Product 01
Customer support agent
Handles customer inquiries in your tone, with your knowledge, inside your stack. Escalates to humans only when it matters.
01 / Product
What it does
A support agent receives inquiries via email, web chat or Slack, classifies them, consults your knowledge base (FAQs, documentation, history), answers what it can resolve directly and escalates to a human the cases that need judgment, emotional tone or commercial decision.
02 / Stack
What the implementation includes
Discovery session to define inquiry categories and escalation criteria
Integration with your channel (email, web chat, Slack, Intercom, Zendesk)
Connection to your knowledge base (Notion, docs, FAQs, history)
Evals on real emails and conversations before going to production
Kill-switch and human fallback documented in the SOW
Contractual inference cap
Full handover at contract end
03 / Fit
When it fits
- ✓You receive ≥30 inquiries a day with repetitive patterns (FAQs, order status, product information)
- ✓You have searchable documentation or FAQs
- ✓You want multilingual support (CA/ES/EN or other) without hiring per language
- ✓You accept that the agent starts with assisted replies and grows to autonomous ones
- ✕Your inquiries are all unique and require empathy or negotiation
- ✕You have no searchable documentation and don't want to create any
- ✕You want to replace the entire support team for cost reasons
- ✕You expect the agent to sell or negotiate on your behalf
04 / Example
A B2B company receives 40 emails a day about order status, product doubts and invoices. An agent replies to 70% directly with accurate information, escalates 20% to humans with prepared context, and flags 10% as cases needing commercial review. The team stops drowning in the inbox and only works on cases that need human judgment.
05 / Signals
Signals you need this product
- 01 Your support team answers many repetitive questions every day
- 02 Customers expect more than your hours allow (nights, weekends)
- 03 Documentation is scattered and answers aren't always consistent
- 04 You're opening multilingual markets and don't have enough people
06 / FAQ
Frequently asked questions
The SOW defines which inquiry categories the agent answers directly, which ones it drafts for a human to validate, and which ones it escalates without replying. As precision is measured, these categories can expand.
Yes. The models we use (Claude, GPT) are multilingual by default. Personality and tone are configured per language in the persona file, editable by the client team without touching code.
Every reply is logged with the information consulted. Incorrect cases are flagged in the weekly evals, which are reviewed to adjust the prompt or expand the knowledge base.
Email (IMAP/Gmail), web chat (own widget or integration), Slack, Intercom, Zendesk, Freshdesk. WhatsApp via Meta Cloud API or Twilio. Other channels reviewed in the SOW.
Usually 4-8 weeks depending on the complexity of the knowledge base and the channels integrated.
The client always owns the data and the conversations. serpixel keeps prompts and configuration during the contract and transfers them fully at the end.
30 minutes, no commitment.
If your team repeats tasks every day, or your website does not bring you clients, a conversation probably pays for itself. No pressure.