The operating system for e-commerce brand operators.
Terminal6 is the agentic OS that replaces the matrix of specialists and point tools every growing brand bolts on. We assemble full brand context across channels, markets, and tactical decisions — and deploy AI agents that reason and act across the whole loop.
1 Bain × Flipkart, How India Shops Online 2025. Projected $170–190B by 2030. 2 Estimated from Inc42 FAST42 2025, 3one4 Capital, DSG Consumer Partners; reconciled with Unicommerce IPO RHP (Aug 2024): 795 enterprise + 2,707 SMB clients. 3 Derived: Amazon 2024 Small Business Report (55K+ US sellers at $1M+) × ~8% at $5M+ ≈ ~4.4K. Shopify (47K Plus merchants globally, ~40% US, typical range $1–10M) adds ~9.5K US at $5M+. Minus overlap ≈ ~10K unique US brands.
Three shifts just converged.
None of the ingredients existed together 24 months ago. All three do now. Nobody has assembled the product yet.
LLMs made cross-silo reasoning cheap.
Claude/GPT-class models produce senior-operator-quality diagnoses at ~$25/brand/month of inference. The economics only work now.
The same product works globally on Day 1.
Shopify + Amazon are the two channels every ambitious brand runs — in India, the US, and everywhere. A $10M brand on Shopify + Amazon in Bangalore has the same decision loop as one in Austin. One product, two markets, from the start.
India has a demand-intelligence whitespace.
Unicommerce, Increff, EasyEcom solve supply — orders, inventory, fulfilment. No Indian player solves demand — the ads × conversion × competitive × tactical layer where brands actually lose money. Greenfield.
Running a brand is a broken decision loop.
Every operational call — raise a bid, pause a campaign, reorder, reprice — moves through the same three stages. Today, every stage is broken.
Observe
Context never assembles. Internal data in 4–6 tools. External data (competitors, search, calendars) scattered. Tactical calls in Slack and standups — nothing captures them.
Decide
Interdependent variables humans can't hold. A conversion drop is inventory × price × delivery × competitor × fatigue — two or three compounding. Rule-based tools hit a ceiling here.
Act
Execution across disconnected tools + people. By the time "pause ads on OOS SKU" lands in the ad tool, 2–3 days of spend is burned on a dead SKU.
Rebuild the loop as one system.
Six layers. Each layer fixes one stage of the loop. Not a dashboard, not a copilot — an operating system where agents work as employees.
Own the data, not the SaaS. Every recommendation, action, override, and outcome stored as a structured decision record — the compounding asset no incumbent holds.
Six specialists, one shared brain.
Not one mega-agent — six focused specialists, each modelled on a human role in a brand team. Each reads shared brand context, reasons through its own playbooks + category cards, and coordinates with the others through structured messages. Phase 1 brings all six online across 10 Indian brands before we scale.
Growth Agent
Watches ROAS, CPC, CTR, TACoS, creative fatigue across Amazon, Google, Meta, Flipkart ads
Acts on Bids, budget reallocation, campaign pauses, creative rotation, throttle signals
Supply Agent
Watches Inventory across FCs, PO statuses, lead times, fulfilment velocity, cover days
Acts on Restock alerts, PO triggers, FC rebalancing, stockout prediction, reorder points
Category Agent
Watches Competitive pricing, BSR, discount calendars, category launches, market share
Acts on Pricing recommendations, discount strategy, competitive alerts, assortment calls
Catalog Agent
Watches Listing quality, keyword rankings, A+ content, rating trends, content compliance
Acts on Listing updates, keyword optimisation, content refresh, review response triggers
Finance Agent
Watches Settlements, fees, GST, working capital, contribution margin per SKU, cash cycle
Acts on Fee reconciliation, cash forecast, P&L alerts, tax compliance, SKU-level P&L
CX Agent
Watches Review sentiment, return reasons, complaint patterns, NPS, delivery experience
Acts on Review response, return root-causing, quality alerts, category team escalations
All six agents are scoped in the canonical schema today. Data layer (L1 + L2) live against Sprig. Growth playbooks written (conversion_drop full, oos_response / roas_drop / low_cover / high_cover skeleton). L3 agent runtime is the core Phase 1 engineering milestone — ship all six agents with working playbooks by Month 6, stress-tested across 10 brands in production.
Each fixes one slice. Nobody converges.
| Category | Fixes | Structural gap |
|---|---|---|
| India OMS Unicommerce, Increff, EasyEcom, Vinculum | Orders, inventory, warehouses, fulfilment | Supply-side only. No demand view — blind to ads, conversion, competitive, tactical decisions. This is the India whitespace. |
| Ad tools Pacvue, Intentwise, Adbrew, AiHello | Ad bids & budgets, one channel at a time | Demand-side only. Blind to inventory, margin, fulfilment. Rule-based — can't reason across interdependent signals. |
| BI / dashboards Metabase, Looker, Domo, internal SQL | Historical reporting | Passive. No decisions, no memory, no actions. Analytics ≠ operating system. |
| AI copilots Horizontal chat, vertical copilots | Natural-language Q&A on data | No state, no memory, no execution authority. A faster search box is not an OS. |
The shape nobody has built. India has mature supply-side tools (Unicommerce IPO'd at ~$130M valuation) and imported demand-side ad tools. Globally, the same story — strong point tools on each side, no unified demand + supply intelligence layer that reasons across both. Each incumbent is structurally anchored to one side of the house; reshaping into a cross-function agentic OS would cannibalise their existing business. We start from the shape the operator actually needs.
Operator expertise, captured and compounded.
The pipeline is not the moat — connectors get commoditised. The moat is a four-layer knowledge system that makes the same playbook produce the right diagnosis across categories, regions, and brands.
The flywheel. Every brand contributes learnings that graduate into category and region cards. The next brand in that category is smarter from Day 1 and cheaper to onboard. No ad tool, OMS, or BI product stores decisions, reasoning, or expertise as structured data — we do.
Mid-market first, aggregators for scale.
Primary ICP
$5–50M GMV brand, fast-growing, team not scaling proportionally. 50–500 SKUs. Shopify + Amazon as core channels (with secondary marketplaces in India — Flipkart, Meesho, Myntra — supported). Ops team of 3–10, founder in weekly decisions. Matrix pain is acute; they know they need leverage. India and US pursued in parallel from Day 1.
Not ICP (yet)
Sub-$5M brands (pain not acute enough, budget too tight). Single-channel Shopify-only brands (too simple). >$100M GMV enterprise (different motion — fight after the moat compounds).
Motion: Land with unified dashboard + morning briefing (lowest-risk wedge). Earn trust over 60–90 days. Expand into approval-based execution, then autonomous execution under policy, then multi-agent coordination. Same product sold into both geographies simultaneously — founder's India network + warm US intros via Opptra & Flipkart alumni.
Matrix → functional. The economics work from day one.
Terminal6's value isn't replacing one analyst. It's changing the organisational shape of the brand — from a function × channel matrix where humans are the connective tissue, to a lean functional team where agents absorb the per-channel work.
Brand saves money Year 1. Pricing scales with value delivered.
The ACV ladder below is directional. Pricing flexes by brand size, geography, and value delivered — US brands typically trend toward the upper half of each band (higher stack costs, more expensive ops teams to replace); India brands toward the lower half. The savings vs existing stack hold in both geographies.
| Brand GMV | Terminal6 ACV | Existing stack cost (tools + 1–2 ops FTEs, annual) | Savings |
|---|---|---|---|
| $5–10M | $2K/mo · $24K/yr | $50–120K | 2–5× |
| $10–25M | $4–6K/mo · $48–72K/yr | $100–240K | 2–3× |
| $25–50M | $8–10K/mo · $96–120K/yr | $200–500K | 2–4× |
| Aggregators 50+ brands | Per-brand 3–10× lower, amortised across portfolio | Multi-million stack + margin drag | 10×+ |
The $2–10K/month ladder is directional — the range we believe the market will settle into. Actual ACV flexes per customer, geography, and value delivered; India may land on the lower half of the band as we learn from early customers.
Unit economics
- Inference: ~$25/brand/mo (Haiku routine, Sonnet complex)
- Infra: ~$110/brand/mo (RDS, S3, connectors)
- 80% of decisions handled deterministically; only novel 20% hits Sonnet
Why the brand pays
- Replaces 1 analyst FTE + 2 point tools
- Cost takeout alone justifies the ACV
- Margin + revenue upside is pure upside
- Same pricing globally — no "India discount"
~76% gross margin at launch. Path to ~85% at scale.
Three variable costs shape the economics: LLM inference, infrastructure, and customer success. LLM is the largest variable cost and the single biggest engineering lever. Fully-loaded margins sit in top-quartile SaaS territory from Day 1 — without claiming the 90%+ pure-variable numbers that would get discounted by any experienced VC.
Engagement volume — Terminal6 is an OS, not a dashboard
An ops team of 5–10 people interacts with the system throughout the day: founders reviewing briefings, growth leads drilling into ads, analysts running ad-hoc queries, supply leads planning reorders. Background agents fire continuously for anomaly detection, policy checks, and cross-channel RCAs.
| Source | Calls/day | Model mix |
|---|---|---|
| Interactive — users asking, drilling, planning (5–10 person ops team) | ~150 | Mostly Sonnet |
| Per-channel RCAs — auto-diagnoses across 5 channels | ~20 | Sonnet |
| Background agent work — monitoring, alerts, policy checks, coordination | ~100 | Mostly Haiku |
| Morning briefing + deep planning (big-context synthesis) | ~10 | Big Sonnet |
| Typical brand | ~280–320 / day | 50% Haiku · 45% Sonnet · 5% Big Sonnet |
Fully-loaded cost per brand per month
Includes customer success and ops overhead amortised across the book — honest numbers, not pure variable cost.
| Cost line | Launch | Mature (Y2) | Scale (Y3+) |
|---|---|---|---|
| LLM inference (Claude) | $460 | $290 | $140 |
| Infra (RDS, S3, Celery, connectors) | $120 | $100 | $80 |
| Customer success (amortised) | $300 | $220 | $150 |
| Ops overhead (DevOps, monitoring, security) | $150 | $120 | $90 |
| Total fully-loaded COGS | ~$1,030 | ~$730 | ~$460 |
Gross margin by ACV tier
| Brand GMV | ACV/mo | Launch GM | Mature GM | Scale GM |
|---|---|---|---|---|
| $5–10M | $2,000 | 72% | 78% | 85% |
| $10–25M | $4,000 | 77% | 83% | 87% |
| $25–50M | $9,000 | 79% | 84% | 86% |
| Blended | ~$5K | ~76% | ~82% | ~85% |
Customer success cost scales with brand size — bigger brands need more hands-on onboarding, tuning, and check-ins — which is why the ceiling tier GM doesn't dramatically exceed the middle tier. Matches high-touch enterprise SaaS patterns.
Three structural levers
- Prompt caching (engineering discipline). Stable prompt portions — system prompt, playbook, category card, brand profile — cached at 10% of input cost, reused across calls. 80% cache hit rate reduces total LLM cost by ~37%. Requires deliberate architecture: bad caching is worse than none (cache writes cost 25% premium).
- Playbook determinism (product maturity). Phase 1 playbooks cover edge cases with Sonnet-heavy reasoning. As playbooks mature over Phase 2–3, more decisions become deterministic Python. Sonnet share drops from ~45% to ~25%, halving inference cost without quality degradation.
- Claude price drops (market tailwind). Inference pricing has halved every ~18 months historically. Not a lever we control, but a consistent structural tailwind that passively brings LLM cost to ~$140/brand/month by 2027–28.
Top operational risk: LLM cost trajectory. If Claude price drops stall, cache hit rates don't reach 80% in production, or usage is 2× heavier than modelled, floor-tier GM compresses to 55–65%. The business still works at those margins but becomes less compelling.
Mitigation: continuous per-brand LLM monitoring; soft limits on power users; pricing flexibility (the $2K/mo floor can shift upward — the ladder is directional, not locked). This risk gets monitored in every board deck.
Day 1: we plug in. Over time: we absorb the stack.
The Day-1 ACV ladder is the entry point, not the ceiling. Every tool in a brand's stack has a price we can absorb as our agents take over its job — starting with marketing tools (which charge 2–5% of ad spend globally). The end-state is Terminal6 pricing against GMV itself, at 1–2% of the brand's online revenue, as the full commerce operating stack.
Intelligence layer
+ Marketing execution
+ Operational tools
Full commerce stack
Why this is inevitable. Once Terminal6's agents are making the decisions, the tools below become commoditised plumbing. A brand running Pacvue alongside Terminal6 is paying twice — once for the brain, once for the hand. By Year 2, ripping out Pacvue and letting Terminal6 execute directly is obvious. By Year 3, the same logic applies to BI, catalog, and operational tools. The pricing evolution tracks what we actually deliver — flat ACV → % of ad spend → % of GMV — not aggressive repricing.
Worked example — $10M brand. Day 1: $24K. Year 2 (+2% of $1M ad spend): ~$44K. Year 3–4 (+0.5% of GMV ops tools): ~$94K. Year 5 (1.5% of GMV as commerce stack): $150K. Roughly 6× LTV expansion per brand without acquiring new customers. CAC stays flat; LTV compounds. This is how Terminal6 becomes a category-defining company, not a point solution.
One market done right, then two markets at scale.
Phase 1 is India-only — 10 brands, full agent suite, 3–6 months, production-grade. We stress-test every agent use case in real brands before we touch US distribution. Phase 2 flips on US in parallel once the system is robust. No Year-1 international overreach.
Depth in India
10 brands across 3–4 categories. Full agent suite live: growth (ads across Amazon + Flipkart + Meta + Google), supply, category, catalog, finance, CX. Objective: stress-test every agent use case in production. Sprig + Basil locked; 8 more through founder network.
India + US in parallel
30–50 brands, split India + US. Scale India to 30+ brands. Onboard first 10–15 US brands via Flipkart / Opptra alumni network. Ship L4 policy engine + L5 execution. Start capturing marketing-tool budgets (Nervous System Stage 2).
Category definition
100+ brands, first aggregator deals. Aggregator accounts (one deal = 50+ brands) kick in. Full 6-layer OS live. Knowledge system compounding — each new brand cheaper to onboard than the last. Nervous System Stage 3 begins.
Why depth before breadth. Running 5 brands badly teaches us nothing we don't already know. Running 10 brands with all 6 agents stress-tests every failure mode — the system breaks, we fix it, ship the fix, and onboard #11 smarter. The agents learn as the brands do. Only when Phase 1 is production-grade do we spend founder time on US distribution — because we'll only get one chance to set US customer expectations, and we're not spending that chance on a half-baked product.
Live data, validated findings, working pipeline.
Validated findings from Sprig's 90 days of data
For the technically curious.
Full 6-layer OS, 16-entity canonical schema, 6 throttle signals, agent roster, policy engine — same document used with AI architecture reviewers and engineering hires.
Built parts of Terminal6 before it existed.
Nitin Chaudhary — 17 years across e-commerce & fintech, spanning category, operations, finance, and technology.
At Opptra (Binny Bansal's venture, current), launched a multi-marketplace fashion business across Amazon AE, Noon, Namshi — scaled three brands to $3M ARR in six months at +15% CM. Built two AI products in production there — a bid-optimisation engine and an inventory replenishment engine — the direct genesis of the Terminal6 thesis. At Flipkart, led planning for categories contributing $2.5Bn GMV and 45% of volumes, grew share 42% → 48% vs Amazon, ran Business Finance for Fashion ($1Bn GMV, 800bps profit), scaled BNPL to 5M MAU.
Founder-led today. Design partners Sprig and Basil Home Solutions providing live data and weekly feedback. First engineering hire planned post-seed.