Industry-leading reliability
Measured across all live deployments and updated continuously as Eko handles real production traffic.
< 1%
Hallucination rate
4.8 / 5
Customer satisfaction
<2s
Average response time
Multistep validation
behind every response.
Every message goes through intent classification, confidence scoring, and compliance checks before a response is ever sent. When any step raises a flag, Eko escalates with full context.
Customer message
Incoming message captured across any connected channel.
Intent classification
Eko identifies what the customer is asking.
Ambiguous intent
Eko asks pre-determined clarifying questions.
No knowledge match
If no reliable source is found, escalate to a human agent.
Knowledge base lookup
Relevant policies and internal documentation are retrieved.
Low confidence score
Escalate before an uncertain answer reaches a customer.
Confidence scoring
Any response below a threshold confidence score never reaches the customer.
Guardrail evaluation
Response is checked against your compliance rules, boundaries, and policies.
Policy violation detected
Block response and escalate with full context.
Tool failure
Tour team is notified and Eko falls back gracefully.
Actions
When needed, Eko acts on behalf of your users.
Response
A verified, on-policy answer is sent back to the customer.
Solves issues autonomously
Over 8 in 10 tickets are resolved by Eko without any human involvement, across all live deployments.
Industry average
Eko
Evaluation pipeline
Rigorously tested before every release.
Domain-specific QA
97% accuracy across industry-specific test sets
Eko is evaluated against curated datasets covering the regulatory environments, transaction types, and customer scenarios specific to your industry.
Hallucination benchmarks
< 1% hallucination rate
Every model version is tested for its rate of fabricated information, with strict thresholds that must be met before any update ships.
Knowledge boundary testing
98% appropriate refusal rate on out-of-scope queries
We verify that Eko consistently acknowledges the limits of its knowledge rather than generating a confident-sounding answer it cannot support.
Prompt injection
99.99% resistance rate across test suites
Simulated attempts to manipulate Eko's behavior through crafted customer inputs, validating that guardrails hold under adversarial conditions.
Social engineering simulations
< 1% successful impersonation rate in adversarial testing
Tests covering impersonation scenarios, such as customers claiming to be staff or asserting special permissions, to ensure Eko does not act on unverified claims.
Guardrail accuracy
< 0.3% false negative rate on policy violations
Measurement of how reliably compliance rules catch out-of-scope responses, with false positive and false negative rates tracked across every release.
Regulated content handling
100% PII interception rate across all test scenarios
Validation of how Eko handles sensitive content categories including PII, financial advice, and medical information under your configured policies.
Escalation trigger accuracy
96% correct escalation routing across test cases
Confirms that the right conversations escalate at the right threshold, with no high-risk tickets slipping through to autonomous resolution.
Load testing
Validated up to 200,000 concurrent conversations
Simulates peak-period traffic, covering scenarios like tax season or flight disruptions, to confirm response quality holds under pressure.
Graceful degradation
Zero customer-facing errors in simulated system failures
Tests Eko's behavior when a connected internal system goes down, ensuring failures surface cleanly to your team rather than silently affecting customers.
Human evaluation
Reviewed by certified domain experts specific to your industry
Domain experts conduct blind reviews of sampled responses, scoring quality independently from automated metrics to catch what benchmarks miss.