Salted Knowledge

The policy changed. Every workflow should already know.

Salted turns approved policies, procedures, and playbooks into one governed source of guidance, so AI agents, human agents, and quality reviews can act on the current rule instead of yesterday's.

Refund policy - threshold raised $30 -> $100
Without Salted With Salted
Approved source
What each workflow follows

Refund policy

Approved
Old $30
Current $100

Damaged items reported within 48 hours qualify for refund up to $100 with delivery confirmation.

AI agent $30
Human agent $30
Quality review $30
Salted Knowledge Applies approved guidance at the point of work
Same approved change. Three workflows land on $100.
AI agent $100
Human agent $100
Quality review $100

The expensive problem is not missing knowledge. It is drift.

Most teams already have the answer somewhere. Trouble starts the moment a refund threshold moves from $30 to $100 and every workflow has to catch up on its own schedule.

The AI agent keeps answering with the old number. Human agents rely on yesterday's playbook. Quality reviews check against a stale standard. Same company, three different answers.

If one team updates the policy and another workflow still follows the old rule, AI has not made the operation smarter. It has made drift faster.

In the conversation

Agents get guidance, not another tab.

When a human agent joins, the relevant policy appears inside the workspace with suggested wording, the source behind it, and why it applies.

Agent Desktop - conversation
Assigned handle time01:55
Direct feedback37.2%
Guidance54%
Sentiment11.5%
Progress28%
Carmen Sanchez Damaged item - refund policy
Today
My order arrived damaged yesterday. Do I qualify for a refund?
Current policy: damage reported within 48 hours qualifies for refund up to $100 with delivery confirmation.
Thanks. It arrived yesterday and is $84, so it is within the current refund policy.

Show your work

Show the source before anyone has to trust the AI.

Every answer, recommendation, or review comment can carry its source trail: which policy, market, channel, and update supported it.

That gives operations real control: approve the source, inspect the guidance, flag the wrong one, and update the policy when reality changes.

Quality review - Conversation 48421
AI answer reviewed

Your damaged item qualifies for a refund because the issue was reported within 48 hours and the item value is under the current $100 threshold.

Knowledge accuracyPass
Policy$100 rule
Used knowledge

Refund policy: damage within 48 hours qualifies for refund up to $100.

UpdatedToday
ChannelUS chat
Available toAI + agent
StatusApproved

Scoping and availability

Scope guidance to your real operation.

Real operations are not one clean FAQ. They have brands, markets, channels, customer types, exceptions, and offers. The same approved source can be made available differently to each workflow.

Turn it on for AI agents and human agents, keep it in scope for quality, gate it on deal-page chat, all without forking the rule.

Knowledge - Refund policy
Brand: Demo Adventures Market: US Channel: Chat + Voice Active
AI
AI agentsAI answers
H
Human agentsIn-conversation guidance
Q
Quality reviewsScoring and comments
D
Deal-page chatEligibility for proactive chat

Proactive AI chat

Eligibility checked before the AI ever opens chat.

On a deal page, the AI agent uses approved guidance: offer rules, customer type, market, and channel, to decide whether to engage or stay quiet.

No rogue discounts, no off-policy promises. The opener a shopper sees is one the rules already cleared.

shop.demoadventures.com/summer-bundle
/
$420
Tyro-X Outdoor bundle
AI assistant - eligibility check
Offer bundle discount eligible
Customer returning, US
Channel: deal-page chat allowed

Hi. The Summit 2 qualifies for our returning-customer bundle. Want me to add the matching footprint at 15% off?

Quality feedback

Quality becomes the alarm system for stale knowledge.

Reviews do more than score conversations. They surface exactly where agents or AI are still using the $30 rule after the business moved to $100, and feed the fix back to the source.

Quality Old threshold detected

A conversation denied an $84 refund using the previous $30 rule.

Knowledge Approved source updated

The refund threshold is corrected once, at the source: $30 -> $100.

Every workflow Correction reaches all three

AI agents, human agents, and quality reviews all move to $100 together.

Governance

Simple for buyers. Reviewable for technical teams.

Salted uses the sources your company approves, finds relevant guidance when AI responds or reviews, and applies it to responses, recommendations, quality reviews, and AI chat behavior.

Security and engineering can review architecture, isolation, retention, model providers, and change control during technical evaluation.

01 Account isolation

Each account's sources stay separate and governed.

02 Scoped availability

Control where each source can be used, per workflow.

03 Change control

Policy updates are visible before they affect AI behavior.

04 Reviewable trail

Every answer and review names the source behind it.

See what happens when a policy changes.

Walk through the $30 -> $100 refund example across AI agents, human agents, and quality reviews in your own operation.

Book a live demo