TIER S BLUEPRINT

Lithuanian Support Triage + Draft Replies

Triage Lithuanian tickets and draft replies your agents actually send - LT quality that generic tools can't match, grounded in your policies.

from €1,990

HELPDESK - LT QUEUE · LIVE TRIAGE

Incoming tickets

#LT-8841 · prieš 2 min

"Sveiki, kur mano užsakymas #7712? Turėjo atvykti vakar…"

INTENT: order status · 0.96→ DRAFTED

#LT-8842 · prieš 1 min

"Reikalauju kompensacijos, nes dėl jūsų kaltės praradau klientą…"

INTENT: compensation claim · 0.91⛔ OUT OF POLICY → HUMAN

Draft reply - #LT-8841 · grounded

"Sveiki! Jūsų užsakymas #7712 išsiųstas liepos 12 d. ir šiuo metu yra pristatymo centre Kaune ⌕ Omniva API. Numatomas pristatymas - rytoj iki 17 val. Jei siunta neatvyks, grąžinsime siuntimo mokestį §4 Pristatymo taisyklės."

grounding ✓ 2 sources PII redaction ✓ tone: your brand ✓

🟡 AGENT APPROVES & SENDS - one click, or edits first. Never auto-sent.

#LT-8842 ▸ no grounded answer in policy → DRAFT BLOCKED → senior agent queue · logged ⛓

today ▸ 78% drafted & grounded · first response <1h · every draft logged

How it works

Connect the helpdesk

Zendesk, Freshdesk, Intercom - read-only to start. Intent taxonomy designed on your actual ticket history, in Lithuanian.

Triage with confidence

Every ticket classified and routed; uncertain intents go to a human, not a guess. Priority and SLA rules applied automatically.

Draft replies that cite sources

RAG over your policies and order data - every claim in the draft links to where it came from. Ungrounded drafts are blocked, not sent.

Agents approve, the model learns

Accept-or-edit per draft; edits feed back. Coverage grows as your taxonomy does - and we report the covered % monthly.

The Four Guarantees™ - this build

Measured value

~30% handle-time cut at reference volume; intent-F1 + groundedness + human-rating eval gate before ship.

Defensible

Routing reason + citations behind every draft; PII redaction; GDPR audit log; Art. 50 AI-assistance disclosure built in.

Self-correcting

Confidence gates uncertain drafts to humans; new-intent and topic drift monitored; draft-acceptance rate feeds the loop.

Yours & everywhere

Your helpdesk, your cloud, full source. MCP endpoint for support agents and copilots.

The number, sized honestly

Reference buyer: Lithuanian e-commerce, SaaS or utility - ~5,000 tickets/month in Lithuanian, ~3 agents, ~15 min handle time, ~4h first-response SLA.

~30% handle-time cut ≈ 2 FTE - the agreed number (conservative)

€60-80k per year in direct labour at this volume

<1h first response on covered intents (from ~4h)

~1.5 mo payback at the PoC tier

Three ways to own it

Tier	What you get	Price
Scaffolding ★	The full repo - LT intent + groundedness eval harness, RAG pipeline, audit log, MCP server. Reference run on public support data.	€1,990
PoC ★★RECOMMENDED	Trained on your LT tickets: intent taxonomy from your real queue, drafts grounded to your knowledge base - intent accuracy and groundedness measured. Code + quality report yours. No guarantee at PoC.	from €6,000
Implementation ★★★	Production: agent-assist in your helpdesk, approve-gates live, coverage + drift monitoring, retraining loop, monthly covered-% report - the agreed number (conservative) attaches here.	from €18,000

★ = engagement depth. PoC is the recommended path: quality proven on your data before production money. The PoC carries no performance guarantee by design; the agreed number (conservative) attaches at Implementation, informed by the PoC report.

What we don't promise

Out-of-policy questions stay with humans - by design. Refund exceptions, legal threats, custom cases: the model detects them and routes them instead of improvising. We report the % of tickets covered every month, so you grow the taxonomy deliberately - coverage is earned, not assumed. And a draft with no grounded source never reaches a customer.

Ready to see your own number?

Request the build: within 48h you get a personal reply with the value sized to your volume.

No commitment · reply within 48h · your data stays in the EU