SUMMY.PRO · TELCO · 2026
// SOLUTIONS · TELCO

Telco support that pays for itself. And keeps customers happy.

We've been deploying AI-powered customer support for the largest telecom operators in Catalonia for years. AI that knows your catalogue, your KB and your processes — escalating to humans only when it matters.

RAG with vector searchLatency < 500 msMulti-collection KBNative human escalation
See full case
Live in production · 24/7/365
// WHAT WE'VE SHIPPED

Real numbers, in production.

Not lab benchmarks. These are the aggregated metrics from the deployments we run today inside Spanish operators.

+3,000USERSserved across Spain
9.6 / 10CSATaverage end-user score
+35,000CHATS IN 2026handled by Summy
80 %AI RESOLUTIONwithout human touch
98 / 100KB ACCURACYcorrect answers per 100
// RAG ENGINE01

Vector search across your knowledge base. In real time.

Our RAG engine doesn't improvise: every reply cites the KB section that supports it. We index multiple collections (products, pricing, tech support, retention, internal FAQs) and route each question to the right collection before generating.

LATENCY< 500 msP95 per request

Retrieval + generation + post-processing. No waits that break a WhatsApp conversation.

SEARCHVectorDense embeddings

Recall by meaning, not keywords. Ask «how do I cancel» → finds the churn section even if it says «outbound number portability».

SCALEN collectionsMulti-domain KB

Products, pricing, tech support, retention, legal — each with its own access rules and tone.

PRECISION98 / 100Hit rate

Audited against real telco conversation sets. Every answer is traceable to its source.

// FOOTNOTEOn-prem or private cloud · Customer data never leaves your instance · Full audit trail per generation.

// AI + HUMANS02

Your team isn't going anywhere. They just stop doing the boring work.

Summy handles tier-1 and escalates to humans when needed. Chats and tickets unified in one inbox, with explicit rules for what AI can touch and what it can't.

01

AI handles tier-1

Standard queries — pricing, coverage, order status, minor changes — answered with the KB and customer data.

02

Sensitive intent detection

Cancellations, complaints, fraud or any flow tagged «human-only» are identified before the AI replies.

03

Native escalation

Chat or ticket moves to the right team — retention, L2 support, legal — with full context and prior thread.

04

Continuous learning

Every escalation marks where the AI must keep learning. The KB updates with the human's reply.

REAL EXAMPLE · RETENTIONCancellation request → AI tier-1 → escalation to retention
CustomerHi, I want to cancel my line.
Summy AIGot it. Before I process it — quick question, what's the reason? I might be able to help without cancelling.
CustomerI pay €32 and my neighbour pays €19 for the same plan.
Summy AIFair. I can see you've been with us 14 months — I can offer you the same €19 plan with no commitment. Works for you?
CustomerI'd rather still cancel.
→ ESCALATED TO RETENTION TEAM · FULL CONTEXT
Marta · RetentionHi, this is Marta from retention. I saw the offer Summy made — let me review your case and come back with a tailored adjustment.
// KB AUTHORING03

We help you write the knowledge base. And tune it for the AI agent.

A KB written for humans doesn't work the same with an LLM. We rewrite it with you — product, support and legal in the room — so the AI gets it right the first time.

  • Initial audit of your current KB: gaps, contradictions, stale sections
  • Restructuring by collection (product, pricing, support, retention, legal)
  • Rewrite tuned for vector retrieval — self-contained chunks, no dangling refs
  • Validation against real conversations before each release
  • Continuous workflow: every human escalation triggers a KB update suggestion
  • Controlled multilingual (ES / CA / EN) from one source
AVERAGE ACCURACY98 / 100correct answers after KB rewrite on telco engagements

// ENTERPRISEKB authoring and rewrite is included in the Enterprise plan only. Conditions apply based on content volume, languages and timelines.

// CONTEXT04

Live in the largest telcos in Catalonia.

We don't sell slides. What you see is in production 24/7, serving real end customers — not pilot users.

“80% of our chats close without a single team member touching them. And the customer leaves with a 9 out of 10 on average.”
Operations teamRegional operator · Catalonia
“When the AI escalates to retention, the conversation arrives with full context. We close deals faster and lose fewer customers to back-office chaos.”
Head of retentionOperator · 2,000 lines
// READY TO TALK

Cut support cost. Without losing the customer.

We set up a closed pilot with your KB and processes in under 4 weeks.

05 · Tell us

An engineer replies within 24 h.

Tell us your monthly chat volume, what processes you want to automate and what KB you have today. We'll come back with an estimated scope and a proposed pilot.