We’re putting $5.5M behind research and development of smaller, specialized AI models for real-world deployment — Read our manifesto →
    All posts

    [Case Study] 3.2 Million Conversations and EUR 13 Million Saved: IKEA's Billie Assistant

    How IKEA automated routine customer questions, reskilled 8,500 contact-center employees, and redirected human expertise toward remote sales and design.

    Conscious Engines

    Industry: Retail
    Organization: Ingka Group, the largest IKEA retailer
    Use case: Customer-service automation and workforce transition to remote selling
    Evidence basis: Ingka Group first-party reporting
    Disclosure: This is an independent analysis by Conscious Engines based on public information. Results are company reported.

    1. Outcome at a Glance

    Between 2021 and 2023, IKEA's Billie assistant handled 3.2 million customer interactions and automated 47% of enquiries. Ingka Group reported nearly EUR 13 million in savings. At the same time, the company reskilled 8,500 contact-center employees for remote interior-design and selling work.

    Key Outcomes

    3.2 million

    Billie interactions

    2021 to 2023 reported in the cited case.

    47%

    Enquiries automated

    Company-reported share.

    Nearly EUR 13 million

    Savings

    Company-reported estimate.

    8,500

    Employees reskilled

    Workforce transition reported in the cited case.

    EUR 1.3 billion

    Remote sales in FY22

    Adjacent channel result.

    MeasureReported resultEvidence note
    Billie interactions3.2 million2021 to 2023
    Enquiries automated47%Company-reported share
    SavingsNearly EUR 13 millionCompany-reported estimate
    Employees reskilled8,500Workforce transition
    Remote sales in FY22EUR 1.3 billionAdjacent channel result
    Remote share of total sales3.3%FY22
    Service availability24/7Customer-access design

    The case is stronger than a simple chatbot story because automation was paired with workforce redesign. Routine questions moved to Billie, while people shifted toward advice and selling where human judgment and taste create more value.

    The EUR 1.3 billion remote-sales figure should not be attributed solely to Billie. It describes the broader remote-sales channel. The public source does not publish a causal revenue contribution from the assistant.

    2. The Operational Problem

    Retail contact centers receive repetitive questions about orders, delivery, stock, returns, and policies. These contacts are high volume and often resolvable through existing systems. At the same time, customers need help with complex design, product combination, and purchase decisions.

    Using trained people for every status question creates queue and labor cost. Automating too aggressively creates customer frustration when the assistant cannot access an order, understand an exception, or transfer context.

    The operating problem is to separate requests by complexity and value. The assistant should complete bounded service tasks. Human coworkers should handle emotional, ambiguous, commercial, or design-intensive work.

    3. What Was Built

    Billie is a customer conversational assistant integrated into IKEA service journeys, operating continuously across supported markets.

    System at a Glance

    Intent and language

    Understand the customer's request and language.

    Customer context

    Identify order, delivery, product, and account state.

    Enterprise retrieval

    Retrieve current policy and product information.

    Action tools

    Complete approved service actions or status queries.

    Dialogue

    Ask for missing information and explain next steps.

    Human handoff

    Transfer complex cases with conversation context.

    LayerFunction
    Intent and languageUnderstand the customer's request and language
    Customer contextIdentify order, delivery, product, and account state
    Enterprise retrievalRetrieve current policy and product information
    Action toolsComplete approved service actions or status queries
    DialogueAsk for missing information and explain next steps
    Human handoffTransfer complex cases with conversation context
    AnalyticsIdentify containment, repeat contacts, and unmet demand

    At retail scale, task-specific routing is essential. Small models can detect language, intent, and entities. Deterministic APIs answer order status. Retrieval grounds policy. A stronger model handles complex explanation. This reduces average inference cost and keeps actions observable.

    The same conversation data can reveal product, delivery, and process defects, provided privacy and retention are governed.

    4. How It Reached Production

    IKEA's deployment suggests a deliberate division of labor between AI and people.

    Automate complete journeys. A useful assistant should resolve the request, not only direct the customer to another page.

    Create a workforce path. Reskilling 8,500 employees gave the program a value story beyond labor removal. Human expertise moved toward revenue and complex service.

    Measure durable resolution. The 47% automation figure should exclude abandoned conversations and rapid repeat contacts.

    Maintain market and language quality. Product names, policies, availability, and consumer rules differ across markets.

    Connect service and sales carefully. Recommendations should be helpful and transparent, with customer consent and fair treatment.

    The business case should separately report avoided service cost, workforce-transition cost, revenue supported by remote selling, satisfaction, and repeat contacts.

    5. What Retail, Hospitality, and Travel Leaders Should Take Away

    IKEA shows that customer AI can support a strategic workforce shift. The assistant handled millions of routine conversations while employees moved toward work where human expertise could influence the purchase.

    A production retail model stack combines multilingual speech-to-text and text-to-speech, voice and chat agents, product and policy RAG, order-system tools, recommendation models, and intelligent human handoff. Models should be routed by intent, risk, and cost.

    The primary metric is cost per durably resolved customer need, with satisfaction and conversion reported separately. The nearly EUR 13 million saving is compelling, but the more durable lesson is to redesign the human and AI operating model together.

    Sources