Responses are assembled from retrieved content with citations, so the bot cannot state a policy it has never been shown
When retrieval returns nothing relevant it says so and offers a person
We score that behaviour on a test set, so it is a number you can check rather than a promise
Most bots treat handoff as the failure case and build it last, which is why customers end up repeating themselves
We define the triggers, pass the full conversation and retrieved context to the agent, and design what the customer sees
A bot that answers 60 percent and escalates well beats one that answers 85 percent and strands the rest
Within a month you have a ranked list of questions your customers ask that your documentation does not answer
That list usually changes the content roadmap more than the bot changes ticket volume
It comes out of the analytics rather than out of a workshop
Froggik
"DESH Team maintained effective communication throughout the project."
Thanks to DESH Team's work, the client saw increased product recognition within the cryptocurrency community. The team managed the...
Viktoriia Bernatska
Co-Founder
ChainCrafters
"I liked their corporate policy and how they turned to customers and their wishes."
DESH Team delivered the project on time, effectively improving the site's UX and flow. The team took the time to understand the cl...
Kolya Vovkun
CEO, Founder
Dropshipping
"I really like how they treat their clients."
DESH Team successfully completed all deliverables; the branding was a great fit for the client's company, and the website was done...
Tetyana Yarchak
CEO
It answers only from retrieved content and cites what it used. When retrieval returns nothing relevant it says so and offers escalation. We score that on an evaluation set built from your real questions, including ones we know it should refuse.
Whatever your team already answers from: help centre, product docs, policies, past tickets, PDFs. Part of the work is turning that into a retrievable knowledge base. Expect a list of contradictions we found in it, because there always are some.
Yes — check an order, book a slot, start a return, update a record. Once it takes actions in your systems it becomes an agent, with the permissions and audit that implies. We usually launch answering first and add actions once answering quality is settled.
Escalation triggers on frustration signals and on any explicit request for a person, and it happens quickly. Arguing with an upset customer is the one thing an automated system should never be allowed to do.
A grounded support bot on one channel typically ships in four to six weeks, including the evaluation set. Additional channels, languages and actions extend it in predictable increments after that.
