Salted AI Agents

AI that stays in the conversation.

Salted AI Agents automate live customer conversations without disappearing when human judgment is needed. The AI can answer, ask structured questions, collect missing details, request guidance, confirm resolution, and feed the outcome into Quality Intelligence. Teams can use Salted automation, connect existing bot or voice-bot logic, or combine both.

Book a live demo See where agents fit
Customer conversation - refund request
Live
CS
Carmen Sanchez Chat - returning customer
AI active
My order arrived damaged yesterday. Can I get a refund?
Yes. Since the damage was reported within 48 hours and the item is under the current limit, I can help start the refund.
Do you need anything from me?
Please upload the delivery confirmation and one photo of the damaged item.
Operating logic Policy, context, state, and action selection
Governed
Customer context Returning, US, chat

Conversation, customer, recent turns, and brand context.

Knowledge Refund policy

Approved knowledge from Salted Knowledge.

Conversation state Refund intake

The agent knows what has already been asked and collected.

Human guidance Ask when needed

The AI can request a decision without dropping the conversation.

01
Reply to the customer Message or email, depending on the channel.
02
Ask or collect Buttons, menus, forms, or a dynamic follow-up question.
03
Guide, close, or collect feedback Ask a human, complete the case, or collect satisfaction feedback.

The hard part is not writing one AI answer. It is keeping the experience continuous.

Customer operations are full of branches: policy limits, market rules, brand exceptions, language requirements, customer status, channel constraints, and escalation paths. A chatbot that answers FAQs and then passes the customer on without context cannot safely run that work.

Salted treats the AI agent as part of the operation. It has to know which brand, language, and channel apply, what state the conversation is in, which action is allowed, and when a human should guide the next step.

That is where Salted AI Agents need more than a response model: quick selections, customer questions, escalation rules, resolution checks, customer feedback, quality review, and knowledge all shape what the AI should do next.

What enterprise CX teams are asking for: AI and human agents working in the same interaction, not handing customers between disconnected or fragmented systems.

AI + human + AI

When the AI needs help, it asks. It does not disappear.

The AI can answer, gather context, and choose actions on its own when the path is clear. When confidence drops or a policy needs judgment, it brings a human into the same conversation with the context and the proposed next step.

The human can approve, adjust, reject, or take over. Either way, the AI stays connected: drafting replies, surfacing policy, showing the next customer question, capturing resolution signals, and feeding Quality Intelligence.

Same conversation - human guidance
1
AI handles the routine work It answers, asks questions, checks state, and prepares the next action.
AI
2
AI asks for judgment The human sees the summary, the proposed action, and the decision point.
Guide
3
Human decides Approve, adjust, reject, invite another person, or take over fully.
Human
4
AI continues with context The customer stays in one conversation while the operation keeps moving.
Continue

Where agents fit

AI Agents are part of the same Salted system.

Teams do not all start from the same place. Some already have automation logic and need Salted to provide the Agent Desktop, routing, and measurement around it. Others run support in Salted and want to automate high-volume contact reasons inside the same platform. AI Agents support either model, or a mix of both: automation configured in Salted and existing automation logic connected to Salted. If your team owns the decision logic, Salted can be the customer conversation system around it. If you want Salted to automate the contact reason, the behavior can be configured in Salted.

Salted automation

Use automation in Salted AI Agents.

Salted automation means workflow and agent behavior configured in Salted: policies, structured customer questions, knowledge, escalation rules, resolution checks, and expected behavior. Your team monitors outcomes in the same platform and quality review flow. This is the product path when the automation behavior should live in Salted.

Best for teams that want bot behavior inside Salted, not in a separate service.
Useful for high-volume contact reasons that can be automated inside the product.
Shares the same state, knowledge, escalation path, and Quality Intelligence as connected logic.
Existing automation logic

Connect existing bot or voice bot logic.

Existing chatbot logic, voice bot logic, or workflow service decides what happens next, including external voice automation that escalates into Salted Agent Desktop. Salted adds the conversation state, human guidance, and measurement around it.

Best for teams that already have bot logic, voice bot logic, or internal automation builders.
Salted provides the state, actions, knowledge context, human guidance, and measurement around it.
Keeps existing decision rules under your team's control.

How it works

The agent chooses actions, not just words.

Each turn is received, checked against the right workflow, and returned as a set of actions for the customer conversation. The agent can send a reply, show structured choices, collect a field, update conversation state, write a note, ask for human guidance, invite another participant, route to a human, ask whether the issue was resolved, collect satisfaction feedback, or complete the conversation.

The point is not to add another interface. The same system can support deterministic policies, LLM-led policies, and topic-aware policies that select the relevant background context for the turn.

Agent turn - action selection
1
Receive the turn Conversation, customer, recent turns, and trigger arrive from Agent Desktop.
Received
2
Load state and context Conversation state, brand context, and approved knowledge are loaded.
Scoped
3
Select the next action Salted automation or existing automation logic decides what should happen next.
Decide
4
Return the action The customer conversation receives a reply, choice menu, guidance request, note, survey, or completion.
Action

Customer-facing actions

Agents can ask, guide, and close the loop.

On a deal page, the agent can use page context and configured rules to open a conversation and ask the first question. In a support flow, it can show choices, collect missing details, confirm whether the issue was resolved, or ask for customer feedback after the case closes.

Because these are actions in the same system, the questions, choices, and messages customers see use the same state, knowledge, human guidance, and Quality Intelligence layer as the rest of the conversation.

shop.example.com/deal/summer-bundle
A
Anonymous visitor No eligible offer
Quiet
R
Returning customer Bundle offer eligible
Engage
V
VIP customer Needs human approval
Review

Offer assistant

The agent checks page, customer, market, and offer rules before it opens chat. If the case is risky, it can stay quiet, ask a safer question, or route to a human.

Page Bundle deal
Customer Returning
Action Ask first question
Hi. This bundle has a returning-customer offer. Want help checking eligibility?
Damaged delivery Collect evidence, apply policy, replace or refund.

Ask for a photo, inspect order context, request approval when needed, then close the loop.

Where is my order Check status and choose the next action.

Answer directly, ask one clarifying question, notify the right team, or route to a specialist.

Reorder and top-up Turn support moments into helpful commerce.

Use customer and product context to suggest a reorder, subscription, bundle, or limited offer.

Resolution feedback Confirm, collect feedback, and feed quality.

Ask whether the issue was resolved, collect satisfaction feedback, and send the signal into quality review.

Built to scale

Start with one capability or run the full customer operations stack.

Salted does not have to replace everything at once. The same platform can evaluate conversations, run Agent Desktop, monitor ongoing bot conversations, show questions, choices, and messages to customers, and automate decisions as teams are ready.

Quality Intelligence

  • Measure conversations against policies, expected answers, and quality criteria.
  • Works even when the contact center or voice stack sits outside Salted.
  • Uses quality reviews, resolution checks, satisfaction feedback, and conversation review as feedback signals.
  • Shows where knowledge, policy, guidance, or automation behavior needs attention.

Agent Desktop

  • Run chat, messaging, email, calls, outbound follow-up, and bot escalations in Salted.
  • Monitor ongoing bot conversations and join when human judgment is needed.
  • Keep conversation state, routing, bot escalations, notes, customer context, and AI guidance in one place.
  • Show customer-facing questions, choices, messages, and follow-up surveys.
  • Connect existing bot, voice bot, or workflow logic when another system already decides what happens next.

AI Agents

  • Automate high-volume contact reasons inside Agent Desktop.
  • Use Salted automation or connected automation logic.
  • Reply, ask structured questions, collect fields, ask for guidance, or complete conversations.
  • Trigger resolution checks, satisfaction feedback, and evaluation through Quality Intelligence.

Buyer questions

What teams need to know before automating.

The practical question is not whether an AI agent can talk. It is which contact reasons are ready, what customers and human agents will see, and how the team will catch mistakes before they become the new process.

Start narrow: one contact reason, one brand, one channel. Review every action against expected behavior, then expand once quality checks pass.

What contact reasons can I automate?

Start with high-volume workflows where the policy, context, and allowed actions are clear: damaged delivery, where-is-my-order, reorder and top-up, offer eligibility, specialist escalation, or resolution feedback.

How quickly can we start?

The fastest path is a scoped start: one contact reason, one brand, and one channel. Review every action in a controlled flow, then expand once quality checks pass.

What will humans see?

Humans see the conversation, customer context, brand and channel context, proposed action, relevant knowledge, collected details, and the path to approve, adjust, reject, or take over.

What will customers experience?

Customers stay in the same conversation. The agent can answer, show choices, ask for missing information, request human help when needed, confirm resolution, and collect feedback.

How do I know it is working?

Review automation actions, completion and escalation patterns, resolution checks, satisfaction feedback, and quality evaluations that compare expected behavior with actual output.

What happens if it is wrong?

Risky turns can ask for human guidance. Reviews show where behavior, knowledge, policy, or workflow needs adjustment, then Quality Intelligence feeds fixes back into the operation.

Why technical teams can trust it

Designed for continuous collaboration, not a disconnected bot.

Salted can support teams using only Quality Intelligence, teams running Agent Desktop, and teams adding AI Agents because the product is designed as one system. For AI Agents, the decision logic can come from Salted automation or from connected automation logic, while brand, language, channel, and conversation state remain shared.

Technical teams can review the actions, questions, choices, messages, context, human guidance, escalation path, and evaluation loop instead of trusting a closed bot.

01 Account, brand, language, and channel scope

State, context, and configuration stay tied to the right account, brand, market, language, and channel.

02 Reviewable actions

The agent returns replies, questions, surveys, notes, or escalation actions the system can inspect.

03 Human guidance

Risky turns can ask for a human decision without losing conversation context.

04 Quality evaluation

Evaluation helps compare expected behavior with actual agent output.

Common questions

How Salted AI Agents fit into customer support automation.

Salted AI Agents are built for contact center teams comparing AI bots, agent assist, and contact center platforms, and who need automation, human guidance, and quality feedback to work together.

What are Salted AI Agents?

Salted AI Agents automate customer conversations by choosing actions: replying, asking structured questions, collecting details, requesting human guidance, routing work to a human, checking resolution, or completing the conversation.

Is Salted only an AI bot?

No. AI Agents are one part of Salted's AI contact center platform. The same platform can include Agent Desktop, knowledge, quality intelligence, reporting, human guidance, and automation.

Can Salted connect to existing bot or voice bot logic?

Yes. Teams can connect existing chatbot, voice bot, or workflow logic while Salted provides Agent Desktop, conversation state, actions, knowledge context, human guidance, and measurement around it.

Can teams use Salted automation?

Yes. Salted automation means workflow and agent behavior configured in Salted: policies, knowledge, escalation rules, structured customer questions, resolution checks, and quality evaluation in the same platform.

Can human agents monitor and join bot conversations?

Yes. Agents can monitor ongoing bot conversations, see the context and proposed next step, then approve, adjust, reject, invite another person, or take over.

Can Salted support multiple brands, languages, and channels?

Yes. State, context, configuration, policies, and knowledge can stay tied to the right account, brand, market, language, and channel.

What happens when the AI needs human judgment?

The AI asks for guidance inside the same conversation. A human can approve, adjust, reject, or take over, and the AI stays connected to continue the work.

See where AI Agents fit in your Salted setup.

Walk through one high-volume contact reason and decide what should be automated: answer, ask, collect, guide, get feedback, close, or connect existing bot or voice bot logic.

Book a live demo