PRODUCT FAQ
Get to know Growze. Start here.
From choosing a plan to daily operations, integrations and customer growth.
Explore capabilities, billing, launch preparation and common questions by topic.
PRODUCT OVERVIEW
Product overview
Understand the platform, agents and the right edition for your team.
What is Growze, and how is it different from a chatbot?
Growze is an AI workforce platform for customer growth. It connects lead generation, engagement, customer service, sales conversion and follow-up. Beyond answering questions, it captures customer needs as leads and records that a team can act on. Company knowledge, conversation context and human collaboration support the next step. CRM, order and booking actions require integrations and permissions defined in a custom project.
How do Social Media and Messaging differ?
Social Media focuses on comments, direct messages and lead capture; Advanced adds competitor comparison, prospect discovery and analytics. Messaging Basic provides AI service, a knowledge base, customer profiles and human collaboration. Enterprise adds intent recognition, qualification, standard follow-up and results reconciliation. Choose Custom Messaging when you need dedicated agents, external systems or workflows designed around your business.
Which businesses is Growze suitable for?
Growze suits teams handling regular comments, direct messages or messaging inquiries and turning those interactions into customer relationships. Examples include matchmaking, education and cross-border commerce. Start with a high-frequency scenario that has clear customer questions, reliable product and service knowledge, and an agreed human handover process. Course inquiries, product and delivery questions, or early-stage advisory consultations are useful starting points.
How do the five modules work together?
Lead generation identifies prospects; engagement starts relevant conversations; service answers questions; conversion supports purchases or bookings; follow-up records evolving needs. Shared customer context, knowledge and conversation history connect these stages. For example, a social inquiry can become a qualified lead, a booking conversation and later a service follow-up. Available capabilities depend on the edition, channels and configuration; automated actions across systems need separate confirmation.
Why describe Growze as an AI sales teammate?
The aim is to turn conversations into useful business outcomes, from first contact through post-purchase service. AI can handle repeated inquiries, organize customer needs and help teams follow up with context. Evaluate results through qualified leads, bookings, orders, renewals or completed services. The vision is Make Growth Easy, not a guarantee of sales. Outcomes depend on your business, data and collaboration with people.
Is Growze an AI model or a business platform?
Growze is a business application and agent platform built on AI models. Models understand and generate language; Growze organizes company knowledge, customer context, channel messages, permissions and business actions into operational workflows. Teams use the workspace to review conversations and improve agents. Model selection, hosting, access and costs are agreed in the technical solution. Using Growze does not mean buying or training a foundation model.
What are an Agent and an AI workforce platform?
An agent is an AI work unit with a specific goal. An AI workforce platform, or Agent OS, organizes agents, knowledge, customer data, workflows and people. Service agents answer questions; marketing and sales agents qualify needs; custom agents handle tasks such as member services or assessments. Each requires its own role, knowledge scope, inputs, outputs, permitted actions and acceptance criteria. Sharing context does not automatically grant another role’s permissions.
What experience does the Growze team bring?
Core team members come from Tencent, Alibaba and ByteDance, averaging over 15 years of hands-on experience and more than 10 years working together across product, algorithms, engineering and marketing. Their experience includes products serving hundreds of millions of users, high-performance computing and B2B services. This informs practical workflows, model and cost choices, integrations and implementation focused on business outcomes.
LEAD ACQUISITION
Lead generation
Recognize intent and capture leads your team can follow up.
How does Growze identify potential customers?
Within connected social comments and direct messages, Growze recognizes needs and purchase intent, asks relevant follow-up questions and organizes the information into leads. It answers the original question before asking for additional details, records confirmed requirements and leaves unknowns for the team to verify. Social Advanced also includes competitor comparison and prospect discovery. Coverage depends on channel capabilities and project configuration.
What information can be captured in a lead?
Useful fields include the product or service of interest, use case, needs, confirmed requirements, unresolved concerns, intent and the next agreed action. Contact details and preferred times should come from information the customer provides or data the business is authorized to use. Keep unconfirmed details marked as unconfirmed. Additional business fields, CRM synchronization and complex routing rules can be assessed as customization.
How do we judge lead quality?
First define a qualified lead with your business team: service fit, location, timing, budget where relevant, specific needs and willingness to continue. Then evaluate conversations and tags against those criteria. A like or message is not automatically buying intent. Review lead volume, qualification rate and downstream conversion together, using a consistent reporting period and sampling real conversations to check AI judgments.
When is Social Advanced useful?
It suits teams already handling social inquiries that want to understand customer comparisons, differentiate their offering and discover more prospects. Bring your target channels, relevant competitors and real customer questions to the evaluation. Confirm the data scope and deliverables before purchase. Competitor comparison and prospect discovery do not provide unrestricted access to every user on any platform.
CUSTOMER ACTIVATION
Engagement
Build outreach, campaigns and operations around customer needs.
What does customer engagement help with?
It helps customers who have already interacted, inquired or expressed interest understand relevant products, offers or services and decide on a next step. For example, explain trial-class content, available times and booking rules to someone who asked about a course. Address decision barriers, explain eligibility and limitations, and let the customer choose. Outreach methods depend on the channel and the agreed solution.
What information is needed before running a campaign?
Prepare the audience, dates, price, eligibility, capacity, booking process, rescheduling, cancellation and refund terms. Explain whether offers can be combined, who maintains the rules and where availability is checked. Update changing offers promptly so AI does not reuse outdated terms. Actual bookings, capacity checks and offer issuance require the relevant business systems and action permissions.
Can Growze send proactive messages?
Only within the channel’s messaging rules, account permissions and customer authorization. Confirm permitted audiences, opt-out handling, triggers, timing, frequency and stopping conditions. Some channels restrict messaging windows or require templates. Connecting a channel does not authorize unlimited bulk messaging. Automated outreach workflows and their stopping rules must be configured within the agreed project scope.
How should engagement results be measured?
Use business-relevant outcomes such as meaningful replies, further inquiries, contact details provided voluntarily or bookings, alongside customer feedback. Agree on the audience, reporting period and success event before starting. Compare similar groups or time periods using consistent definitions and review conversations to understand why progress stalled. Linking outreach to orders or bookings requires reliable result data from the relevant business systems.
Are broadcasts, welcome messages and keyword replies supported?
Private or group broadcasts, Moments posts, welcome messages, keyword replies and content libraries can be assessed within a custom project. Availability depends on each channel and account permissions. Define the audience, approved content, sending time, frequency, opt-out handling and human follow-up. These workflows are not included in every standard plan and do not permit unlimited messaging.
CUSTOMER SERVICE
Customer service
Prepare reliable knowledge for multilingual, multimodal inquiries.
What questions can the AI answer?
It can answer product, service, usage, pricing-policy and common pre-sales or after-sales questions using current company knowledge. Order status, tracking, bookings and live membership information require connected systems; static documents cannot establish current status. Special discounts, disputes and requests beyond standard rules should go to authorized people. Separate knowledge questions, live-status checks and decisions when defining the service scope.
What should we prepare for the knowledge base?
Prepare accurate product and service descriptions, FAQs, pricing and campaign rules, after-sales policies and statements the AI must not promise. Mark scope, version, effective date and the responsible owner. Explain ambiguous rules with examples. Test real questions before launch, review conversations afterward and replace outdated material. Supported formats, import methods and maintenance workflows are confirmed during setup.
Which languages and message types are supported?
Growze supports multilingual communication and text, images, voice and files, subject to channel capability. Mini-program and contact cards may be supported in WeChat or WeCom projects. Test real samples for terminology, tone, language detection, formats, file sizes, accents and noise. Confirm human handover languages and verify product, shipping and return information. Unclear or specialized content should be clarified or handed to people.
SALES CONVERSION
Sales conversion
Recognize buying intent and agree on the next step.
Can Growze replace a sales team?
Growze can handle repeated questions, identify needs, organize information and progress standard steps so people can focus on high-value conversations. Complex proposals, negotiations, exceptional terms and trust-dependent decisions still require human judgment. Define responsibilities together and expand automation based on evidence from real use. Treat Growze as a teammate, not a replacement for every sales role.
How does Growze recognize and progress purchase intent?
It considers the product of interest, intended use, price concerns and willingness to act, then responds using company knowledge. Messaging Enterprise adds intent recognition and qualification. Clarify the use case, address concerns with verified terms and ask whether the customer wants a product page, further advice or a booking. The next step should follow your business rules and the customer’s explicit choice.
Can AI place orders, book appointments or change orders?
These are business actions requiring custom agents, external APIs and appropriate permissions. A knowledge-base service alone does not grant write access. Verify identity, the target record and conditions, explain and confirm key actions, then call authorized interfaces. Report the actual result: a submitted request is different from a completed action. Define success feedback, failure handling and human escalation before launch.
Does Growze guarantee a conversion rate?
No. Results depend on the offering, customer sources, pricing, workflows and human collaboration. A demo or case does not guarantee outcomes for every business. Define qualified leads, attribution windows, data sources and team involvement, then compare similar scenarios and review customer feedback. A pilot can measure inquiry quality, leads, bookings and sales. Performance-based agreements need explicit attribution and reconciliation rules.
CUSTOMER FOLLOW-UP
Follow-up
Track needs and status across repeat purchases, renewals and CRM.
What is recorded for customer follow-up?
Customer needs, tags, current status and agreed follow-up help the next team member understand the conversation. Record confirmed conditions, open questions, bookings or submitted requests, who follows up, when and what materials are needed. Keep pending states distinct from completed work. Additional fields and a broader view combining orders, membership or fulfillment can be assessed through system integration.
How do standard and custom follow-up differ?
Messaging Enterprise includes standard follow-up and results reconciliation within existing capabilities. Custom follow-up covers triggers from orders, membership or approvals, multiple systems, special review steps and bespoke task rules. Map the process from trigger to completion and identify the data, people and systems required at each step. Confirm the exact standard configuration before purchase; assess complex workflows through Custom Messaging.
Can Growze help with renewals, repeat purchases and retention?
Yes, by discussing experience, usage concerns, expiry and new needs rather than only sending promotions. Resolve obstacles, explain valid options and confirm the customer’s wishes. An inquiry is not consent to renewal or a charge. Automatic expiry detection, balance lookup or extensions require connected data and explicit business rules. Outreach must also follow the channel’s conditions.
Can follow-up records sync with our CRM?
CRM integration can be evaluated within a custom project. Agree on customer identifiers and deduplication, field mapping, read and write directions, update timing, conflict rules and failure notifications. Test these behaviors during integration acceptance. Feasibility depends on available APIs and authorization; similar names alone should not be used to merge customer records.
DAILY WORKSPACE
Workspace and knowledge
Explore daily work, knowledge maintenance, team roles and reporting.
What can the workspace show?
Depending on edition and configuration, it can show service metrics, growth trends, follow-up tasks, channel status, recent conversations and agent activity. Customer, tag, lead and order information helps teams decide what to do next. Business-result reporting requires integrations or result submissions from your systems. Confirm modules, filters, exports and metric definitions during selection; a preview image does not imply every feature is included.
How can teams review customers, tags and conversations?
Customer management, tags and conversation history help teams continue work and find recurring questions or weak answers. Before taking over, review needs, prior replies and pending actions. Check AI classifications and keep unconfirmed information separate. Review samples by customer, channel, time or status where supported. Filters, saved views, retention and export options depend on the plan; cross-channel history requires reliable identity matching.
How is the knowledge base maintained?
Upload, import, edit and delete capabilities depend on the actual system interface and supported formats. Assign a business owner and organize content by product, policy and effective date. When rules change, update old documents, scripts, agents and workflows together, then test real questions. Growze can assist within the agreed scope. Dynamic inventory and order data should come from business systems, not a document repository.
Is there a limit on knowledge-base entries?
Custom plans do not cap the number of knowledge entries. Standard allowances depend on the selected plan. Unlimited entries do not mean unlimited storage, file size, processing capacity or model usage; confirm those separately. Large knowledge bases need clear categories, versions and owners to avoid duplicate or conflicting rules. Provide sample documents and expected scale to evaluate retrieval and answer quality.
How do team access and permissions work?
The system supports sign-in, sign-out and account assignment. Define administrator, AI operations, sales, service and read-only responsibilities, then confirm the roles, permissions and seats offered by your plan. Limit configuration, customer access and export rights to the roles that need them. Human seats are different from the channel accounts used for plan billing. Organization structures, approvals and audit capabilities require separate confirmation.
CHANNELS & ACCOUNTS
Channels and accounts
Confirm channel scope, collaboration and account responsibilities.
Which social and messaging channels can be connected?
Social channels include Douyin, Xiaohongshu, WeChat Channels, Kuaishou, TikTok, Instagram, Facebook, YouTube and X. Messaging channels include WeChat, WeCom, WhatsApp, Messenger, Telegram and LINE. Confirm account types, access permissions and whether you need comments, private messages or other interactions. Test receiving, replying and human handover on each channel; capabilities are not identical across platforms.
Can our website, mini program or app connect?
These are integration directions Growze can assess around your consultation entry points and business workflows. Explain where customers start a conversation, whether login, product or order context is needed, and what should happen afterward. The solution defines required development, available connection methods and delivery scope. Connecting a custom application may require implementation work.
How should we choose the number of accounts?
List the actual channel identities you plan to connect; do not confuse them with human seats. Standard plans default to one account. The total is the current monthly unit price multiplied by account quantity and purchased months. Included replies are shown per account per billing cycle. Confirm channel identity mapping, cross-account sharing, migration and quantity changes before purchase. Custom account scale is agreed in the solution.
What is required before connecting a channel?
Prepare the channels, account quantities, authorized administrators and target conversations. Confirm account types and platform permissions, authorize through the formal setup process and define permitted data and actions. Test a complete flow: incoming messages, appropriate replies and human continuation. Platform rules, account status and interface availability determine what can actually be enabled.
Can channels share agents, knowledge and customer data?
A custom project can assess shared agents, knowledge, tags and rules. Shared content does not automatically share identity, permissions or workflows. Keep market, language and channel differences; define reliable customer matching; and test message formats, windows, limits and handover separately. Start with one channel, then expand. Document the sharing scope and access rules; do not act on an unverified cross-channel identity match.
Who supplies and maintains channel accounts?
Usually the business supplies valid, authorized accounts, entity information and API access. Growze handles setup, integration testing and support within the project scope. Agree on owners for approvals, credentials, permission changes, expiry and incident notification. Third-party platforms control reviews, risk checks and restrictions; connecting Growze does not grant additional platform permissions.
PLANS & BILLING
Plans and billing
Understand plan fees, reply allowances and custom-project costs.
How are plans priced?
Standard plans are billed per channel account per month. Live prices and included replies come from the same catalog as the workspace. Choose 1, 3 or 12 months, with one month selected by default. Total price equals the monthly unit price × account quantity × months. Prices are in CNY. Custom development, integrations, third-party fees and service scope follow the formal quote; Custom Messaging is quoted individually.
What does each plan include?
Social Standard includes account management, direct messages, comment replies, lead capture and knowledge-base training. Social Advanced adds competitor comparison, prospect discovery and analytics. Messaging Basic includes AI service, a knowledge base, basic tags, customer profiles and human collaboration. Messaging Enterprise adds intent recognition, qualification, standard follow-up and results reconciliation. Custom Messaging covers agreed modules, dedicated agents, external systems, workflows and reports. Confirm exact configuration and deliverables in the service terms.
How many AI replies are included?
See the live plan cards for the allowance per account, issued each membership billing cycle. Custom usage is agreed separately. The unit is an AI reply, not a customer, lead or entire conversation. One inquiry may involve multiple replies. Estimate usage with real conversation lengths and volume, then confirm any sharing rules. Unused included replies expire at the end of their billing cycle.
How do we buy additional usage?
Buy AI Conversation Packs shared across Social Media and Messaging. You need an active membership in either product to purchase. Current pack sizes and prices are shown in the live catalog and workspace. Compare your remaining included allowance with expected extra replies before choosing a pack. Purchased credits never expire or automatically renew; any remaining balance is retained if membership expires.
How is Custom Messaging priced?
Pricing considers account scale, dedicated agents, workflows, integrations, reports and delivery scope. A performance-based arrangement may be agreed separately. Prepare business objectives, success criteria, existing systems, permitted actions, reporting and deployment requirements. Confirm implementation stages, acceptance, ongoing support and attribution rules. Development, implementation, service and third-party costs follow the agreed quote rather than a standard plan’s monthly price.
What happens after I select Get started?
Choose a plan, account quantity and duration. If needed, sign in with WeChat, then confirm the order and pay in the workspace. The server verifies the actual price. Payment and entitlement fulfillment are separate steps: check the order status until the membership is issued. Channel connection, company knowledge and business validation still need setup. Custom plans require a prior discussion and quote; booking a demo is a separate action.
Can I purchase multiple plans of the same product type?
You can make repeated new purchases of the same plan; each creates an independent membership with its own accounts, validity and included replies. Renew to extend an existing membership instead. Different tiers in the same product cannot be purchased concurrently as new memberships: use the upgrade path for Social Standard to Advanced, or Messaging Basic to Enterprise. Social Media and Messaging are separate products and may both be active.
How do renewal and upgrade differ?
Renewal extends the original membership. Upgrade changes an active membership to a supported higher tier immediately after payment and fulfillment: Social Standard → Advanced, or Messaging Basic → Enterprise. Account quantity stays the same and a new billing cycle begins. Pay the new plan in full first. The selected membership’s old current and future cycles are revoked and remaining included replies expire; other memberships and purchased credits remain. Resolve pending renewal or upgrade orders first. No automatic debit occurs. Advanced and Enterprise have no further upgrade; contact an advisor for Custom.
Is unused value refunded when upgrading?
Yes. Refunds are calculated against the actual amounts paid on the original orders and returned to the original payment method. For the current cycle, its amount is multiplied by the lower of the remaining-time ratio and remaining-included-replies ratio, rounded down to cents; if no replies were included, use time alone. Unstarted future cycles are fully refundable; ended cycles are not. Separate original orders produce separate refund orders, unless the amount is zero. Checkout shows an estimate, recalculated at upgrade activation. Refunds do not offset the new payment. Track them in order details; refund failures do not undo the upgrade. Purchased credits remain. Arrival time depends on WeChat and the bank.
Which AI replies are used first?
Usage first draws from the current membership billing cycle’s included replies. Any shortfall draws from purchased conversation credits. Included replies expire at cycle end. Purchased credits never expire and are shared across Social Media and Messaging. If membership expires, the purchased balance remains and can be used again after membership is restored. The workspace displays included and purchased balances separately.
Is membership required to buy conversation credits?
Yes. You must have an active Social Media or Messaging membership to purchase AI Conversation Packs. Purchased credits are retained after membership expires. Restore an active membership in either product to use the remaining credits again. The purchased balance is shared across both product types.
When do benefits appear after WeChat payment?
An order may first show that payment is received and benefits are processing. Membership validity or purchased credits appear after fulfillment completes. Check membership status, expiry, current-cycle replies and conversation-credit balance in the SaaS workspace. If fulfillment remains pending unusually long, keep the order number and contact support rather than paying again.
How do tokens differ from AI replies?
Tokens measure model input and output; reply counts measure AI responses. Different replies consume different token amounts, so there is no fixed conversion. Standard website plans use account fees and reply allowances shown in the catalog. Custom projects may specify token or other usage billing according to model, context and integration requirements. Confirm units, periods, alerts and overage handling. This explanation does not add an undisclosed token fee to standard plans.
How are custom development and running costs estimated?
Assess one-time work such as business analysis, agent design, integrations, workflows, reports and acceptance separately from ongoing accounts, model usage, hosting, third-party services and support. Also define knowledge preparation, training, optimization and change requests. Provide monthly conversations, knowledge scale, languages, agents, systems and target launch date for an itemized estimate. Custom estimates do not alter standard plan prices or benefits.
How are performance-based results attributed and reconciled?
Agree on valid results, deduplication, attribution windows, AI and human contributions and reliable source data before agreeing fees. Define how refunds, cancellations, duplicates and missing data affect settlement, and set review periods and a dispute process. Chat volume or AI-predicted intent alone is not proof of a sale. Performance fees and revenue share apply only where the custom contract explicitly specifies them.
CUSTOM DEVELOPMENT
Customization and integrations
Connect systems and configure dedicated agents, branding and workflows.
What can be customized or integrated?
Modules, dedicated agents, workflows and reports can connect CRM, products, orders, memberships, services, support tickets, collaboration tools and BI. Define goals, data sources, inputs, outputs and authorized actions for each integration. Specify failure, duplicate-request and human-handover behavior, as well as rollback or compensation for data changes. These are areas for assessment, not a promise that every system has a ready-made connector.
What do we need to provide for integration?
Provide system names, API documentation, field definitions, test environments and business and technical contacts. List permitted reads and changes, process conditions, success states and review requirements. Agree on identity keys, status values and test examples, and separate test permissions from production access. If APIs are unavailable, assess alternatives before committing to schedule and cost.
How is AI’s action scope controlled?
Define which data may be read, what may be changed and which actions need customer or human confirmation. The process is understand the request, confirm permission and execute. Distinguish submitting a request from directly changing a record. Money, benefits and personal-data changes need appropriate authorization. Define stopping points for missing information, denied permissions and uncertain results, with a tested handover path.
What happens if an external system fails?
Explain whether a request has not been submitted, is pending confirmation or has failed. Never describe an unknown result as success. The integration must define which actions can be retried safely and how duplicate changes are prevented. Preserve necessary request details for an authorized person to investigate. Test failure feedback, notification, retry and compensation paths before launch.
Which dedicated agents can be built?
Examples include membership service, order service, product sales, university application guidance and AI assessments. Each requires business and technical evaluation. Define its audience, objective, knowledge, data, inputs, outputs, APIs, confirmation steps and acceptance examples. Shared customer infrastructure does not replace specialized knowledge or testing. Important decisions should be reviewed by people with appropriate authority and expertise.
How are product, inventory, price and order details kept accurate?
Read dynamic facts from connected product, inventory, order, delivery or after-sales systems. A knowledge base explains policies but cannot establish live stock, status or final transaction amounts. Validate identity, record keys, timestamps and field meanings. Write actions need separate permissions and confirmation. If data is unavailable or delayed, explain the limit and escalate rather than guessing. Your business system remains authoritative for prices, stock, payment and order results.
Can we connect through APIs or our own apps?
API and business-agent integration can be assessed for websites, mini programs and apps. Confirm authentication, customer identity, permitted data, request and callback formats, rates, timeouts, duplicate handling and errors. Discuss test environments, examples, any required SDKs, hosting and maintenance. Available API documents and scope are confirmed during technical evaluation; custom systems may still require development.
Can we use our own branding and domain?
Custom solutions may include your name, logo and dedicated domain on agreed pages and entry points. Prepare brand assets and confirm domain ownership, DNS, certificates, hosting and any applicable registration requirements. Assign responsibility for renewals and updates. Branding or a dedicated domain does not automatically mean private deployment, and does not change third-party channel interfaces.
AI & HUMAN COLLABORATION
AI quality and human collaboration
Set answer boundaries, confirmations and human handover rules.
How is answer reliability improved?
Use verified company knowledge, explicit rules, real-question testing and conversation review. Pricing, refunds and timing should come from current policies or business-system results. Before launch, test common, ambiguous and exception cases. During use, sample decision-sensitive replies. After changes, rerun the original and related questions. Human review remains necessary for complex or uncertain cases; no AI system is guaranteed error-free.
How do people take over complex issues?
Conversation management and human collaboration let teams continue when customers request help, information is insufficient, disputes arise or actions fail. Define the responsible role, routing and notification. Pass the question, verified facts, prior actions and pending items so the customer does not repeat everything. Automatic handover conditions and assignment methods depend on the configured solution.
Can Growze guarantee zero errors or hallucinations?
No. Knowledge, context, rules, system checks and human collaboration reduce risk but require validation in each use case. Assign authoritative sources for pricing, stock, orders, refunds and contracts. Configure clarification or handover when knowledge or permissions are missing. Add errors to regression examples and expand automation only as results justify it. Money, benefits and significant commitments should retain appropriate confirmation and review.
What if AI cannot verify an answer?
It should identify the missing information, ask only necessary questions, offer a verifiable help path or hand over to people. Missing company knowledge requires an authoritative answer from the business. Uncertain system results should remain pending until checked; distinguish not submitted, awaiting confirmation and failed. Capture the corrected answer and escalation path afterward to improve future handling.
How are complaints, refund disputes and sensitive issues handled?
AI can clarify the request, organize known facts and explain next steps. Responsibility, compensation, major disputes or decisions beyond its authority belong to authorized people. Define handover triggers and provide the relevant conversation, verified status and pending actions. Do not permit unapproved compensation promises, contract changes or professional legal, medical or financial decisions. Actual assistance follows verified policies, permissions and project configuration.
GETTING STARTED
Launch and collaboration
From pilot evaluation to acceptance, training and improvement.
How do I book a demo and start using Growze?
Book a demo through the website and bring your industry, customer sources, channels, account scale and main problems. For standard plans, choose Get started, sign in and confirm payment in the workspace. Custom projects begin with workflows, APIs, scope and acceptance planning. Before live use, prepare current knowledge, assign a human handover owner and validate channels and scenarios. Trial, support and launch arrangements follow the agreed terms.
What should we prepare and test before launch?
Prepare authorized channels, product knowledge, prices, after-sales policies, brand tone, prohibited commitments and a handover owner. Integrations also need documentation, a test environment and data permissions. Test normal, ambiguous, out-of-scope, malicious and sensitive questions; message handling, identity and handover; and action success, timeouts, denied access, duplicates and failures. Agree on reproducible acceptance cases and launch criteria, then begin with a limited rollout.
How long does launch take, and how is it improved afterward?
Timing depends on channel readiness, knowledge quality, account scale and customization. Confirm scope first, then plan configuration, validation and rollout. Standard setup differs from multi-system development. Start with a defined scenario, test real questions and exceptions, then expand using conversation reviews, knowledge updates, team feedback and business metrics. Support and response arrangements are agreed before work begins.
What are the stages of a custom project?
Typical stages are business and data assessment, solution and quote, contract, configuration and development, integration testing, pilot, training and handover, then launch. Each stage should have clear outputs and acceptance criteria. Growze can help map workflows within the agreed scope, while your team supplies actual policies and business and technical owners. Complex projects benefit from staged acceptance and gradual expansion.
Is training and launch handover available?
Training for administrators, AI operators, sales and service teams can be included in the project scope. Agree on format, sessions and materials. Cover account and channel management, knowledge maintenance, conversation review, lead follow-up, takeover and escalation. Practice a complete scenario with your own business data and appoint owners for knowledge, monitoring and urgent issues. Service response commitments follow the agreement.
Who maintains the system after launch?
Your business maintains policies, product and service information and outcome feedback. Within the agreed scope, Growze supports agent, retrieval and workflow configuration. Establish regular reviews of answer quality, handovers, lead quality, conversion and satisfaction. Provide necessary redacted examples, expected answers and timing, then retest changed behavior. Optimization frequency, changes and response arrangements follow the contract.
What should we confirm before signing?
Share your industry, business model, channels, scale, systems, target automation, budget and launch expectations. Confirm plan, channel accounts, human seats, agents, knowledge maintenance, APIs, training, acceptance and schedule. Clarify billing units, periods, third-party fees, invoices, support commitments and change requests. Also cover data rights, permissions, export, deletion and exit responsibilities. For custom token or performance billing, define usage, attribution and reconciliation separately.
Can we arrange a demo or a small pilot?
Book a demo on the website. To evaluate fit, discuss a pilot with one channel, a valuable scenario and redacted sample data. Pilot duration, fees and support require agreement. Bring real questions and rules, then observe knowledge use, context, handover, status lookup and lead capture. Agree on success criteria and human involvement before expanding. A narrow, measurable workflow is often the best first test.
DATA & PRIVACY
Data security and privacy
Confirm data access, deployment, audit and compliance requirements.
How are security, privacy and deployment agreed?
Evaluate access, authentication, transport, storage, logs, backups, redaction and human access during the project and contract process. Collect only authorized data needed for the task. Confirm where applications, models, databases and backups process data, how long it is retained and how exit and deletion work. Raise private deployment, residency or certification requirements early and verify formal evidence; they are not automatically included.
Can other customers or unrelated staff access our data?
Tenant isolation, roles and API authorization control access in multi-tenant or multi-project systems. Confirm isolation for queries, files and endpoints, internal customer and export permissions, and conditions for operational support access. Specify necessary approval and access records. Exact granularity and audit features depend on deployment; include critical access-control scenarios in acceptance testing.
Can demos and tests use real customer data?
Prefer synthetic or redacted data that preserves the needed structure. If real data is necessary, confirm authorization, purpose, access scope, retention and whether external models or services receive it. Remove unrelated identity, contact, payment and sensitive fields. Agree on deletion and backup handling after testing. Do not supply an unauthorized full production database for a demonstration.
Who owns the data, and can it be exported or deleted?
Ownership, permitted use, export, deletion and backup treatment belong in the contract or data-processing agreement. Confirm who can view and export, available formats and scope, retention periods, deletion verification and handover at service end. Include backup and statutory-retention responsibilities. Raise residency, regulated-industry or special export requirements during solution design rather than assuming them from a feature page.
How is cross-border data handling assessed?
Consider customer location, local requirements, authorization, channel policy, transmission paths and model-provider terms. Identify what data enters, where applications, databases, logs, backups and models process it, and who is responsible for each purpose and retention period. Restrictions on regions, hosting or models should be stated early. Do not assume one arrangement works for every country or region.
Are logs, action tracking and audit reports supported?
Conversation and action records support review. Login, administration, API calls, exports and audit reports can be assessed in the project scope. Specify events, fields, actor identity, confirmation steps, interface results and human handling. Agree on retention, access, exports and redaction. Having conversation history alone does not imply a full audit capability.
Are customer conversations automatically used for model training?
Improving knowledge, workflows and answer quality from feedback is different from training a model on customer data. Define which redacted examples may be used, who can access them, what model services receive content, and the relevant retention and deletion terms. Any training use needs an agreed purpose and authorization. Updating a knowledge base does not mean the model has automatically been trained.
How can Growze support a security and compliance review?
Both teams can review business processes, data flows, permissions, interfaces, hosting and support. Cover collection purpose, access, transport, storage, backups, deletion, action authorization, human escalation and third parties. State sector and regional requirements early, particularly for finance, healthcare, education, minors and cross-border activity. Verify security measures and credentials through formal materials; your legal, security and business owners should review suitability before launch.
INDUSTRY SCENARIOS
Industries and use cases
Choose a useful first scenario and measure business value.
Which scenarios are good starting points by industry?
Choose frequent work with clear rules, measurable value and a human handover path. Advisory services can qualify needs and prepare context for consultants. Education can handle courses, policies, booking intent and member guidance. Cross-border commerce can answer multilingual product, shipping and return questions and query connected order data. Cases illustrate possibilities, not guaranteed outcomes. Use assistance and human review for complex or regulated decisions.
How should an agent be designed for a specific business?
Map the customer journey and objectives, then define the role and permitted scope. Document audience, trigger, knowledge, inputs, outputs and expected outcomes. Specify prohibited actions, confirmation, handover and exception handling. Build normal, edge, failure and multilingual acceptance cases. Maintain industry knowledge, scripts, workflows and tests separately, and share customer context only within channel and data permissions.
TROUBLESHOOTING
Troubleshooting
Diagnose answer, context, knowledge and channel issues.
What should we do about an inaccurate answer?
Keep the necessary conversation and authoritative answer, then identify whether the cause is knowledge, context, channel parsing, live data or agent rules. Escalate decision-sensitive pricing, refunds or benefits to a person first. Correct the source or rule and retest the original and similar questions. For support, provide redacted examples, time, channel, expected behavior and record identifiers; exclude unnecessary sensitive information and credentials.
Why does AI ask for information already provided?
Check identity matching, conversation continuity, information extraction and writes to business systems, not just wording. Verify that the right fields contain the known facts and the next agent can read them. Define when old information or important actions require reconfirmation. Retest multi-turn conversations after fixes. Give people summaries and pending questions during handover, and explain necessary security checks to customers.
Why does AI use old content after a knowledge update?
Check whether the document is active, old versions remain, and the agent uses the intended knowledge space. Outdated scripts, workflows or external system data may also be responsible. Verify scope, dates, duplicates, processing and indexing status. Test the original and rephrased questions, and check live sources for prices or stock. If needed, share document version, update time, agent and redacted examples with support rather than repeatedly uploading duplicates.
What if a channel cannot connect or send messages?
First identify whether authorization, receiving or replying fails, and test a single plain-text message. Check account permissions, approval and status, messaging windows and limits, then credentials, callbacks, subscriptions, network and responses. Verify images or voice separately against supported formats. Record redacted errors, time, channel, account identifier and request ID for support. Avoid repeated operations that might trigger platform restrictions or duplicate business actions.
Channel capabilities, plan scope and custom deliverables follow the formal service description and quote. Products, courses and orders in demonstrations are illustrative.
LET’S TALK
Have a specific use case?
Tell us about your needs. We will help confirm the right plan, integrations and delivery scope.
Explore plans first
