AI support · SoliqFrom 41% to 89% first-contact resolution in three months
A global D2C brand fielding 60k support conversations a month replaced six tools with one Soliq-grounded support copilot, lifting FCR 48 points without losing a single edge case to hallucination.
A stylised excerpt of the production UI.
The problem we walked into.
Six surfaces (web chat, WhatsApp, email, Instagram DMs, returns portal, voice IVR) each running on a different vendor. CSAT inched up year-over-year but cost per ticket grew 31%. The pre-existing chatbot resolved 18% of contacts and was the leading driver of negative sentiment.
How we attacked it.
- 01
Audit of 12 months of ticket data, clustered into 31 intent classes, found that 7 intents drove 68% of volume.
- 02
Soliq deployed in the customer's AWS VPC. RAG indexed across help-center, returns policy, last 24 months of resolved tickets (with PII redacted), and live order data via the OMS API.
- 03
Custom support copilot deployed across all 6 surfaces in week 4. Eval set of 480 graded conversations regenerated weekly from real traffic.
- 04
Guardrails: never make promises about delivery dates we cannot verify, never invent SKUs, never quote prices in non-USD without conversion confirmation.
- 05
Human handoff: detection of escalation signals (frustration sentiment, complex policy edge cases, refund > $200, accessibility needs).
- 06
Continuous deployment: prompt versions A/B tested weekly with eval gating.
The tools, picked deliberately.
“Soliq moved our support resolution rate from 41% to 89% in three months. The eval discipline they bring is rare. Most vendors hand you a demo; Techliphant hand you a system you can operate.”
Want the full case PDF?
We share the unabridged story: architecture diagrams, eval set excerpts, runbooks, all under mutual NDA.
More cases.
Ready when you are
Let's build something exceptional.
Tell us about your business, your stack, and the problem you are trying to solve. We respond with a clear next step usually a 30-minute discovery call, no fluff.