- ACD
- Automatic call distributor. The telephony component that routes incoming calls to the right queue or agent based on rules such as skill, language or priority. It decides who picks up, not what they say.
- Agentic AI
- AI that does more than answer. It plans steps, calls tools and takes actions toward a goal, such as looking up an account and proposing a payment plan. In regulated work, agentic AI needs a write gate so that its actions are checked before they execute. See VoiceAI agents.
- AHT
- Average handle time. The average length of a contact from answer to the end of after-call work, including talk, hold and wrap time. Lower is cheaper, but only if resolution does not suffer.
- Audit chain
- A log where each entry is cryptographically linked to the one before it, so any change or deletion breaks the chain and becomes visible. On Spine every action and decision is appended to an HMAC hash-chained, tamper-evident log. See trust and governance.
- Auto QA
- Automated quality assurance. Software that scores every conversation against a scorecard instead of a small human sample, and routes disputed scores to a QA lead. See Auto QA.
- Barge-in
- The ability of a voice agent to stop speaking and listen when the caller interrupts. It makes a conversation feel natural instead of scripted. In supervisor tools the same word describes a supervisor joining a live call.
- Calibration
- The practice of having several reviewers score the same calls and compare results so everyone applies the scorecard the same way. With automated QA, the machine takes part as one more scorer and people keep the final word.
- CCaaS
- Contact center as a service. A cloud platform that provides routing, telephony, recording and agent desktops on subscription, instead of equipment on site. AI agents usually connect to it over SIP or WebRTC. See integrations.
- Containment
- The share of contacts an automated agent resolves without a transfer to a person. Useful, but a high containment rate means little if customers call back, so read it next to FCR.
- CSAT
- Customer satisfaction score. Usually collected from a short survey after a contact asking how satisfied the customer was. It measures the individual interaction rather than the overall relationship.
- Data residency
- Where customer data is stored and processed, geographically. Many regulators and clients require data to stay in a given country or region. Spine offers residency options including India (Mumbai region). See model router.
- Disposition
- The outcome code recorded at the end of a contact, such as promise to pay, callback requested or wrong number. Good dispositions drive reporting and next steps. After-call summaries can set them automatically for review.
- FCR
- First contact resolution. Whether the customer’s issue was fully resolved on the first contact, with no need to call back. It is one of the strongest links between service quality and cost.
- Golden call
- A reference conversation with agreed, correct handling and scores. A suite of golden calls is rerun whenever an agent, prompt or model changes, to catch regressions before customers do. See simulation and evaluation.
- Guardrail
- A rule that limits what an AI agent may say or do, such as verifying identity before disclosing a balance or refusing to discuss topics outside scope. Guardrails shape the conversation. A write gate controls actions.
- Hallucination
- When a language model states something that sounds confident but is not true or not supported by its sources. Grounding answers in approved knowledge with RAG and testing with simulation reduce the risk.
- Intent
- What the customer is trying to achieve on a contact, such as checking a balance, disputing a charge or setting up a plan. Detecting intent early lets an agent take the right path and helps leaders see why people call.
- IVR
- Interactive voice response. The menu system that greets callers with prompts such as press one for billing. Traditional IVRs route calls. Voice agents can understand free speech and resolve the request instead.
- LLM
- Large language model. An AI model trained on large amounts of text that can understand and generate language. It is the reasoning engine inside modern voice and chat agents.
- MCP
- Model Context Protocol. An open standard for connecting AI agents to tools and data sources in a consistent way. In Agent Studio you can import tools over MCP rather than building each integration by hand.
- Mini-Miranda
- The disclosure a US debt collector must give to identify the communication as an attempt to collect a debt and to say that information obtained will be used for that purpose. LiveAssist can nudge agents to read it. See collections.
- Model router
- A layer that sends each task to the most suitable AI model based on cost, latency and risk. It lets you mix open-weight and commercial models and switch models without rebuilding agents. See model router.
- Next best action
- A live suggestion to a human agent about what to do or say next, based on the conversation so far, the customer’s account and your policies. See LiveAssist.
- NPS
- Net Promoter Score. A measure of loyalty based on how likely customers are to recommend a company, on a scale of zero to ten. It reflects the whole relationship, not one call.
- Open-weight model
- An AI model whose trained weights are published, so it can be run on infrastructure you choose. It gives more control over cost, privacy and residency. Spine runs open-weight models, for example Qwen, on our own GPU capacity in Google Cloud.
- PCI DSS
- Payment Card Industry Data Security Standard. The security standard for any organization that stores, processes or transmits card data. For contact centers it means keeping card numbers out of recordings, transcripts and logs.
- Promise to pay
- A commitment from a customer to pay a stated amount by a stated date, recorded on a collections call. Promises made and promises kept are both tracked. See collections.
- RAG
- Retrieval augmented generation. A technique where the AI looks up relevant passages from an approved knowledge base and answers from them, with sources. It keeps answers current and reduces made-up content. See knowledge assist.
- Ramp gate
- A release control where a new agent version first takes a small share of live traffic and receives more only while its scores hold. If quality drops, the ramp stops. See simulation and evaluation.
- Redaction
- Removing or masking sensitive data, such as card numbers or personal identifiers, from recordings, transcripts and logs. Spine redacts automatically so sensitive details are not stored where they should not be.
- Regulation F
- The US rule issued by the Consumer Financial Protection Bureau to implement the Fair Debt Collection Practices Act. It covers how often and when collectors may contact consumers, which channels they may use and what validation information they must provide.
- Right-party contact
- Reaching the actual person responsible for an account, confirmed by verification, rather than a relative, a colleague or a wrong number. Disclosure rules usually apply only once right party is confirmed.
- Sentiment
- A reading of the customer’s emotional tone during a conversation, from negative to positive. Tracked across a call it shows where things turned. Tracked across many calls it shows which issues frustrate customers. See insights.
- Simulation
- Testing an agent with synthetic callers who follow scripted and adversarial scenarios before any real customer hears it. It is how difficult paths, such as an angry or vulnerable caller, get tested safely. See simulation and evaluation.
- Speech to text
- Technology that turns spoken audio into written text in real time. It is the ears of a voice agent and the source of live transcripts in assist tools.
- TCPA
- Telephone Consumer Protection Act. The US law that restricts automated and prerecorded calls and texts, especially to mobile phones, without the right consent, and limits calling hours. Consent checks before dialing are how contact centers stay within it.
- Text to speech
- Technology that turns written text into a spoken voice. Modern voices sound natural, handle pauses and emphasis and can speak many languages, including Hinglish and other Indian languages.
- Warm transfer
- A transfer where the receiving person gets the context of the call, and often an introduction, before taking over. The customer does not repeat themselves. It is the opposite of a cold transfer to a queue.
- Wrap time
- The after-call work an agent does once the customer hangs up: notes, dispositions and follow-up tasks. Automated summaries shorten it by drafting notes for the agent to review. See after-call summaries.
- Write gate
- A checkpoint that holds any consequential action an AI agent proposes, such as a payment plan, refund or account change, until it passes a policy check or a named person approves it. Then it executes with a signed audit entry. See trust and governance.