autonomous resolution in pilot 2, against a business-case minimum of 60%
of HALO livechat conversations are resolved fully autonomously (trailing 90 days)
of order-cancellation requests are handled autonomously by the Cancel Agent
CameraNu is one of the largest photo and video retailers in the Benelux, with around 40,000 products in its webshop and physical stores across the Netherlands and Belgium. Started on Urk, it has grown from an online-only retailer into an omnichannel business that combines its webshop with in-store expert advice for hobbyist and professional photographers and videographers.
Camera gear is complex: specs, lens and body compatibility, and personal preferences all play a part, and a lens can cost thousands of euros. Customers want the right advice before they buy. For a lean team that handles both sales and service, that is a lot to deliver on every chat, call and email.
CameraNu was looking for a way to take repetitive, first-line questions off its customer service team's plate, so staff could spend more time on the advice-heavy conversations only a camera specialist can really help with.
Working with CM.com, CameraNu built its HALO chatbot, internally nicknamed Flits, on top of the Agent Inbox service desk CameraNu had already used for years. After two pilots and a lot of hands-on script tuning, HALO now resolves around 60% of all livechat conversations fully on its own, with some agents, like order cancellation, resolving nearly 90% of requests without any human involvement.
CameraNu’s customer service team of around 22-24 FTE handled almost every question manually over phone and email. Livechat existed only as a bonus channel, staffed by one or two colleagues when capacity allowed.
"Before we had HALO handling things, you'd fall back on the classic question, where's my package, what's the current status, and that all still had to be looked up manually. It created a lot of pressure on the department, with a lot of repetitive questions, without us really being able to use our team's customer service knowledge where it mattered most."
AI Specialist, CameraNu
CameraNu had already run its service desk on CM.com's Agent Inbox for around eight years, so when it set out to automate first-line service, it still ran a formal comparison rather than assuming HALO was the obvious choice. It built a matrix of five AI chatbot vendors and went deeper with three, including a parallel internal pilot of another AI assistant for email while HALO handled livechat. Two things ultimately tipped the decision toward HALO: staying inside CM.com's existing data and tooling landscape instead of bolting on another vendor, and HALO's guarantee that a conversation could always be handed to a live agent whenever the AI couldn't fully resolve it, a non-negotiable safety net for the team.
CameraNu's HALO setup is built and maintained largely by its own in-house AI Specialist, organized into agents across four pillars: product & orientation, order handling, compatibility & specification questions, and an escalation layer that can hand off to a live colleague at any point.
Agents for product orientation, order status, cancellation, and camera/lens compatibility, several built and fine-tuned directly by CameraNu's own AI Specialist
A choice-guidance module reusing CameraNu's product-filtering tool to ask shoppers the same questions they'd answer on the website and surface the top matches
New agents increasingly built from scratch using Claude via MCP, including a custom dashboard to monitor agent performance and tune scripts
Not every customer knows exactly which camera or lens fits their needs. CameraNu's Product Advice Agent asks the same guided questions used in the website's choice-guidance filter (intended use, budget, and preferences like size or lens type) and returns a short, reasoned shortlist instead of the full catalog. Where a choice still isn't obvious, it explains the trade-offs between options so customers can decide with confidence.
Asks structured questions on usage, budget, and preferences
Matches answers against CameraNu's live product catalog
Recommends a short, reasoned shortlist instead of a generic list
Explains trade-offs between options so customers can decide with confidence
Customers often want to know not just whether a product is available, but where. CameraNu's Stock Agent checks live inventory across the webshop and, on request, a specific physical store, so customers can choose between home delivery and in-store pickup. If an item isn't in stock nearby, the agent can point to the next-best store or split an order, reserving what's available in-store while the rest ships separately.
Checks real-time stock across the webshop and physical stores
Helps customers choose between delivery and in-store pickup
Suggests the nearest alternative store when one location is out of stock
Splits an order automatically when only part of it is in stock
CameraNu told CM.com what it missed in HALO: a faster way to build agents. CM.com delivered a Claude connection through MCP, an open standard for linking AI to business systems, and Ask HALO, where you describe the agent you want in plain language.
Andy now builds and changes agents by working with Claude connected to HALO. He also built his own dashboard with Claude to see how each agent performs and to tune scripts on real conversations.
His first rebuild was the order-status agent: one task, the right context, three tools. Measured the morning after go-live, response time fell from 5.5 to 6 seconds to 3 seconds, and success rose from 45% to 67%. These are first results, Andy adds, and more conversations need analyzing. Output also depends on good instructions and the right knowledge.
Flits takes the first line; colleagues pick up the rest in the same thread in Agent Inbox, with full context. About 35% of chats end in a justified handover to sales or service, routed automatically. Email, phone and WhatsApp still run on people only; phone moves to CM.com voice next. Colleagues mostly see what Flits couldn’t solve, so CameraNu now also shows the team what it handles well on its own.
The results speak for themselves
of HALO livechat conversations resolved fully autonomously (trailing 90 days)
pilot 2 result vs 60% minimum
autonomous success rate for the order-cancellation agent
success rate for the camera/lens compatibility-check agent
HALO on livechat now accounts for roughly a quarter of all incoming customer service conversations, a channel CameraNu barely had before. Autonomous resolution has grown to 60% over the trailing 90 days, building on the 62% CameraNu already hit in pilot 2, well above the 60% minimum set in the business case. Cancellations are handled autonomously nearly 90% of the time, and camera/lens compatibility questions around 70%. Phone volume dropped around 20% and email 5–20% during the pilot, as more questions moved to chat.
Introducing HALO also required a change in how the team thought about the chatbot. When it was first announced to 50-60 employees at once, some found it unsettling. The first question for many was simply, "what does this mean for my job?" CameraNu's own lesson: next time, introduce a change like this more gradually rather than all at once.
A vocal minority stayed skeptical of chatbots in general and was quick to point at any wrong answer as proof it "doesn't work." But colleagues who were closely involved in the pilots, helping build scripts and giving feedback, grew into strong advocates, and now see the whole project very differently. The chatbot, Flits, even has its own place in CameraNu's customer service org chart, introduced to the team as a colleague to work alongside, not around.
"We don't do this to replace people. It's an addition to their work, and it gives the team the chance to actually spend their time on the tasks they enjoy most."
AI Specialist, CameraNu
CameraNu's AI ambitions extend well beyond livechat. Areas currently being explored include:
Native voice automation via HALO, as telephony moves onto CM.com's voice capabilities
Expanding automation to WhatsApp and email, gated on livechat first reaching a 70–75% autonomous success rate, per CameraNu's own internal criteria
A HALO–BigQuery integration to analyze how much the chatbot contributes to conversion, at the customer level
Product-feed and on-page context integration, so HALO knows which product a visitor is viewing instead of relying only on hard-coded values
Soon, HALO won't just help customers track and manage their orders. It will help advise them on the right camera or lens in the first place, moving from pure service into sales support.
"If you give something attention, it works better. We were early adopters of HALO, so in some ways we were figuring it out from both sides at once. It really came down to trial and error, testing and fine-tuning, to find out what actually makes a good agent." Andy Luijendijk, AI Specialist, CameraNu
Select a region to show relevant information. This may change the language.
