Deploy
Getting started
What a pilot looks like week by week, what we need from you, and how success is judged.
Most engagements begin with a scoped pilot — one avatar, one surface, one outcome metric, 30 to 90 days. It exists to answer one question: does this work for your users, on your content?
Week 1 — scope and avatar
- Agree the use case and the single surface it launches on (web, app, or kiosk).
- Pick an avatar from the catalog, or start a custom one.
- Agree the metric the pilot is judged on. Usually first-contact resolution, containment rate, or completion rate — pick it now, not at the end.
- Decide the deployment shape and, if on-prem, get cloud access sorted.
Week 2 — integration
- Load your knowledge base and review the first answers together.
- Wire the language model — yours or ours.
- Connect any system the avatar has to read from or write to.
- Style the surface: branding, languages, subtitles, idle behaviour.
Week 3 onward — live
- Open it to real users, starting narrow.
- Review transcripts weekly for the first month. This is where most of the quality comes from: real questions expose gaps that no amount of internal testing does.
- Measure against the metric agreed in week 1.
Kiosk deployments add hardware procurement and physical installation, which usually makes them a 3–4 week start rather than 2.
What we need from you
| Item | Why | When |
|---|---|---|
| Knowledge sources — documents, FAQs, policies, a site to crawl | The avatar's answers come from here | Week 1 |
| A subject-matter reviewer | Someone who can say "that answer is wrong" with authority | Throughout |
| Brand assets — logo, colours, tone-of-voice notes | Visual and verbal fit | Week 1 |
| Language model access, if you bring your own | Provider, model, and a key | Week 2 |
| Integration details, if the avatar reads or writes systems | Endpoints, credentials, test environment | Week 2 |
| Cloud project access, for on-prem | We provision inside your account | Week 1 |
| Site survey, for kiosks | Power, network, mounting, ambient noise | Week 1 |
The two things that most often slow a pilot down are knowledge sources arriving late and no empowered reviewer being available. Both are worth fixing before kickoff.
What we deliver
- A working deployment on the agreed surface.
- Configured avatar, voice, knowledge base, and conversation flows.
- Access for your team to review conversations.
- An outcome review at the end of the pilot against the agreed metric.
After the pilot
Pilots convert into an annual production deployment — hosted or on-prem — and everything built during the pilot carries over. Commercial terms are quoted per deployment shape rather than per seat; see the pricing page.