Service flow: how to design from scratch

What is service flow and why design it
Service flow is the path that a demand takes from entry to closing: which channel receives, who takes over, when to transfer, who schedules, what deadline, and where to register. It organizes responsibilities, priorities, and rules for transshipment between bot, AI, and human. Without a defined flow, calls fall into generic queues, conversations are transferred manually, and the team wastes time asking who should answer.
Coordinators who receive requests on WhatsApp, Instagram, e-mail, and chat at the same time face lost conversations, clients repeating information, and teams overwhelmed by the lack of clear distribution criteria. The service flow solves this by documenting entries, screening, assignees, scheduling, and SLA in a single configurable process.
This guide guides the flow design from scratch, without requiring a complex system, starting with manual mapping and converting each step into an executable rule.
How to design service flow in six steps
1. Map all entry points and contact reasons
List where customers log in: WhatsApp, Instagram Direct, Messenger, email, website chat, phone. For each channel, identify the five most frequent contact reasons by looking at conversations from the last 30 days. Group similar messages: product question, order status, technical problem, complaint, compliment.
Create a table with channel, reason, and estimated volume. This survey reveals where the highest volume of demands is and which require immediate response. Reasons that appear in more than 30% of conversations on a channel are priority candidates for automatic screening.
If the company serves multiple channels without a unified history, start by manually recording the volume of each reason in a shared spreadsheet over two weeks. This temporary tracking already allows you to identify patterns before implementing the tool.
2. Define which sector or person answers each demand category
For each mapped contact reason, assign a responsible person: sales, support, financial, retention, legal. Avoid generic categories such as “general care”. Whenever possible, direct it to those who have the knowledge and autonomy to solve without the need for a new transfer.
Identify cases where two sectors need to collaborate. In these scenarios, define an initial assignee and objective criteria for scheduling. Example: technical support assumes a complaint about a defective product, but takes legal action only when the customer mentions a lawsuit or refund greater than the amount of the order.
List those responsible by name or function, not by abstract department. If only three people answer technical questions, write down the names. This clarity prevents conversations from becoming ownerless.
3. Establish automatic sorting and distribution criteria
Screening separates conversations by urgency, type, or customer profile before assigning them to someone. Use keywords, entry time, source channel, and previous history as routing criteria.
Example of simple criteria: messages with “cancel” or “I didn't receive” go to the retention queue; messages with “price” or “deadline” go to sales; messages after business hours enter the waiting queue with an automatic response informing a return.
Omnichannel platforms such as [BRAND] (URL0) allow WhatsApp, Instagram, Messenger, Telegram, email, and website conversations to enter a single unified inbox, with full customer context and attribution rules by sector or skill. This prevents attendants from having to switch between tabs and apps to keep up with demands.
If there is no screening tool yet, use documented manual rules: “Messages received on Instagram about the product go to seller A; complaints on WhatsApp go to coordinator B.” Manual rules are tentative, but they work better than random distribution.
4. Design transshipment rules between bot, AI, and human attendant
Transshipment is the transfer of a conversation from automation to a person. Define in which situations the bot should pass the conversation: when it doesn't understand the question, when the customer explicitly asks for a human, when the demand requires decision-making, when there is a serious complaint, or when the bot's response time exceeds the agreed limit.
Bots based on fixed flows work well for simple screening, data collection, and FAQ response. AI agents interpret natural language and seek answers based on knowledge, but require continuous curation to avoid wrong answers. Many operations combine both: rules bot does initial screening; AI answers frequently asked questions; human assumes negotiation, exception, and conflict.
Document the attempt limit before transshipment. Example: if after three message exchanges the bot didn't resolve it, automatically transfer it to the attendant with the entire context of the conversation preserved. Never make the customer repeat what you've already said.
[BRANDcombines service automations with the possibility of human participation] (URL1), allowing the same workspace to manage bots, AI agents, and human queues without losing history.
5. Map the internal scaling flow
Escalation is the transfer of a conversation between levels of authority or knowledge. Define when a level 1 attendant must pass on the demand to a coordinator, specialist, or manager. Common criteria: amount of the claim above a certain limit, technical impossibility to resolve, legal requirement, request for a business exception.
Set maximum time for each level to respond before climbing automatically. Example: if the coordinator doesn't respond within 2 hours, escalate to manager with notification. This prevents forgotten conversations.
List who can make decisions without asking for authorization. If the attendant can offer a discount of up to 10%, document this in the flow. Autonomy reduces transfers and speeds up resolution.
6. Convert rules to SLA and set up alerts
SLA (Service Level Agreement) is the response time commitment by type of demand. Define first response time (how long until someone or something responds for the first time) and resolution time (how long until the demand closes).
Examples of SLA by channel and urgency:
- WhatsApp and chat: first response within 5 minutes; resolution within 24 hours for operational demands and up to 7 days for cases requiring analysis.
- Instagram Direct: first response within 10 minutes.
- E-mail: first response within 4 hours; resolution within 48 hours.
Set up automatic alerts when SLA is close to expiration. Platforms with [real-time analytics] (URL2) allow monitoring SLA compliance by channel, attendant, and reason for contact, facilitating intervention before the deadline expires.
Test the SLA for two weeks before communicating to the customer. If the team is unable to meet 80% of the goals, review the deadline or reinforce screening and automation.
Visual service flow model
A functional service flow can be represented in a simple diagram:
Entry → Screening → Responsible → Resolution or Scheduling → Closing
Practical example for e-commerce:
- Login: Customer sends a message on WhatsApp or Instagram.
- Automatic screening: Bot identifies the keyword (“tracking”, “exchange”, “product question”).
- Distribution: Demands about tracking go to the post-sales line; product questions go to sales; exchange requests go to SAC.
- Initial service: Level 1 attendant responds with a deadline and next steps.
- **Scheduling (if necessary) **: If the customer asks for reimbursement above the order amount, schedule it to the coordinator.
- Closing: Register the solution in the CRM, mark the conversation as resolved, and send a satisfaction survey.
Document the flow in a visual format (flowchart, Miro, Lucidchart) or textual (spreadsheet, Notion, shared document). The important thing is that anyone on the team can consult and understand the path of each type of demand.
How does the flow work outside office hours
Outside of working hours, define what happens with received messages. Common options:
- Automatic reply informing return time: “We received your message. Our office hours are Monday through Friday, from 9 a.m. to 6 p.m. We will return by the next business day.”
- Immediate screening and waiting queue: Bot collects the contact reason and adds a conversation to the corresponding queue for opening the next working day.
- AI support for frequently asked questions: AI agent answers questions about hours, address, exchange policy, and order status; demands that require human decision enter the line.
If the company does not operate 24 hours a day, never leave a message unanswered or promise an immediate return that will not be fulfilled. Transparency about the deadline reduces anxious recontact.
Where human service enters the automated flow
Automation solves repetitive, predictable, and factual demands. Humans assume when there is ambiguity, emotion, exception, negotiation, or decision that requires judgment. Define the action limits of each layer:
- Rules bot: Data collection, screening, objective FAQ, appointment confirmation, order status already tracked.
- AI Agent: Response to questions based on documentation, knowledge base, and company public policies.
- Human Attendant: Complaint, business exception, interpretation of a specific case, empathy in a delicate situation, securities negotiation.
The handoff (passage of context) between AI and human must preserve complete history: previous messages, data already collected, stage of the process. Never make the customer repeat their name, order number, or reason for the contact.
Platforms that offer [unified inbox] (URL3) allow attendants to see everything that was said before transshipment, eliminating rework.
What happens with the demand that needs to continue after the first response
Not every conversation ends in a single message exchange. Demands that depend on another area, technical analysis, or third-party approval require follow-up. Define:
- Who is responsible for following up until closing: The conversation cannot be left without an owner after the transfer.
- Where to register intermediate status: CRM, internal system, or observation field on the service platform.
- When to notify the customer about the going: Do not leave the customer without news for more than 48 hours. If resolution takes time, send interim updates.
- How to alert the team about due deadlines: Automatic alerts reduce forgetfulness.
Companies that integrate service with sales CRM are able to follow the complete customer cycle in one place, avoiding duplication of information and loss of context. [BRANDincludes sales CRM with opportunity pipeline] (URL4) in the same service workspace, allowing the team to view conversation history, business proposal, and negotiation status without changing systems.
How to know if the service flow is working
Flow health indicators:
- First response time by channel and type of demand: How long the customer waits for someone or something to respond for the first time.
- Percentage of conversations resolved without transfer: The higher, the better the initial screening.
- Average number of transfers per conversation: More than two transfers indicate a poorly designed flow.
- 24-hour contact rate for the same reason: Customer who returns with the same question was not answered the first time.
- SLA compliance per attendant and per queue: Identify bottlenecks before becoming a complaint.
Review these indicators weekly for the first 60 days after implementing the flow. Adjust screening, assignees, and scheduling criteria as needed.
Example: What changes when the flow ceases to be manual
**Before the documented flow: **
- Messages come through WhatsApp, Instagram, and email without organization.
- Attendants ask in the internal group “who takes this customer?”.
- Conversations are transferred two or three times until they reach the one who really solves it.
- Customer repeats information with each transfer.
- There's no SLA registration or delayed demand alert.
- The team works harder, solves less.
**After the flow implemented: **
- Bot performs initial screening and redirects to the corresponding queue.
- The attendant receives a conversation with full context: source channel, identified reason, previous history.
- Transfers only take place when necessary, with defined criteria and preserved context.
- Automatic alert notifies you when SLA is close to expire.
- Indicators show volume per queue, response time per attendant, and resolution rate by type of demand.
- Team resolves more conversations in the same working time.
The flow that matters is what continues after the conversation
The call flow doesn't end when the conversation closes. What happens to the information collected? Does it feed CRM, trigger commercial action, generate internal ticket, update knowledge base, or get lost in chat history?
Companies that connect service, CRM and automation are able to transform conversations into sales opportunities, churn alerts, and strategic data. BRANDwas designed to centralize this cycle: conversations from multiple channels feed the CRM, triggers trigger automatic actions, and analytics reveal patterns that guide decisions.
Designing the service flow is the first step in operating with predictability. Documenting responsibilities, criteria, and SLA reduces improvisation, accelerates response, and allows you to scale without losing control.
Frequently Asked Questions
**How many service queues should an operation have? ** The number of rows depends on the number of sectors with different responsibilities. It is recommended to start with three to five rows (sales, support, financial, retention) and adjust according to volume and specialization. Single queue simplifies distribution, but increases the chance of an uninformed attendant taking on the wrong conversation.
**How to define the first response SLA? ** Analyze customer expectations by channel: chat and WhatsApp require a response within 5 minutes; Instagram Direct within 10 minutes; email within 4 hours. Test the SLA internally for two weeks before communicating to the public. If the team does not meet 80% of the goals, review the deadline or reinforce the screening.
**Is it possible to design flow without a service system? ** Yes. Start by documenting the flow in a spreadsheet or diagram: input channel, reason, responsible, deadline, and next action. Implement rules manually until volume justifies automation. Documented manual flow already reduces unnecessary transfers and ownerless conversations.
**When should the bot transfer to a human attendant? ** Transfer when the bot doesn't understand the question after three attempts, when the customer explicitly asks for a human, when there is a complaint, value negotiation, business exception, or decision that requires judgment. Never keep the customer stuck in the bot due to the obligation to deflect.
**How to prevent the customer from repeating information in transfers? ** Configure handoff with context passage: message history, data already collected, process stage, and reason for the transfer must accompany the conversation when changing responsible. Platforms with a unified inbox provide complete context for the hiring attendant.
**What is the difference between scheduling and transfer? ** Transfer moves the conversation between sectors with different responsibilities (sales for support). Scheduling moves between levels of authority within the same sector (attendant to coordinator). Both must preserve context and have documented objective criteria.
**How to measure if automatic screening is working? ** Track the percentage of conversations distributed correctly at the first assignment and the number of manual transfers after screening. If more than 30% of conversations need to be reassigned, review keywords, routing criteria, and contact reason mapping.
**Is it possible to have a different SLA by type of customer? ** Yes. Companies can define a differentiated SLA for VIP clients, serious complaints, or regulatory demands. The important thing is to have objective classification criteria and to ensure that exceptions do not block the overall queue. Set up specific alerts to track compliance with each SLA level.
References
- [Service flow: how to draw yours from scratch (with template) | Octadesk] (URL5)
- [URA of service: how to bring the customer to the right sector | Omnichannel] (URL6)
Share
Open Instagram and paste the link in your post/story.