How IKEA Turned 8,500 Call Centre Workers Into Interior Design Advisors Using AI

The starting point

Ingka Group is the largest of the twelve IKEA franchise operators, running the majority of IKEA's 460+ stores worldwide. Their customer service operation handles millions of enquiries annually across dozens of markets. The volume is enormous, and the pattern is familiar to anyone who has run consumer-facing operations at scale: the vast majority of inbound contacts are transactional. Order status. Delivery tracking. Store opening hours. Missing items. Returns.

These are necessary interactions, but they are low-value for the customer and expensive for the business. Every call centre agent handling a delivery status check is an agent not doing something more useful. The cost-to-serve is high, customer satisfaction on routine queries is mediocre (people would rather self-serve), and staff turnover in contact centres is a persistent problem because the work is repetitive.

IKEA had tried AI before. Ask Anna, an early chatbot developed with Artificial Solutions around 2005, operated in 21 languages but was eventually shut down. It could not handle direct questions properly and customer satisfaction was poor. That experience left the organisation wary of chatbot hype but also informed about what a better version would need to do.

By 2021, the technology had matured enough to try again. Ingka Group launched Billie, an AI-powered chatbot named after IKEA's iconic Billy bookcase range, across its customer service channels. The goal was straightforward: let AI handle the routine queries that did not need a human, and free up capacity for something better.

What happened next is what makes this case worth studying.

What they built

Billie: the AI chatbot

Billie uses natural language processing to understand customer questions, provide product information, and resolve common service enquiries. It operates 24/7, handles multiple conversations simultaneously, and works across Ingka Group's markets.

The results were significant. Billie resolved 47% of all customer enquiries without human intervention, handling 3.2 million interactions and generating nearly EUR 13 million in cost savings. For a customer service AI deployment, those are strong numbers.

But the more interesting number is the other one. 53% of enquiries could not be resolved by the chatbot. In most organisations, that would be framed as a failure rate to be reduced. Ingka Group treated it as data.

The insight from failure

When the team analysed what Billie could not handle, a pattern emerged. Customers were not just asking about orders and opening hours. They were repeatedly reaching out for home planning and interior design help. They wanted advice on room layouts, furniture combinations, storage solutions, and styling. These requests required design expertise, active listening, and personalisation that a chatbot cannot replicate.

This was latent demand that had been invisible for years. When call centre agents were occupied with routine queries, there was no capacity to notice the pattern, let alone act on it. Billie handled the routine work and exposed what customers actually wanted from a human interaction.

As Parag Parekh, Chief Digital Officer at Ingka Group, put it: "This level of personalisation is not only going to continue to improve but will enhance customer satisfaction and increase loyalty overall."

The reskilling programme

Rather than cutting headcount, Ingka Group launched a structured reskilling programme. 8,500 call centre workers were retrained as remote interior design advisors. The new competencies included remote design consultation, digital retail sales, relationship building, and complex problem-solving for unique customer needs.

The logic was sound. These workers already had customer empathy and deep product knowledge from years of handling IKEA queries. They understood the catalogue, the common pain points, and how customers talk about their homes. Retraining them as design advisors built on existing strengths rather than starting from scratch.

Jesper Brodin, CEO of Ingka Group, was direct about the broader opportunity: "Six out of ten people will need to be retrained for this new economy, but it is a great opportunity. Most of the jobs they're doing right now are soul crushing."

In the UK, the service launched in April 2023 with tiered pricing: GBP 25 for a 45-60 minute video consultation with product recommendations, and GBP 125 for three workspace consultations including a floorplan and 3D visuals.

The results

A note on the EUR 1.3 billion figure: this represents the entire remote selling channel, not solely revenue attributable to the reskilled workers. Some of this revenue existed before the AI and reskilling programme. The number represents the scale of the opportunity that the programme feeds into, not a direct before-and-after comparison.

What makes this case interesting

The chatbot failures were the insight, not the problem. Most organisations deploying AI in customer service focus on increasing the resolution rate. Get that 47% up to 60%, then 70%. Ingka Group did something more valuable: they studied what the AI could not do and found a new revenue channel hiding in the unresolved queries. The 53% failure rate was not a weakness to fix. It was a demand signal to act on.

A cost centre became a revenue channel. Customer service is traditionally a cost-to-serve line item. Every interaction costs money and the goal is to minimise cost per contact. IKEA flipped this. By automating the transactional work and redeploying humans into design advisory, the customer service function moved from pure cost centre to revenue generator. The remote design channel now contributes 3.3% of total revenue with a target of 10%.

No redundancies reframed the entire narrative. The typical AI-in-customer-service story ends with layoffs. IKEA's version ends with 8,500 people doing more interesting, higher-value work. This is not just good PR. It changes how the rest of the organisation responds to AI. When frontline staff see AI as a route to a better role rather than a threat to their current one, adoption accelerates and resistance drops. That is a change management advantage that compounds across every subsequent AI initiative.

The predecessor failure made the second attempt better. Ask Anna's failure a decade earlier gave IKEA institutional knowledge about what goes wrong with customer-facing AI. They knew a chatbot needed to handle real language patterns, not just FAQ matching. They knew the fallback to human agents needed to be seamless. The second attempt succeeded partly because the first one failed visibly enough to teach the organisation what mattered.

The challenges

Attribution is not clean. The EUR 1.3 billion headline is for the full remote selling channel, not a direct measurement of revenue generated by reskilled workers. Multiple commentators have noted that the causal link between "deployed chatbot" and "generated EUR 1.3 billion" is overstated in some coverage. The real picture is more nuanced: AI enabled a redeployment that fed into a growing channel. The programme accelerated growth, but did not create the entire channel from nothing.

The chatbot still fails on the majority of queries. 47% resolution means 53% of customers either need to be transferred to a human or abandon the interaction. For a mature AI deployment, that is a significant limitation. The narrative reframing (failures as demand signals) is clever and genuine, but it does not change the fact that most customers hitting the chatbot do not get their issue resolved by it.

No published satisfaction or retention data. Ingka Group has not released customer satisfaction scores for the Billie chatbot or the design advisory service, nor employee retention data for the reskilled workers. The story is compelling at the strategic level but thin on outcome metrics beyond revenue and cost savings.

Not all transitions are equal. The 8,500-person reskilling number is impressive, but public reporting does not address how many workers successfully transitioned, how many left during the process, or whether the design advisor role suits everyone who previously handled transactional calls. Large-scale reskilling programmes always have a distribution of outcomes.

Lessons for your programme

Map what AI cannot do before you scale what it can. IKEA's biggest insight came from the chatbot's failures, not its successes. Before you optimise your AI's resolution rate, study what it is failing at. Those failures may contain demand signals, process gaps, or opportunities that are only visible when the routine work is stripped away. The Task Decomposition Worksheet (04a) in Section 4: Opportunity Identification is designed to break down operations into tasks that AI can handle and tasks where human judgment creates disproportionate value.

Design the workforce transition before you deploy the technology. Ingka Group did not deploy Billie and then scramble to figure out what to do with displaced workers. The reskilling programme was planned alongside the technology rollout. If your AI programme will change what people do (and most will), the people plan needs to be as developed as the technology plan before you go live. The Workforce Impact Model (15c) in Section 15: Designing for Transformation helps you map which roles are affected, how they change, and what the transition path looks like.

Treat automation as role elevation, not role elimination. The framing matters. "We are automating customer service" and "We are freeing our people to become design consultants" describe the same programme but produce completely different organisational responses. The first triggers resistance. The second builds momentum. This is not spin. It requires genuine investment in new roles that are better than the old ones. Section 11: Change Management, Scaling, and Adoption covers how to design communications and change strategies that build trust rather than erode it.

Let the first failure inform the second attempt. IKEA's Ask Anna chatbot failed publicly. A decade later, that failure was an asset. The organisation knew what bad looked like and could design Billie to avoid the same mistakes. If your first AI experiment does not work, document what you learned and feed it into the next one. The Experiment Outcome Canvas (07g) in Section 7: The Experimentation Framework captures both what worked and what did not, so that learning compounds rather than evaporates.

Sources

• AI and Remote Selling Bring IKEA Design Expertise to the Many (Ingka Group Newsroom, 2023)
• IKEA Uses Artificial Intelligence to Transform Call Center Employees Into Interior Design Advisors (PYMNTS, 2023)
• IKEA NLP and AI-Powered Billie Chatbot Brings Increasing Benefits (Retail Technology Innovation Hub, 2023)
• IKEA's Jesper Brodin Thinks AI Is Inspiring and Deeply Dangerous (Inc., 2023)
• IKEA Is Retraining Call-Handlers as Interior Design Advisors (World Economic Forum, 2023)
• IKEA CEO Jesper Brodin on AI (Yahoo Finance, 2023)
• How IKEA Turned AI Failures Into Revenue (Fluent Support, 2024)
• Can Customer Service Bots Get Companies In Trouble? (Employment Law Review, 2023)