AI-Powered Customer Support Automation for Global Transportation Service

lime · Tech · 2024 · source Agentic: no · promotional bias: high · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Global micromobility customer support

Lime needed to scale multilingual customer support across email, phone, app, and web channels while prioritizing critical safety/compliance issues and reducing manual routing.

Constraints:

Architecture

Lime uses Forethought's support automation suite: Triage predicts language, category, and severity tags to route tickets into queues; Solve interprets chat/email intent and searches knowledge articles plus resolved tickets for automatic answers; Workflow Builder runs RPA-guided workflows for damaged vehicles, charges, payments, and receipts; complex questions route to the help-center web form and human support agents.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  customer-support-channels["customer-support-channels"]
  forethought-triage["forethought-triage"]
  language-and-priority-queues["language-and-priority-queues"]
  forethought-solve["forethought-solve"]
  knowledge-and-ticket-retrieval["knowledge-and-ticket-retrieval"]
  workflow-builder-rpa["workflow-builder-rpa"]
  help-center-web-form["help-center-web-form"]
  human-support-agents["human-support-agents"]
  customer-support-channels --> forethought-triage
  forethought-triage --> language-and-priority-queues
  language-and-priority-queues --> human-support-agents
  forethought-solve --> knowledge-and-ticket-retrieval
  help-center-web-form --> human-support-agents
Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component Role Interacts with
customer-support-channels Email, phone, Lime app, web form, chat widget, and help-center inputs. forethought-triage
forethought-triage Predicts language, category, severity, custom tags, and service levels for routing. language-and-priority-queues
language-and-priority-queues Routes tickets to agents with the right language skills and prioritizes critical accident or compliance cases. human-support-agents
forethought-solve Intent interpretation and automated answer system for chat and email requests. knowledge-and-ticket-retrieval
knowledge-and-ticket-retrieval Search over help articles and previously resolved tickets for automated customer responses. —
workflow-builder-rpa Preset RPA workflows integrated with internal tools for damaged vehicles, charges, payments, and receipts. —
help-center-web-form Fallback intake for complex questions that need human support. human-support-agents
human-support-agents Specialized support agents for complex, critical, or escalated cases. —

Data flow

A customer submits a support request through chat, email, app, phone, or web. Triage predicts language, category, and severity, applies custom tags and triggers, and routes the case to the right queue or priority. Solve can answer simple chat/email intents by searching articles and prior resolved tickets, Workflow Builder handles preset RPA flows, and complex cases fall back to the help-center web form and support agents.

3 supporting quotes
“Triage layers the language and category classifications, and works internally to determine different service levels based on severity.”article, stated
“With Forethought Solve, they’re able to interpret the intent of a customer’s request, through their chat widget or email”article, stated
“Using Workflow Builder, Lime has also built out Robotic Process Automation (RPA) guided workflows.”article, stated

Why this architecture

Lime needed classification, self-service, and workflow automation because manual all-purpose queues could not scale across countries, languages, and criticality levels. Triage fixes routing and priority, Solve deflects simple multilingual cases with retrieval, Workflow Builder automates repeatable internal tasks, and human agents remain the backstop for complex inquiries.

3 supporting quotes
“Every support agent handled every type of ticket because there was nothing in place to prioritize and route specific tickets to specific agents.”article, stated
“Being a global company, Lime receives support tickets in many different languages.”article, stated
“If the customers’ question is complex, they’re automatically routed to the help center web form”article, stated

Models used: Forethought AI models

Scale

Alternatives considered (roads not taken)

All-purpose manual support queues

Rejected because: Manual queues failed to route tickets by expertise, language, or criticality.

“Every support agent handled every type of ticket because there was nothing in place to prioritize and route specific tickets to specific agents.”article, stated

Agents toggling between Google Translate and cases

Rejected because: Triage and Solve support language-specific routing and multilingual service.

“agents often toggled between Google Translate and their cases to respond to common inquiries.”article, stated

First-in-first-out handling for accidents and compliance issues

Rejected because: Triage prioritizes critical accident or city official compliance tickets over routine charge questions.

“a rider had been in an accident, or a city official messaged about a compliance issue, their tickets were handled in the order they were received.”article, stated

Another automation finalist

Rejected because: Forethought offered the best ROI balance of cost and performance among four final contenders.

“During the evaluation, there were 4 final contenders. For us, the ROI benefit was the most important factor, and Forethought had the best balance of cost and performance.”article, stated

Patterns

confidence-thresholding (support-routing)

Predicted language and category tags determine whether tickets route to specific queues and priority levels.

“when Forethought predicts a specific language and category, the case is routed to a specific queue based on those tags.”article, stated

rag-retrieval (self-service-answering)

Solve searches knowledge articles and resolved tickets to answer chat or email requests without an agent.

“instantly search their entire database of knowledge articles and previously resolved tickets to provide the most accurate response–all without the help of an agent.”article, stated

tool-use-orchestration (support-workflow-automation)

Workflow Builder integrates preset RPA rules with internal tools to automate damaged-vehicle, payment, and receipt tasks end to end.

“These RPA-guided workflows are automated from beginning to end through pre-set rules.”article, stated

human-in-the-loop-gate (complex-case-handoff)

Complex customer questions are routed to the help center web form and then to human support agents.

“If the customers’ question is complex, they’re automatically routed to the help center web form, which is then passed on to a support agent.”article, stated

safety-policy-enforcement (critical-ticket-priority)

Language/category/severity classifications prioritize accidents and compliance issues ahead of less urgent billing questions.

“a customer’s inquiry about what to do in an accident can now be prioritized over a customer who has a question about a charge from two weeks ago.”article, stated

Tradeoffs

How to select an automation partner.

Chose cost/performance ROI balance with Forethought over another finalist with less favorable ROI balance.

Rationale: Lime prioritized ROI while also valuing a partnership model for scaling automation.

“For us, the ROI benefit was the most important factor, and Forethought had the best balance of cost and performance.”article, stated

Which cases should be automated.

Chose simple, routable, and workflow-based cases over automation of every support inquiry.

Rationale: Automation saves agent time at scale, while complex cases still route to human support.

“deflecting simple questions helps save agents’ time in a massive way.”article, implied

Gotchas (what bit them)

Manual all-purpose support queues did not scale globally.

Root cause: Every agent handled every ticket type, and multilingual tickets required agents to toggle between translation tools and cases.

Resolution: Triage routes tickets by language and category to queues with appropriate speakers and trained agents.

“Every support agent handled every type of ticket because there was nothing in place to prioritize and route specific tickets to specific agents.”article, stated

Critical safety/compliance tickets were not prioritized.

Root cause: Tickets were handled in arrival order regardless of accident or city-official urgency.

Resolution: Lime layered language/category classifications with severity service levels and custom queue triggers.

“even if a rider had been in an accident, or a city official messaged about a compliance issue, their tickets were handled in the order they were received.”article, stated

Lime lacked self-service channels before automation.

Root cause: Support inquiries came through email, phone, app, and web form and required follow-up by support agents.

Resolution: Solve and Workflow Builder introduced self-service answers and automated workflows while preserving handoff.

“Lastly, Lime didn’t have a chat widget or any self-service support channels.”article, stated

Evaluation

Lime tracks automation share, annual ticket volume, automatic tagging coverage, predicted tag volume, time-to-first-response reduction, cost savings, and customer satisfaction effects.

Metrics: case automation rate · annual tickets · automatic tagging rate · predicted tags · time to first response · cost savings · customer satisfaction

“Lime automates roughly 27% of cases that come through email and web channels with Solve.”article, stated
“98% of support tickets are tagged automatically, resulting in a 77% reduction in time to first response.”article, stated

Outcomes

Lime operates in 200-plus cities and supports riders globally across multiple languages and support channels.

“With more than 1.7 million tickets per year across all channels, deflecting simple questions helps save agents’ time”article, stated
“With Triage, Lime automates the case tagging process to predict over 2.5 million tags for language and category.”article, stated