Which handoff type occurs when a customer returns after previously interacting with automation?

A continuity contact. It is defined by the customer coming back to an issue an automated system already touched, in a later session rather than the current one. It is distinct from the other three types: an escalated contact is a live interaction the automation could not finish and passed upward within the same session, an explicit handoff is the customer directly asking for a human, and specialized routing sends a case sideways to a specific team based on skills, language or authority. The continuity contact is the hardest of the four to handle well, because the context needed is in the past and must be retrieved, which requires persistent case history linked to a durable customer identity across channels.

What is a continuity contact in customer service?

A continuity contact is a customer returning after an earlier automated interaction on the same underlying issue. They asked an AI agent about a refund yesterday and are back today because it has not arrived. Nothing necessarily failed the first time; the customer simply came back. The defining property is that the relevant context sits in a previous session, so the system has to retrieve it rather than read it from the current conversation. The failure mode is treating the returning customer as a new contact and asking them to explain everything again, which is the single most common complaint about automated support.

What is the difference between an escalated contact and an explicit handoff?

The difference is who decided. An escalated contact is system-initiated: the automation determined it could not resolve the issue, because it lacked a grounded answer, lacked authority for the action, detected negative sentiment, or hit a policy rule requiring human confirmation. An explicit handoff is customer-initiated: the person asked for a human. They also fail differently. Escalations fail through context loss, where the human starts by re-asking what the customer already answered. Explicit handoffs fail through friction, where the company makes the request hard to fulfil in order to protect deflection metrics, and converts a routine request into an angry one.

What is specialized routing and how is it different from escalation?

Specialized routing sends a case to a particular queue, team or person based on what the case requires: a language, a product area, an account tier, a regulatory qualification or a spending authority. Escalation moves a case up a tier of seniority or capability. The simplest way to hold the distinction is direction: escalation goes up, specialized routing goes sideways to the right expertise. A Spanish-language billing query routed to the Spanish-speaking billing queue is specialized routing even though no one escalated anything. Its characteristic failure is bouncing between queues, with each reassignment adding delay the customer experiences as silence.

What context should be passed to a human agent during an AI handoff?

Six things. The full transcript including the automated portion, not just a summary. Verified identity and its authentication state, so a customer who already verified is not asked again, which research identifies as a leading driver of post-escalation satisfaction. Every action the automation already took, such as account lookups or attempted refunds. Detected intent and current sentiment. A handoff reason tagged to a taxonomy, such as knowledge gap, phrasing gap, missing authority or policy rule, so handoffs become analysable. And prior case history for this customer, which is what converts a continuity contact from a cold restart into a continuation.

When should an AI agent hand off to a human?

On five triggers. First, immediately whenever the customer explicitly asks, with no qualifying question in between. Second, when the AI cannot ground an answer in the knowledge base, because the alternative is inventing one. Third, when the action requires authority the agent does not have, such as an exception to policy or a refund above a threshold. Fourth, when sentiment turns clearly negative, since continuing to automate an angry customer compounds the problem. Fifth, when a business rule requires human confirmation before a sensitive action, which should be enforced in the platform rather than requested in a prompt. Handoff rate is not a failure metric; an AI that never hands off is either handling only trivial cases or getting things wrong.

All articles

AI & AUTOMATION·September 20, 2026·10 min read

AI-to-Human Handoff Types Explained: Continuity, Escalated, Explicit and Specialized Routing

When a customer comes back after an automated interaction, that is a continuity contact, not an escalation. The distinction matters because the four handoff types need four different designs. Here is each one defined, with the context each must carry.

D

Delyt Team

delyt.ai

IN THIS ARTICLE

01The four handoff types, defined
02What has to travel with every handoff
03Designing each type well
04The metrics that tell you whether handoffs work
05How Delyt handles this

Short answer. When a customer returns after previously interacting with automation, that is a continuity contact. It is the handoff type defined by the customer coming back to an issue that an automated system already touched, and it is distinct from an escalated contact, which is a live interaction the automation could not finish and passed upward in the same session. The four types in common use are continuity contact, escalated contact, explicit handoff and specialized routing.

These four get used loosely and often interchangeably, which is a problem because they fail in different ways and need different designs. A continuity contact fails when history is lost. An escalated contact fails when context is lost. An explicit handoff fails when it is slow. Specialized routing fails when it sends the case to the wrong place. Same symptom to the customer, four different root causes.

The four handoff types, defined

Type	What triggers it	Timing	Primary failure mode
Continuity contact	Customer returns after an earlier automated interaction	A later session	Prior history not retrieved, customer starts over
Escalated contact	Automation cannot resolve and passes upward	Same session	Transcript and verified state not carried across
Explicit handoff	Customer asks for a human	Same session, immediate	Slow or refused, customer has to ask repeatedly
Specialized routing	Case needs specific skills, language or authority	Either	Wrong queue, case bounces between teams

Continuity contact

A continuity contact is when someone comes back after an automated interaction has already happened. They asked the AI agent about a refund yesterday, the agent gave them an answer, and today they are back because the refund has not appeared. Nothing failed in the first interaction. The customer simply returned.

The defining characteristic is that the relevant context is in the past, not in the current session. Which means the system has to go and get it. The failure mode is specific and extremely common: the returning customer is treated as a brand new contact, the earlier automated exchange is invisible, and they are asked to explain the whole thing again. From the customer's point of view this is the single most irritating pattern in automated support, because it proves the company was not listening the first time.

Why continuity contacts are the hardest type to get right

The other three handoffs happen inside one live session, so the context you need is already in memory. A continuity contact requires persistent, retrievable history linked to a durable customer identity across channels and across time. If someone asked the AI a question on WhatsApp last Tuesday and emails today, the system has to resolve two different identifiers to one person and surface the earlier exchange. Most chatbot deployments cannot do this at all, because the bot session ends and takes its memory with it.

Escalated contact

An escalated contact is a live interaction the automation could not complete, passed upward to a human or to a more capable tier within the same session. The customer did not ask for this; the system decided. Triggers are usually one of: the AI could not find a grounded answer, the request needs authority the agent does not have, sentiment has turned negative, or a policy rule requires human confirmation before the action.

The failure mode is context loss. The human receives a queue entry that says "customer needs help" and begins by asking questions the customer already answered, including identity questions they already passed. Research on escalation quality consistently finds that authentication continuity is one of the largest drivers of post-escalation satisfaction: a customer verified before the handoff should stay verified after it.

Explicit handoff

An explicit handoff is the customer asking for a person. "Agent." "Talk to a human." "This isn't helping." It is the simplest type to detect and the one companies most often get wrong on purpose, by making it hard in order to protect deflection metrics.

That is a bad trade. A customer who asks for a human and is looped back into automation does not become a deflected ticket, they become an angry ticket, often arriving on a public channel instead. The rule worth following is that an explicit request is honoured on the first ask, without a qualifying question in between. If no human is available, say so immediately with a real timeframe rather than continuing to automate.

Specialized routing

Specialized routing sends a case to a specific queue, team or individual based on what the case needs: a language, a product area, an account tier, a regulatory qualification, a spending authority. It is a routing decision rather than an escalation, and it can happen with or without automation being involved beforehand.

It differs from escalation in direction. Escalation goes up a tier; specialized routing goes sideways to the right expertise. Its failure mode is bouncing: the case lands in the wrong queue, gets reassigned, and each hop adds delay while the customer hears nothing.

What has to travel with every handoff

Regardless of type, the receiving party needs six things. Missing any one of them forces the customer to repeat themselves, which is the outcome every handoff design exists to prevent.

1The full transcript, including the automated portion. Not a summary alone, the actual exchange, so the human can see exactly what the customer was told.
2Verified identity and its state. If the customer authenticated, that must persist. Re-asking for an order number or the last four digits after a handoff is the most damaging single thing you can do to post-escalation CSAT.
3Actions already taken. Account lookups performed, refunds attempted, records updated. The human needs to know what has already happened so they do not repeat it or contradict it.
4Detected intent and sentiment. What the customer wants, and how they feel about it by now.
5The handoff reason, tagged to a taxonomy. Knowledge gap, phrasing gap, missing authority, policy rule, customer preference, negative sentiment. This is what makes handoffs analysable rather than just survivable.
6Prior case history for this customer, which is the one that turns a continuity contact from a fresh start into a continuation.

Designing each type well

Type	What good looks like
Continuity contact	The customer is recognised across channels, prior automated exchanges surface automatically, and the opening line references the earlier interaction rather than asking them to restate it.
Escalated contact	The human opens a pre-populated workspace with transcript, identity, actions taken and a reason tag. No re-authentication. The customer is told a human is joining and roughly when.
Explicit handoff	Honoured on the first request, no qualifying questions. If no one is available, an immediate honest timeframe and a promise that is kept.
Specialized routing	The routing decision is made once, from case attributes rather than guesswork, and reassignment is logged so bouncing shows up in reporting.

The metrics that tell you whether handoffs work

Handoff rate by reason tag. A rising knowledge-gap share means the knowledge base needs work; a rising phrasing-gap share means the AI is failing to understand rather than failing to know.
CSAT on escalated conversations specifically, measured separately from overall CSAT. Blended CSAT hides exactly where handoffs do their damage.
Repeat-contact rate within 7 days, which is the direct measure of continuity-contact failure. If people keep coming back about the same thing, the first interaction did not resolve it.
Re-authentication rate after handoff. This should be close to zero and usually is not.
Reassignment count per case. Anything above one hop is a specialized-routing problem.
Time to human after an explicit request. Measure it from the first ask, not the last one.

How Delyt handles this

Delyt threads every channel onto one customer record, so a returning customer is recognised whether they come back on WhatsApp, email or Instagram, and prior automated exchanges are part of the same case rather than a separate bot log. Handoffs carry the full transcript, the verified state and the actions already taken, and the execution engine can be configured to require human confirmation before sensitive actions, which turns a class of risky automations into designed escalations rather than accidents. Every action is audit-logged, so handoff reasons are analysable rather than anecdotal.

Handoffs that do not make customers repeat themselves

One customer record across every channel, full context on every handoff, and AI that escalates by your rules rather than by guesswork. From $29/mo flat, no per-resolution fees.

See how it works

FREQUENTLY ASKED QUESTIONS

AI vs Human Support: When to Automate, When to Stay Human

Mar 18, 2026·11 min read

Industry & Opinion

The end of the chatbot era

Jun 19, 2026·6 min read

AI & Automation

How to Automate Customer Support With AI: A Practical Guide for Small Teams in 2026

Mar 22, 2026·12 min read

READY TO SEE IT IN ACTION

Faster responses. Smarter routing.
Less work for your team.

Start free trialExplore features