klax IQ is klaxify's local-first ecommerce intelligence system. It recognizes routine intent, follows context across a conversation, evaluates approved patterns, and can answer from an active playbook without sending the same request to an outside AI model.
One connected klax IQ workflow
The klax IQ page starts with a plain-language view of the complete workflow: understand, ground, resolve, and improve. The launch checklist shows what your workspace still needs, while routing thresholds, compiler evidence, learned skills, and playbooks remain available under Advanced klax IQ controls.
The individual intelligence layers no longer make separate interpretations of the same customer message. klax IQ creates one tenant-scoped resolution packet containing the customer’s requests and goals, conversational phase, references, missing clarifications, verified sources, communication approach, safety blockers, and paid-reasoning requirement. Reply generation, inbox guidance, relevant actions, mobile responses, and audit history use that shared decision record.
This keeps follow-ups and corrections connected, prevents later steps from quietly changing the detected intent, and allows a safe local skill to avoid an outside call when the required facts are already verified. Shopify permissions, approval policies, and action validation are still evaluated at execution time.
The packet also supplies a resolution graph and one pre-send verifier. Each request must connect to current tenant evidence or live commerce context, and the final wording must match the current customer and actual sender. The same verifier gates human-assisted sending, automatic sending, mobile replies, quality review, autonomy readiness, and verified first-reply reporting.
The three IQ layers
- Core IQ starts with 100 versioned ecommerce support skills across orders, delivery, returns, exchanges, refunds, order changes, products, checkout, subscriptions, accounts, loyalty, and feedback. It can recognize several questions in one message instead of forcing the request into one broad category. Core IQ contains no merchant or customer data.
- Workspace IQ uses only your organization's policies, products, voice, verified customer context, accepted replies, and active playbooks. This information is isolated from every other tenant.
- Fast Lanes let routine requests skip an outside model only when confidence, approved knowledge, required facts, privacy, and risk checks all pass. Requests that change commerce state remain governed.
Managers can open klax IQ to enable or pause each layer, set the Fast Lane confidence threshold, and choose which safe skills may use a Fast Lane.
Prepare klax IQ for launch
Choose Open guided IQ setup on the klax IQ page to complete a full-screen, step-by-step interview. The flow writes into the same Workspace IQ, employee identity, knowledge, product intelligence, action policy, evaluation, and autonomy systems used at runtime.
- Describe the business, customers, markets, service promises, and first goals.
- Connect and sync Shopify.
- Review Shopify policies and product descriptions before they become trusted evidence.
- Crawl the public website for FAQs, guides, warranties, and other missing knowledge; confirm only accurate pages.
- Configure AI employee voice, empathy, identity, instructions, and sign-off.
- Choose automatic-action, approval, and amount-limit guardrails.
- Build a de-identified, tenant-private accuracy baseline from historical requests.
klax IQ stays supervised while required information is incomplete. Completing the interview does not bypass live action validation, permissions, approval rules, or skill-level accuracy gates.
Customer and employee names remain dynamic in learned replies. klax IQ stores {{customer.first_name}} and {{agent.name}} placeholders instead of memorizing names from approved examples. At send time, klaxify inserts the current customer and the human or AI employee who actually sends the reply. Every outbound email receives a personalized greeting and a sender-matched signature.
Contextual conversations
Workspace IQ tracks the current conversational phase, including new requests, follow-ups, customer corrections, acknowledgements, and topic changes. It gives the responding AI employee explicit guidance to continue naturally, avoid repeating settled information, and replace a mistaken assumption when the customer corrects it.
References such as “that one” or “is it still active?” are treated as conversational clues, not verified facts. klaxify still uses current tenant data and connected commerce records before making a claim or taking an action.
This example is a product explainer, not a live score. In your workspace, the response skill shows its own approved evidence, readiness, source status, and review requirement.
Training signals
When a teammate approves an AI-assisted reply, klaxify records whether the pattern was accepted and updates the tenant’s learning evidence. Safe response patterns can become proposals or active playbooks according to the workspace training setting.
When a teammate edits a recommendation, klaxify also classifies the correction—such as personalization, sender identity, an incorrect fact, an incorrect action claim, missing coverage, or tone. This helps strengthen the right behavior without teaching customer or employee names as reusable content.
Resolution compiler
The resolution compiler converts approved outcomes into testable Workspace IQ skills. Each compiled skill records the customer-language examples it learned from, the verified facts and versioned sources it requires, confusing counterexamples, prohibited completion claims, applicable actions, and its risk level.
A compiled skill can avoid outside generation only when every requirement is satisfied and the intent is safe for local handling. Missing knowledge, stale evidence, risky commerce changes, or language that claims an action completed keeps the skill supervised and routes the conversation through the normal reasoning and approval path. Managers can review compiler coverage and blockers on the klax IQ page.
The learning is isolated to your organization. Another tenant’s replies, products, policies, and outcomes are never used as your workspace facts.
Self-improvement is bounded: approval strengthens a specific answer skill, while an edited reply records the correction and response alignment. Sensitive requests, unsupported claims, and completed-action statements do not become trusted merely because they appeared in a reply.
Multi-question replies
When a customer asks several questions in one message, klaxify learns one answer skill for each request instead of saving the entire reply under a single topic. Greetings, sign-offs, and brand voice remain separate from the reusable answers.
For example, one approved email can strengthen three independent skills:
- Shipping countries
- Refund and return policy
- Product-specific ingredients
Open the proposed skills to review the answer, original customer question, supporting sources, confidence, and number of approved examples. A superseded combined proposal is not used for future replies.
Automatic reply safeguards
A multi-question reply can only be sent automatically when every detected request has current supporting context, a matching response skill, and a complete answer. If one answer is missing, klaxify blocks automatic sending and keeps the response in human review.
Answer splitting, feedback measurement, playbook learning, caching, and coverage checks run inside klaxify. An outside AI model is used only when the message is ambiguous or the final response needs language generation that cannot be safely assembled from approved skills.
Autonomy increases only after enough reviewed examples establish a conservative reliability floor. The raw acceptance percentage is not sufficient by itself. A recent wrong-action correction, critical verifier failure, or manager pause trips the skill's circuit breaker and returns it to review. Managers see these results as a short readiness summary first; detailed thresholds and regression cases remain under Advanced klax IQ controls.
What never becomes trusted automatically
An approved sentence is not proof that a refund, cancellation, replacement, or other transaction happened. Commerce actions still require live validation, the current action policy, and execution through the connected provider. Sensitive or low-confidence requests route to an outside model or human review instead of guessing.
Open klax IQ to inspect local decisions, external fallbacks, accepted outcomes, and playbook proposals.
Ask klax IQ
Tenant managers can open Ask klax IQ from the app header. The floating conversation bar answers operational questions from the active workspace, such as:
- What needs attention right now?
- Which actions are waiting for approval?
- Find order #1045.
- Show replies that need quality review.
- Which VIP customers are waiting?
The bar starts with deterministic, tenant-scoped tools and normally uses $0 in outside AI credits. A local scope gate declines questions unrelated to ecommerce, customer support, the store, or klaxify before an outside provider can be contacted. The bar is read-only: it can organize evidence and open a governed workflow, but it cannot bypass validation or approval policy.
