Business knowledge

How does AI learn your business?

Choose the files, website pages, FAQs and product information the agent may use. Approve sources, review answers, update changed information and agree what happens when a question needs a person.

Sculpture of website, document and FAQ panels linked by brass pipes to a glowing approved-answer block on a plum marble base

An approved source, an answer and an update

Illustrative walkthrough — sample source wording and answers, not a product screenshot or customer record.

  1. 01

    Choose and approve a source

    A service FAQ approved by its owner says: “Preparation instructions are on our service page.” Keep its page reference and exclude drafts or internal-only material.

  2. 02

    Test a question against that source

    Ask “How should I prepare?” The answer points to the approved preparation instructions. If the source does not cover a requested exception, the assistant should not guess.

  3. 03

    Route an unanswered exception

    A question about a special requirement goes to the business’s configured contact path with the question and source context. A person confirms what applies.

  4. 04

    Update and retest

    When preparation instructions change, the owner approves the replacement source, retires the old wording and repeats both the routine question and exception test before sharing the revised answers.

What counts as approved business knowledge

You decide what the agents may use. Nothing enters the answer set without being approved first.

Uploaded files

Policies, manuals, service descriptions, price lists, onboarding material and internal guides you choose to include.

Selected website pages

Crawled pages from your own site, scoped to the sections you approve rather than everything published.

FAQs

Existing question-and-answer content, including the answers your team already sends by email.

How retrieval and grounding work

Grounding is a retrieval step before generation: the agent finds relevant approved passages first, then writes an answer constrained to what it found.

  1. 01

    Ingest

    Approved files and pages are processed into searchable passages that keep their source reference.

  2. 02

    Retrieve

    Each question triggers a search across the approved set for the passages that actually match the request.

  3. 03

    Ground

    The answer is composed from those passages, in your language, rather than from general model knowledge.

  4. 04

    Withhold

    When nothing relevant is found, the agent says so and follows the configured next step instead of improvising.

  5. 05

    Log

    The conversation is recorded so the answer and the knowledge behind it can be reviewed.

Source review and human oversight

Grounding is a process, not a one-time upload. These review habits keep answers dependable.

Review answers at the source

Trace a questionable answer back to the material it came from and fix the material.

Scheduled refresh

Re-approve or replace sources when pricing, policy, hours or scope change.

Gap log

Track questions with no approved answer and close them deliberately.

Oversight responsibilities are shared: you own what is approved and current, we own the configuration and the review tooling.

Connects with your favorite tools

Connects with the tools modern teams already use — scheduling, storage, CRM, and ticketing — so conversations turn into actions in your systems.

  • Calendly
  • Google Drive
  • Slack
  • HubSpot

Integrations are configured for your workflow during setup; not every integration is enabled by default.

More about your knowledge setup

What counts as approved business knowledge — additional considerations

Product and service data

Catalog details, options, inclusions and eligibility rules that answers must respect.

Helpdesk and portal content

Support articles and internal documentation you want customers or staff to receive consistently.

Exclusions

Material you explicitly keep out of the answer set, such as drafts, internal-only notes or outdated policy.

Source review and human oversight — additional considerations

Owner per topic

Assign who signs off on each knowledge area so accuracy has an owner.

Spot checks

Sample real transcripts on a regular cadence rather than only after a complaint.

Change process

Agree how updates are requested, approved and published so nothing drifts silently.

Connects with your favorite tools — additional considerations
  • Salesforce
  • Zendesk
  • Shopify
  • WordPress
  • WhatsApp
  • Zapier
  • Microsoft Teams
  • Discord
  • Stripe
Why grounding supports every channel

Support, sales, phone, SMS and digital human experiences all read from the same approved set, so improving knowledge improves every surface at once.

Support

Consistent policy answers instead of per-agent interpretation.

Sales

Qualification that respects real scope, service areas and inclusions.

Phone and SMS

Callers and texters get the same answer a website visitor reads.

Digital human

Guided explanations stay inside approved material.

A practical first knowledge set

You do not need to upload everything. A focused first set is faster to launch and easier to keep accurate.

  • Your top 30 to 50 recurring questions with their approved answers
  • Policy documents that dictate what the agent may commit to
  • Service or product descriptions for the workflow you are launching first
  • The escalation rules that define when a person takes over
  • A named owner for each of the above
AI Agents Trained on Your Business — for Support, Sales, and Lead Capture

Train agents on your website, documents, FAQs, products, and workflows. Deploy them across chat, portal, digital-human, phone, and SMS experiences — with usage and answer quality visible from the platform. Crawl websites, upload docs, FAQs, and product info — and publish answers grounded in your approved knowledge.

  • Supported source types: .pdf, .docx, .csv, .html, FAQ
  • Approved business knowledge — files, selected website pages, FAQs and product data you sign off on.
  • Source review — trace any answer back to its source, then correct the source itself.
  • Human handoff — configured triggers move the conversation to your team, not a guess.
  • Configured workflows — routing, destinations and escalation are defined with you during setup.
  • Consistent answers — train agents on one knowledge base so website chat, portal, phone, and SMS all give the same approved answer.
Train, connect and customize, then deploy

No coding required. Train, customize, and deploy in minutes.

01 · Train Your AI

Upload documents, crawl your website, add FAQs and product info.

02 · Connect & Customize

Integrate with Calendly, Google Drive, CRMs. Customize personality.

03 · Deploy Everywhere

Go live on your website, phone lines, and messaging platforms.

Usage and answer quality visible from the platform
  • Chat + Portal — Phone & SMS channels
  • Grounded — Knowledge-based answers
  • Credits — Usage tracked per channel
  • Audit — Review & improvement workflows
Answers come from approved business knowledge

Every agent answers from sources you approve — uploaded files, selected website pages, FAQs and product information — so replies match the way your business actually operates.

Approved sources only

You choose which documents, pages and FAQs the agent may use. Content you have not approved is not part of the answer set.

Source review

Review transcripts and the knowledge behind answers, then correct or retire a source when policy, pricing or process changes.

Human handoff

Configured escalation rules route sensitive, unclear or high-value conversations to your team instead of guessing.

Service scope and configuration

An AI agent is only as trustworthy as the knowledge behind it. Grounding means every answer is retrieved from sources your business has approved — and that you can review, correct and update as the business changes.

  • Approved files, pages, FAQs and product data
  • Retrieval-grounded answers, reviewable at the source
  • A defined update and oversight process

Business knowledge questions

What is grounding in plain terms?

The agent searches your approved material for relevant passages and writes its answer from those passages, instead of answering from general knowledge.

Which sources can we use?

Uploaded files, selected pages from your website, FAQs, product and service data, and helpdesk or portal content you approve.

How do we keep knowledge current?

Assign an owner per topic, re-approve sources when something changes, and use transcript review to catch answers that no longer match reality.

What happens when no approved source covers a question?

The agent states that it does not have that information and follows your configured next step, such as capturing details or escalating to your team.

Do we have to upload our entire knowledge base first?

No. Most first deployments start with the recurring questions and policies for one workflow, then expand as that workflow proves out.

Bring your policies, FAQs and product information. We will map the first knowledge set, the owners and the review cadence.