We replaced manual lead handling and email-based customer flows with a Next.js webshop on a WooCommerce catalogue and an n8n + OpenAI lead classifier. Sales recovered 30–40 minutes of daily admin per person.
Background
The client is a 25–30 person Central European automotive SME, with a B2B parts catalogue and an advisory service line. Strong in the market, but the digital customer flow lagged behind competitors: the catalogue lived in PDFs, leads came in by email, and the sales team classified them by hand.
During the first meeting the managing director mentioned three problems: 1) 40–60% of new leads weren't followed up within 24 hours, 2) the sales assistant spent 8–10 hours per week on email triage, 3) maintaining the existing catalogue ate developer time.
Challenge
The core problem wasn't crystal clear on the client side. They knew lead handling was slow, but it wasn't measured. In the early weeks we invested time in process discovery: reconstructing a typical lead journey from sales conversations, from CRM data that hadn't been mined, and from quote-flow analysis.
The lead types turned out to be well-segmented (technical question, commercial enquiry, complaint, partnership offer), and the sales team expected a different SLA per type. The existing flow didn't support that. The catalogue site, a WordPress-based information page, didn't carry real-time stock data, generating more email traffic.
Approach
After discovery we ran two streams in parallel: 1) a Next.js webshop with the WooCommerce API so the catalogue would render with current stock and leads would come in via a structured form, 2) an n8n workflow with OpenAI integration that classifies inbound forms (and the existing email archive) into four categories.
We kept the scope intentionally narrow: no full CRM build, no migration of everything, no upending of the sales workflow. We extended the existing Outlook + Excel routine with two new layers that freed up time.
Solution
The solution has three components. (1) Next.js webshop frontend (App Router, TypeScript, Tailwind) with prerendered catalogue pages and live stock from the WooCommerce REST API. Static regeneration every hour gets Lighthouse to 96/100 on mobile. (2) A structured lead form, accessible from every product page, that lands on an n8n webhook on submit. (3) The n8n workflow calls OpenAI GPT-4o to classify the lead, drafts a follow-up email for the sales assistant, and creates an entry in the right HubSpot pipeline.
The OpenAI prompt is version-controlled in the git repo, with an eval set of 60 real, manually labelled leads. The regression test runs on every prompt change, and if accuracy drops below 90% it blocks the deploy.
Tech decisions
We chose Next.js over WordPress because the catalogue content is structured data (product, price, stock) — ideal for static generation + ISR, not dynamic CMS publishing. We kept WooCommerce because the back-office team has used it for years and didn't want to swap — we just consume the API.
We picked n8n over Make or Zapier because the workflow is self-hosted (running on the client's DigitalOcean VPS), and we can export the full code into git. OpenAI GPT-4o ended up in production because without few-shot examples it already gave good precision on the four lead categories. Anthropic Claude was an alternative, but OpenAI's price–accuracy ratio was better on this use case.
Bumps
Two main difficulties. First: the sales team didn't trust AI classification at the start. In the first weeks the workflow only proposed a category, while the sales assistant verified manually. After two weeks and 80 leads the precision sat steady at 94% and the team began handling it automatically.
Second: we hit the WooCommerce API rate limit at peak hours. We added an edge cache layer with Vercel KV and refresh stock only every 10 minutes. The client agreed, because real-time stock isn't critical for the B2B buyer — the sales follow-up is the place to fine-tune.
Results
Three months after go-live we measured. Median lead response time dropped from 18 hours to 4 hours (the under-90-minute bucket grew from 12% of all leads to 41%). Classification accuracy stabilized at 94% (on the eval set). The sales assistant freed up 35–40 minutes of daily email triage.
Webshop traffic launched with 22% mobile share; after 90 days it was 38%. Lighthouse score 96/100 on mobile, 99/100 on desktop. Organic search position for core product categories moved 8–12 places up (Search Console data).
Stack
The best decision was not trying to revolutionize the whole process. We could measure the small steps, and the team adopted the AI piece once they saw with their own eyes that it worked.
Managing director — Automotive SME
The quote represents typical client feedback; it is not a verbatim citation.
Related
A RAG chatbot with pgvector and OpenAI GPT-4o, integrated into an existing Shopify storefront. 38% of support tickets close automatically; customer satisfaction didn't drop.
Read detailsWe rebuilt the landing site for a 12-person Hungarian B2B SaaS startup on Next.js with static generation, hreflang HU/EN and PostHog A/B testing. Free-trial conversion went from 2.1% to 3.4% in 6 weeks.
Read detailsCase studies are based on real projects, anonymized at our clients' request. Numbers are approximations; specific business data remains confidential.