Chatbot Design Systems
Deflection counts the people you stopped. Resolution counts the people you helped. We design, build, and tune assistants against the second number — with the conversation design, escalation paths, and evaluation that make it hold up in production.
What you can rely on
They were built as a deflection tool, and customers can tell within two messages.
A bot that traps someone in a loop scores well on deflection and terribly on renewal. Measuring the wrong thing produces exactly the experience you are trying to avoid.
Answering from a help centre is table stakes. Without authenticated access to order, account, and booking systems it cannot resolve the requests that actually generate contact.
Handover arrives with no context, so the customer repeats themselves and the agent starts cold. The bot has added a step rather than removed one.
Without transcript review, evaluation sets, and per-intent quality tracking, quality drifts silently and the first signal is a complaint.
Start here
A production-grade assistant on one channel, live in thirty days.
What is included
Fixed budget and timeline from day one. If the pilot does not hit the agreed resolution target, we tell you plainly rather than reframing the metric.
Book a Discovery CallDeliverables
Design artefacts and running software, not a prompt in a text file.
Persona, dialogue, tone, error recovery, and the writing that makes an assistant usable.
Building and integrating the assistant across web, mobile, messaging, and in-product surfaces.
Tool-using agents that take real actions, with guardrails and human checkpoints.
Telephony and voice interfaces, with latency, barge-in, and speech recognition tuning.
Content structuring, retrieval strategy, and keeping answers correct as sources change.
A structured review of an existing assistant with a prioritised improvement plan.
Moving off a legacy or vendor-locked bot without losing intents or history.
Ongoing transcript review, eval expansion, and iteration against live quality data.
Teaching your writers and engineers to own conversation design after we leave.
Week 1
Transcript and ticket analysis to find what people actually contact you about, and which of it can be resolved end to end.
Weeks 2–3
Persona, flows, failure handling, and escalation. Reviewed with your support team before a line of code is written.
Weeks 3–4
Integration, guardrails, and the evaluation harness. Quality is a measured number before launch, not after.
Ongoing
Staged rollout, live transcript review, and weekly iteration against resolution and satisfaction.
Engagement models
Indicative starting points, not quotes. You get a written price for your specific scope before any work begins.
Lowest-Cost Entry
Structured review of an existing assistant, with a prioritised fix list.
Starting Price
From $4,500
Best For
You already have a bot and it is underperforming.
One channel, one integrated system, live and measured in thirty days.
Starting Price
From $24,000
Best For
Proving the case before committing to a full programme.
Multi-channel assistant across several systems, with full evaluation tooling.
Starting Price
From $75,000
Best For
Replacing or substantially expanding a support and service channel.
Ongoing transcript review, evaluation, and tuning against live quality.
Starting Price
From $5,000 per month
Best For
A live assistant whose quality must keep improving.
Every option is scoped and priced in writing before delivery begins. No surprise invoices or automatic upgrades.
FAQ
The things prospective clients ask us most often about this service.
Ask us something elseClients rarely need only one of these. Combining them under one contract is what removes the handover gaps.
Thirty minutes with the engineer who would run your project. You leave with a scope, a cost range, and a recommended next step — whether or not you work with us.
A 30-minute call with an engineer, not a salesperson. You leave with a scope, a rough range, and a recommended next step — whether or not you work with us.
