Field research
8 implementations, 5 patterns: what AI chatbots really do
Eight Chatwize implementations, eight sectors — paint shops, mortgage advisors, media production, glazing, hospitality supply, cleaning solutions, drilling materials, custom packaging. Outwardly different; under the hood the same five patterns appear every time. Below: the breakdown, with the numbers.
8
customers in this meta-analysis
≥35%
of questions outside office hours — everywhere
1 day
average time from setup to live
Pattern 1 — 35-65% of conversations happen outside office hours
Innovi sits at 36%. Bestel-verf at 40%. Onlineborenkopen 35%. Horecasupply 38%. Maatwerkverpakkingen 64%. Noova 65%. In every case at least a third, often more. That's not consumer-behaviour noise — that's structurally underserved demand. People buy, compare and reconsider in the evening. If you only respond between 9 and 5, you miss the peak.
What this means
Don't sell a chatbot as “% of work automated” — sell it as “customers who otherwise wouldn't have been served.” That's the real ROI.
Pattern 2 — 1 day to live, 2-4 weeks to a stable number
Hypadvies and Onlineborenkopen launched in a day. Maatwerkverpakkingen in a week. Belned hit 100% resolution within month one. Technical go-live is consistently fast. But the automated-resolution percentage stabilises only after 2-4 weeks of iteration — and that's where most teams trip: launching and never looking back.
What we schedule on every implementation
- Conversation log review at 7 days (30 min)
- Conversation log review at 14 days (30 min)
- First report at 4 weeks — only commit to a KPI from there on
Pattern 3 — product links beat product descriptions
At Bestel-verf a customer asks: “Which paint suits outdoor wood?” The bot doesn't dump a paragraph — it links directly to two product pages. Same at Onlineborenkopen: “Which drill for concrete?” → two links. At Horecasupply it goes via an article-number API: type the code, get the page.
Long product descriptions perform poorly inside a chat window. Customers don't read them. A short explanation plus one click to the right page converts better, faster and more cleanly.
Pattern 4 — the bot scales with your catalogue
Horecasupply has 35,000+ products. At that volume you can't pour individual product pages into a chatbot — that's why the custom article-number API was built. It's not an exotic integration; it's what any webshop beyond a few hundred SKUs ends up needing. For smaller catalogues (Bestel-verf, Belned) manually linked product sources do the job.
Rule of thumb
Under ~500 SKUs: manual sources. Above that: webhook or API. Building a complex integration too early costs more time than it saves.
Pattern 5 — a name and a persona drive engagement
The Belned chatbot is called Lynn. Customers behave measurably differently towards a bot with a name than towards “the chatbot.” Not because they think they're talking to a human — they know it's AI — but because the name sets the tone. It takes five seconds and noticeably affects how long conversations run and how much feedback customers leave.
What to take into your own rollout
- Plan for 24/7 traffic from day one — not office hours.
- Budget 2-4 weeks of iteration. Block the slots in your calendar now.
- Build around product links, not product descriptions.
- Decide upfront whether your catalogue is manual or API — don't switch mid-way.
- Give the bot a name. It's not a branding thing, it's a tone thing.
Ready to make this happen for your team?
Book a short demo and we'll show how Chatwize fits your customer questions, channels and processes.