialist consensus, 2026-05-06. Captain met with a customer yesterday who said: “I’d rather see small benefits for my staff first — once they see the value, I’ll find it easier to invest.” This document is the design that captures those customers.

Verdict

Build it. R2,400/mo + R3,500 once-off onboarding. Name it “YBA Edge Foothold.”

This is not a profit tier — it’s a paid sales discovery rung. Volume + pipeline beats trying to sell R6k cold. Targeted economics: 10 Foothold customers convert to R24,600/mo recurring after 90 days, plus R35k onboarding once-off, vs. ~R12k/mo from converting 2/10 cold-sold R6k Office Brain prospects.

Pricing & ladder

R1,800 between rungs. No customer ever faces a R6k jump cold.

Onboarding fee R3,500 once-off is critical — covers VPS provisioning + 2hr discovery + tool connections, and prevents churn-after-setup eating the margin.

What’s IN Foothold (5 things, no more)

1. 09:00 daily Telegram brief per staff member — calendar + emails-that-need-you + commitments owed. Replaces ~20 min of inbox triage. This is the single biggest week-1 win — it’s the reason the whole tier works.
2. Email triage to Telegram — “you have 3 emails that need a reply today” with one-tap drafts.
3. WhatsApp / text → CRM logger — voice or text to a number, auto-files into existing Pipedrive / HubSpot / Sheets.
4. Friday weekly digest — what got done, what slipped, who was busy, who wasn’t. Whole-team email.
5. One custom n8n workflow per quarter — included. Their pick. The upsell hook.

What’s NOT in Foothold: voice agents, document RAG, approval bots, multi-user dashboards, Twilio outbound, Deepgram. Those are the upsell rungs.

The first 7 days for staff (this is what we sell)

• Day 1 (Mon): Captain installs the container, connects email + calendar + their existing chat tool. 5-minute walkthrough video — no training session, no workshop. Just “watch this, your phone will ping tomorrow at 9.”
• Day 2 (Tue) 09:00: First daily brief lands on every staff phone. Owner’s brief includes “pipeline at risk” — clients gone quiet >7 days. Staff start replying “can it also include X?” — that’s the signal it’s working.
• Day 3 (Wed): Brief now includes yesterday’s commitments (“you told Sipho you’d send the quote”) pulled from sent email. One staff catches a forgotten follow-up and saves a deal. They tell the owner.
• Day 4 (Thu): Add Telegram /find command — search across email + drive in one go. Stop tab-switching.
• Day 5 (Fri) 16:00: First weekly digest to owner. They see their business in one screen for the first time.
• Days 6–7: Brain stays quiet. Captain reviews logs, tunes noise out of next week. No staff contact.

By end of week 1: every staff member has opened Telegram before their inbox at least 3 times. That behavior change is what they’re paying for.

Discovery call script (15 minutes Captain can actually run)

1. “Walk me through your morning — first 30 minutes at your desk.”
2. “Which 3 tools does your team live in every day?”
3. “Who would be the first person to push back on AI, and why?” (We’re hiring this person as the canary.)
4. “If one task disappeared from your team’s week, which one?”
5. “Have you tried any AI tools before? What stuck, what didn’t?”
6. “What does ‘this is working’ look like in 30 days — for you, not for me?” (Get success criteria in their own words.)
7. Close: “I can have the daily brief running for your team within 5 working days. R2,400/mo + R3,500 onboarding, month-to-month, kill-switch any time. Want to start Monday?”

Yes / no / “let me think” (= soft no, move on).

Onboarding agreement (one page, plain English)

• Tools we’ll connect: [list]. Read-only access to start.
• Point of contact: one person, their WhatsApp number. Captain talks to them, not the whole team.
• Success criteria (theirs): 2 sentences in the customer’s own words from question 6.
• Cadence: daily brief 09:00, weekly digest Friday 16:00. Two 20-min check-ins (end of week 2 + week 4).
• Payment: debit order, monthly, first month pro-rata.
• Kill switch: 7 days written notice, container archived, data exported on request. No lock-in.
• What we will NOT do: send email on their behalf, post to social, touch financials — until explicitly added.

Expansion triggers (Foothold → Tier 2)

Two of the five = book the upgrade conversation.

• 3+ “can it also do X?” requests from staff in 30 days
• Staff using /ask command 10+ times/day across the team
• Customer asks to connect a 4th tool (CRM / accounting / WhatsApp Business)
• They’ve referred someone unprompted
• They ask about “letting the brain reply for me”

Technical architecture (per tenant)

Each customer gets a 3-container pod on the existing VPS:

Shared infra: Traefik HTTPS, one hardened Postgres (schema-per-tenant + RLS), Coolify orchestration.

Capacity: ~600 MB + 0.4 vCPU steady-state per pod. Current 32 GB / 8 vCPU box realistically holds 25–30 paying tenants before sharding to a second Hostinger node and routing by hash(slug) % 2.

Brain memory pattern: markdown wiki on disk per tenant (/srv/tenants/{slug}/vault/). Identical to Captain’s own brain. No vector DB at this tier — cheap BM25 retrieval via ripgrep is sufficient and feels magical because the vault compounds over weeks.

LLM routing: YBA-billed (not BYOK — BYOK kills the “it just works” promise). Default Haiku 4.5; brain auto-routes to Sonnet 4.6 on complex prompts (multi-step, comparative reasoning, >2k input context). Per-tenant LLM cost at 50 msgs/day: ~$9/month. Comfortably below floor.

Provisioning: provision_tenant.sh {slug} — one script, ~5 min compute + ~20 min tenant OAuth flow = onboarded. Later wrapped in n8n workflow triggered by ERPNext “contract signed” event.

Per-tenant URL: https://{slug}.tenants.ybaedge.com

Data isolation (defense in depth)

1. Filesystem: vaults at /srv/tenants/{slug}/vault, owned by tenant_{slug} UID. Brain container runs as that UID. No tenant can read another’s disk.
2. Network: per-tenant Docker bridge network. Only Traefik + shared Postgres reachable from inside.
3. Postgres: schema-per-tenant + dedicated role with USAGE only on its own schema. RLS as belt-and-suspenders on shared audit tables.

Separate Postgres-per-tenant is over-engineered at 30-tenant scale. One well-tuned Postgres + schema isolation is the right boundary.

Conversion math (10 customers, 90 days)

Realistic SA SMB assumptions:

• 3 upgrade to Tier 2+ within 90 days (30%)
• 2 churn in months 1–2 (20%) — usually because their tooling was too messy to integrate cleanly
• 5 stay flat at Foothold (50%)

Month 4 onwards revenue from 10 Foothold customers:

• 3 × R4,200 = R12,600
• 5 × R2,400 = R12,000
• Total: R24,600/mo + R35k onboarding once-off

Versus selling R6k Office Brain cold: realistically close 2/10 = R12k/mo. Foothold beats it on volume + pipeline + reduced sales cycle.

Risk to watch (the one Captain isn’t seeing)

Hidden support cost. R2,400/mo customers ask the same volume of “help me, my Telegram bot stopped working” questions as R6k customers — but at 40% of the revenue. 20 Foothold customers × 30 min/month each = 10 hours/month of unpaid ops drag.

Mitigations (codify in contract):

• Shared Telegram support group, not Captain’s DM
• Async-only Tier 1 support — same-day response, not real-time. Real-time SLA is Tier 2+ feature.
• Week 4 check-in is non-negotiable — Captain reviews open-rates on the daily brief. If <60% across the team, he ships ONE new automation that week. Engineer the journey to never let the customer feel they’ve seen the whole product.

Operational risks + mitigations

• Noisy tenant: CPU/RAM hard limits per container + Postgres statement_timeout = 30s per tenant role. One runaway loop cannot starve neighbors.
• Postgres SPOF: nightly pg_dump per schema to Backblaze B2, weekly full-cluster snapshot, hot-standby on a second Hostinger box before tenant #10.
• Vault corruption: vaults are git-init at provision time, brain commits after every write. Recovery = git reflog. Off-box: nightly rsync of /srv/tenants/ to B2.
• LLM provider outage: brain falls back to a cached “I’m degraded — Captain has been notified” response and pages Niki via Telegram.

Failure mode to engineer around

Months 2–3 churn driver: novelty wears off, staff stop opening the brief, owner stops seeing value, cancels.

Counter: The journey must always have a “what’s coming next” — never let the customer feel they’ve seen the whole product. Foothold is the appetizer, not the meal. The Week 4 check-in (mandatory) is the moment to ship the first automation the staff canary asked for. Use the customer’s own asks to design the upgrade path, not a fixed roadmap.

Build sequence (Niki, 2–3 days once approved)

1. provision_tenant.sh — full automation script
2. ybaedge/tenant-brain:v1 Docker image (FastAPI + Anthropic + vault tooling)
3. ybaedge/tenant-tgbot:v1 Docker image
4. Vault template — pre-filled CLAUDE.md, index.md, log.md skeletons
5. Traefik wildcard cert for *.tenants.ybaedge.com
6. Onboarding wizard page on ybaedge.com/foothold/onboard — handles the 4 OAuth flows
7. Foothold landing section on ybaedge.com with pricing + the discovery-call CTA
8. ERPNext “contract signed” → n8n workflow → provision_tenant.sh trigger

What Captain needs to approve to unblock

1. Price + name: R2,400/mo + R3,500 onboarding + “YBA Edge Foothold”?
2. Architecture: 3-pod tenant + schema-per-tenant Postgres + ~25-30 capacity?
3. LLM model: YBA-billed (Haiku default, Sonnet escalate), not BYOK?
4. First customer: the customer from yesterday’s meeting — or do we build the infrastructure first and onboard them as customer #1 next week?