Resolve, not just answer
A chatbot that only answers FAQs moves the work rather than removing it. Our agents act: they look up orders, check entitlements, process returns and refunds within policy, reschedule appointments, update account details and create or update tickets in your helpdesk, all through your systems' APIs.
Escalation design
Human escalation is designed up front, not bolted on. The agent hands off when:
- The customer asks for a person.
- The request is outside policy or above a value limit, for example a refund over a set amount.
- The agent's confidence is low or the case type is new.
- Signals suggest vulnerability, a complaint, legal risk or strong frustration.
- The same issue has already been contacted about several times.
The handover goes to the right queue with a summary, the steps already taken, relevant records and a suggested next action. Outside staffed hours, the agent tells the customer what will happen next and when, and creates a prioritised ticket.
Channels and systems
- Channels: web and in-app chat, email, WhatsApp and other messaging apps, and voice.
- Helpdesks: Zendesk, Salesforce Service Cloud, ServiceNow, Freshdesk and Intercom.
- Back-office: order management, billing, CRM and ERP systems through their APIs.
- Knowledge: help centre articles, policies and internal guides, retrieved with citations so answers can be checked.
Quality and safety
- Answers are grounded in your approved knowledge and policies, with sources recorded for each response.
- Actions are limited by policy and value thresholds, and validated before they run.
- Every conversation, action and handover is logged for QA review.
- An evaluation set of real past conversations is re-run before every change.
- Your team reviews samples and overrides, which feed back into the agent's instructions.
How an engagement runs
We analyse a sample of recent contacts to find the request types that are high volume and resolvable within policy, then build and test against those. You get a working prototype on your data within 3 to 5 weeks and production within about 60 days, starting with a limited set of request types and channels. You own the code and models; work is scoped as fixed-price milestones.
Frequently asked questions
What share of contacts can an AI agent resolve?
It depends on your mix of requests and how much the agent is allowed to do in your systems. We measure it on your own contact data during discovery rather than quote a generic figure, and track resolution rate, escalation rate and customer satisfaction after launch.
How does the handover to a human work?
The agent transfers the conversation to the right queue in your helpdesk with a summary, steps taken, relevant records and a suggested next action, so the customer doesn't have to repeat themselves. Outside staffed hours it sets expectations and creates a prioritised ticket.
Can it issue refunds or change orders?
Yes, within limits you set. Actions below a threshold and inside policy can be completed automatically; anything above it or outside policy is escalated for approval.
Does it work with our existing helpdesk?
Yes. We integrate with Zendesk, Salesforce Service Cloud, ServiceNow, Freshdesk, Intercom and others through their APIs, so tickets, macros and reporting stay in the tools your team already uses.
How do you stop it making things up?
Answers are grounded in your approved knowledge sources and system data, with the source recorded. If the agent cannot find a supported answer, it says so and escalates rather than guessing.
Can it handle multiple languages?
Yes. The agent can converse in the languages your customers use, while escalations and internal summaries are written in your team's working language.