Smartphonehoesjes

The challenge

Before Mim, Smartphonehoesjes ran a familiar playbook: a 14-person support team, a scripted chatbot built on a massive decision tree, and recurring capacity crises every time the holidays or summer vacations came around.

The scripted chatbot managed to route about 45% of customer traffic, but only 25% of those conversations were resolved without a human handover. Every Black Friday and Christmas, volumes spiked and SLAs fell apart. Sickness days went up after peak periods. Summer vacations squeezed the team from the other direction: fewer agents, no way to deny people their time off.

"It was a lot of text work. We just added buttons and buttons and buttons."
Jessica Schulz, Customer Service Specialist at Smartphonehoesjes

The company tried filling gaps with colleagues from other departments, "in-house heroes" drafted in to handle emails and returns. Those same colleagues kept asking one question: why isn't this automated?

They were right. The same queries came in again and again. The scripted chatbot could only do so much, and the team couldn't keep scaling with people. International expansion made it worse, new markets meant needing language coverage, but early-stage order volumes couldn't justify dedicated hires per country.

"We cannot always keep adding international colleagues, because in the beginning the order volume is not high enough to have enough work for one agent. But if you have just one agent, what do you do during lunch? When they're sick? On vacation?"

‍

Finding the right tool

When Smartphonehoesjes started looking for an AI agent in spring 2025, the team ran into a problem most companies face: every vendor looked the same.

‍

The rollout

Jessica and her colleague Mascha started deliberately small–Dutch market, simplest possible use case, minimal setup.

‍

From content to actions

The real shift came when the team stopped using Mim as a knowledge base and started connecting it to backend systems.

‍

The results

Less than a year after going live, the numbers tell a clear story.

The gap between AI and human CSAT scores is just 3 percentage points. Customers regularly don't realise they're interacting with AI at all.

"We hear a lot from customers–they say, 'Oh, I didn't even notice it was an AI assistant. Afterwards I saw the disclaimer, but I couldn't tell.'"

The team reduction from 14 to 8 full-time agents happened while the business continued to grow: more orders, more markets, more customer questions. The remaining team members handle higher-complexity cases while Mim takes care of the repetitive volume.

‍

Staying honest about the challenges

Jessica is candid about the rough edges. As Mim took on more responsibility, the team noticed it would sometimes overpromise, telling a customer someone would follow up, when no follow-up was triggered. This showed up clearly in CSAT data. The quantity numbers looked great, but quality needed attention–"Customers gave us feedback that the AI bot experience was brilliant, but somebody should come back to them afterwards and it didn't happen."

Jessica brought the issue to her executive team, showed them how Dixa was actively addressing it with a new overpromise detection feature, and confidence held. That confidence had been building steadily, one of the company's owners had pushed for AI adoption early on, and monthly reports tracking containment and CSAT side by side kept the momentum going.

"They always deliver what you promise–that's the biggest compliment, I guess."

‍

Jessica's advice

Start small. Don't over-plan. The executive team will want a detailed roadmap with projected numbers for every use case, resist that.

Real customers will phrase things differently than your test cases, and the results only compound if someone is consistently reviewing conversations, refining instructions, and adding new tools–"You have to be in the game, really strict on what's going good, what's not going so good. Without that, you cannot be surprised if the quality is not so good."

‍

What's next

Smartphonehoesjes is preparing to launch an Italian webshop with 100% AI-driven customer service from day one. No Italian-speaking agents. No local hires. Mim will handle all customer conversations on email, chat, and WhatsApp in Italian, while the existing team uses Agent Copilot for the edge cases that need a human touch.

If Italy works, the playbook repeats across more markets. Between Mim handling conversations in any European language and Agent Copilot letting agents respond in languages they don't speak, new countries become a logistics question, not a hiring one.

The team is also adding Facebook, Instagram, and WhatsApp as channels, with AI on the phone channel planned for later this year. Email containment targets are set to climb from 51% to 70%.

"We want to make it AI-driven, automation-driven. That's how we grow."

‍

From 25% containment to 82% in under a year — while cutting the team in half. And the next market launch won't require a single hire. Book time with our team to see how Mim could work for yours.