Skip to content
Contact-Centre Automation

Contact-Centre Automation

You look after the Unified Contact Center and are tired of a supervisor dragging every WhatsApp conversation onto somebody’s desk by hand. This guide and its two parts build the one rule most contact centres need — these chats go to that team — in a way the routing engine will actually run, and shows what BigLedger does with a new conversation before you do (less than you think). Building it takes about twenty minutes; with a bot, seeing it fire takes up to fifteen more.

The procedure at a glance. It depends on whether a bot is linked to the endpoint the customer writes to.

With a bot linked, the rule is built on its own page, Contact-Centre Routing with a Bot:

  1. Create the skills and add them to the default queue (Step 1).
  2. Link each agent to a skill, and check they hold the applet at rank MEMBER (Step 2).
  3. Create a CUSTOM rule under SKILL_BASED_TASK_ASSIGNMENT_AUTOMATION_RULE, with one condition (Step 3).
  4. Give it the ASSIGN_TO_TEAM action and the team (Step 4).
  5. Link the rule to the queue — the step everyone misses (Step 5).
  6. Send a test message and wait up to fifteen minutes for the poll (Step 6).

With no bot linked, a different rule set routes the task the moment it is created, and Contact-Centre Routing with No Bot builds it: make sure the ASSIGN_TO_TEAM action exists, create TASK_TEAM_ASSIGNMENT_AUTOMATION_RULE_SET, add a CUSTOM rule on the number’s virtual contact with the team as its action, and send a test message.

Read the next section on either path. The two after it explain the engine a bot hand-over runs; the no-bot engine is on the applet page.

Meet GadgetSphere

GadgetSphere Online Sdn Bhd (GSO) answers customers on WhatsApp and web chat from 9 to 9, with six agents in two teams: Orders (four people — where is my parcel) and Technical (two — the laptop will not boot). About 250 conversations arrive a day, 900 on a launch day, and today a team leader hands every one out by hand.

What every tenant already has

The contact-centre tables are populated by the platform’s database migrations, so your tenant has routing configuration before anyone opens the applet. Unless somebody here has already been in and changed it, this is exactly what you will find:

Seeded recordCodeWhat it means in plain English
One queueDEFAULT_QUEUEThe holding area every task goes into
One skillDEFAULTA label a task can carry and an agent can hold. It points at the default queue
One rule set (the applet calls it an Automation Rule)SKILL_BASED_TASK_ASSIGNMENT_AUTOMATION_RULEThe family of rules a bot hand-over uses, unless its Assign To Agent element names another
One ruleDEFAULT_RULEA parent rule with no conditions at all, priority 10. The engine uses it by name when nothing else matched
One rule actionDEFAULT_RULE_ACTION_LINK_…Ties the default rule to the action below
Four actionsASSIGN_TO_ANY_AGENT_WITH_LEAST_TASKS (the default), ASSIGN_TO_AVAILABLE_AGENT_WITH_LEAST_TASKS, ASSIGN_TO_ANY_AGENT_AT_RANDOM, KEEP_IN_THE_QUEUEWhat a matched rule does. “Least tasks” means fewest assignments in the last 24 hours; “available” means the agent’s browser is connected and marked available right now; “keep in the queue” puts the task back for the next run

So out of the box the intent is: every task goes to whichever agent has had the fewest tasks today. Read on for why, out of the box, that never happens.

How a chat becomes a task on someone’s desk

    flowchart LR
  AR[Automation rule set]
  R[Parent rule]
  RC[Child rule = one condition]
  RA[Rule action]
  A[Action]
  Q[Queue]
  QR[Queue–rule link]
  SK[Skill]
  LS[Agent–skill link]
  U[Agent login]
  T[Task]
  TQ[Assignment]
  G[Team]
  R --> AR
  RC --> R
  RA --> R
  RA --> A
  QR --> Q
  QR --> R
  SK --> Q
  LS --> SK
  LS --> U
  T --> Q
  TQ --> T
  TQ --> U
  TQ --> G
  

Every arrow is a link between records. Read it left to right with a conversation in hand:

  1. The chat-flow creates the task. This section describes a bot hand-over; a task from an endpoint with no bot is routed differently (the warning below). The bot flow that answers the customer has an element — assign to agent or invoke automation rule — and when the conversation reaches it, a task is created carrying the element’s skill and a bag of facts about the conversation: the channel code, the endpoint, the virtual contact, the project, the skill.
  2. The skill decides the queue. The task’s queue is the skill’s queue; a skill with no queue → the default queue. No skill on the element → no task at all.
  3. The task waits for the poll. Creating the task drops a row on a processing queue. The platform’s own round reads that queue on every pass, ten tasks at a time, so the wait is one round of the platform clock, longer behind a backlog. The */15 row the Scheduler shows for the poll is not what does the work: as installed it carries no settings, fires, and reads nothing (how a task is routed; what a scheduled job’s row does).
  4. The engine picks a rule. It collects the rules linked to the task’s queue (the Rule tab on the queue, under Task Router), keeps the parent rules in the rule set the task carries, sorts by priority — highest number first — and takes the first whose conditions all hold. A parent rule with no conditions never holds; the seeded default rule is reached only by name, when no linked rule matched.
  5. The rule’s action runs. For a STANDARD rule the engine reads the action code and does that. For a CUSTOM rule it ignores the action code and reads the target: a team, or one person.
  6. Who counts as an agent. For the “assign to agent” actions the candidates are the people linked to a skill on the task’s queue (Configurations → Agents → Skill) who also hold the applet at rank MEMBER. Nobody linked → the action fails → the default rule is tried with the same empty list → the task stays CREATED, on no queue: only Task Router → Tasks shows it.
  7. The assignment is written: the task’s status becomes AGENT_ASSIGNED or TEAM_ASSIGNED, and its events record which rule matched (RULE_MATCHED) and what happened.

Point 6 is where a fresh tenant stops. The seeding links no agent to a skill and no rule to the queue, so the engine finds no candidates and leaves the task where it was, with no error. Step 5 fixes that for your team rule, which needs no agent links; Step 2 is the safety net for every task your rule does not match.

Two of the three Automation Rule choices on the screen — TASK_TEAM_ASSIGNMENT_AUTOMATION_RULE_SET and OUTBOUND_TASK_ASSIGNMENT_AUTOMATION_RULE_SET — are never seeded. A bot hand-over reads only the set its Assign To Agent element names, the skill-based one unless another is picked, so put its rules under SKILL_BASED_TASK_ASSIGNMENT_AUTOMATION_RULE.

The team-assignment set is not idle, though: it routes a task from an endpoint with no bot linked, at the moment the task is created, and with no such set that task waits unassigned for a person to pick it up. Its rules are in Unified Contact Center → A task from an endpoint with no bot, and how to send every chat on one number to one team without a bot is Contact-Centre Routing with No Bot.

What a condition can test, and what an action can do

A condition is a child rule: a field, an operator, a value. The screen offers four fields — Channel, Project, Virtual Contact and Skill — and six operators: equal, not equal, less than, less than or equal, greater than, greater than or equal. For these four fields only equal and not equal make sense. All conditions on one rule must hold together (AND). There is no “contains”, no “or”, and no condition on the message text.

An action is what a matched rule does, and the two places that offer one do not agree. The four seeded actions run through a STANDARD rule; the Action screen is off the menu, so they are all you have unless someone created more. “At random” always picks the first candidate.

The Rule Action drop-down lists the same action records, and gives ASSIGN_TO_TEAM a team picker. ASSIGN_TO_TEAM works only on a CUSTOM rule, where the engine reads the team you picked; on a STANDARD rule the rule fails over to the default. CREATE_ISSUE never runs (LB8-P05). Which codes the engine handles is on Core concepts.

So the rule you want — WhatsApp goes to Technical — is a CUSTOM rule with a Channel condition and an ASSIGN_TO_TEAM action, linked to the default queue. That is the procedure with a bot.

What success looks like

Whichever path you built, open the test task under Task Router → Tasks and read its events from the bottom: TASK_CREATED, then RULE_MATCHED naming your rule, then TEAM_ASSIGNED. With a bot, the last two arrive within fifteen minutes of the message; with no bot, as soon as the task appears. Each sibling page says what a missing line means on its path (with a bot, with no bot).

Common mistakes

SymptomCauseFix
Every task stays CREATED, and no TEAM_ASSIGNED or AGENT_ASSIGNED line follows RULE_MATCHEDWith a bot: no agent is linked to a skill on the queue, so every action finds no candidates. With no bot: no rule in the team-assignment set assigned itWith a bot: Step 2 — and check the applet rank is MEMBER. With no bot: Steps 2 and 3. Every cause of this symptom, bot or no bot: A chat came in and nobody got the task
Old tasks pile up and nothing ever removes themA sweep on every round of the platform clock copies tasks untouched for fifteen days onto an expired-task queue, but the step that would close them is switched off, so nothing is closed, deleted or clearedClosing a task is a person’s job, with no safety net. Make complete or cancel the last action of every conversation, and give a supervisor an end-of-shift pass with the bulk COMPLETE / CANCEL actions. It is working when the queue empties at the end of a shift — the Task and Agent reports count the same open tasks

The mistakes that belong to one path are on its page: with a bot, with no bot.

Related documentation

Last updated on