Dazza.
Dazza. Consider it
handled.
Building an AI customer service business. From a familiar name to the product behind every answer.
Inside the project- Relationship
- Built & owned by Wall & Fifth
- Our role
- Strategy, identity, design & development
- Stage
- Early access
01The idea
Good service takes
more than an answer.
A customer asks a simple question. Behind it sits a website, a policy, a booking, an order—or a person who needs to step in.
We built Dazza to bring those pieces together. An AI agent that understands the business, carries out permitted work and hands over with the context intact.
As our own venture, we shaped the whole experience: the name and identity, the customer conversation, the team’s workspace and the systems that make it a business.
of UK businesses with fewer than ten employees reported using AI in September 2026. [1]
more issues resolved per hour, on average, in a study of human support agents using AI assistance. [2]
The opportunity: make capable AI easier to put to work.External research, with scope and sources below. These are not Dazza performance figures.
02The product
One conversation.
Everyone on the same page.
The customer gets a clear next step. The team gets the knowledge, controls and context behind it.
A clear handover.
When a teammate takes over, Dazza cancels its unsent reply. The conversation keeps one owner and one direction.
Answers with a source.
Business knowledge supports the reply. Source references and corrections give the team a way to check and improve it.
Getting started
From your website
to your Dazza.
Setup begins with what the business already has. A guided sequence turns that context into an agent the team can shape, test and publish.
- 01Read the business
- 02Shape the voice
- 03Test the answers
- 04Publish deliberately
03The identity
A name you can
ask for help.
Short, familiar and capable. We gave a technical product a personality people could feel comfortable talking to.
A lot more sorted.
A recognisable signature.
Inter Tight / The voice
Helpful
by nature.
Personality, with a purpose.
Warm yellow brings the energy. Confident typography gives it clarity. The asterisk becomes a familiar presence across the website, the product and the conversation.
04The motion
Small moments.
A more human rhythm.
The wait, the reply, the handover. We designed each transition to make the conversation easier to follow.
Making the wait feel considered
A quiet signal that Dazza is working on an answer.
An answer as it happens
Words appear as they arrive, so the customer can start reading.
A person takes it from here
The name and avatar change together when a teammate joins.
Explore the complete motion system 8 interactions, considered together
Opening the conversation
A small invitation, with a clear place to begin.
Keeping a natural rhythm
New messages arrive without interrupting the conversation.
Helping someone get started
Useful prompts make the next step easier to find.
A clear response to an action
The send interaction acknowledges the customer’s input.
Giving useful details a shape
Structured answers make prices and key facts easier to scan.
05The engineering
Simple on the surface.
Considered underneath.
The friendly interface sits on a system of knowledge, published versions, permissions and human ownership. Explore how the pieces connect.
Start with what the business knows.
Website pages, help content and policies give Dazza its context. The runtime selects relevant knowledge and can cite the source. Corrections written by the business take precedence over imported material.
Website · Help content · Policies · CorrectionsAstro
Fast, readable product pages and documentation.
Next.js + React
One workspace for setup, conversations and control.
Postgres + Neon
Connected records, managed through Drizzle ORM.
Claude + adapters
Grounded replies, tool use and provider choice.
Stripe · Resend · Twilio
Billing, email delivery and voice conversations.
Vercel
Separate releases for the website and application.
The decisions behind the stack
Separate jobs, separate releases.
Astro serves the marketing site; Next.js 15 and React 19 support the authenticated application. Separate deployment roots let content and product evolve independently. TypeScript connects the implementation, with Tailwind CSS used in the application.
Context before generation.
Small knowledge bases can be supplied in full. Larger ones use ranked passages and query expansion. Business-authored corrections take priority. Provider adapters support a business’s own Anthropic, OpenAI or compatible provider, subject to its plan.
Permission belongs in the application.
Workflow conditions, identity checks and approvals are evaluated in code. The model can interpret a request; the application controls execution. These controls constrain actions while testing, corrections and human review address answer quality.
A deliberate release process.
Setup generates eight test questions. Saved questions and multi-turn simulations help teams inspect behaviour without executing live actions. Publishing creates a numbered version, with a route back to an earlier configuration through a reviewed draft.
06The complete build
A product.
And the business
around it.
We built the parts that make Dazza usable beyond a demo: onboarding, subscriptions, documentation, mobile SDKs and the tools to keep the brand moving.
Connected web experiences
The marketing site and the service application.
Mobile SDKs
iOS, Android, React Native and Flutter.
Editable brand templates
Six collections, from social posts to presentation slides.
Built, launched.
And still learning.
Dazza is in early access and runs on Wall & Fifth’s own website. The figures here describe what we delivered. Activation, answer quality and successful handovers are the measures that matter next.
Research & project notes.
The evidence behind the story.
Office for National StatisticsBusiness insights and impact on the UK economy
September 2026 adoption of at least one AI technology: 29% of UK businesses with fewer than 10 employees; 51% of businesses with 250 or more. BICS Wave 165, a weighted voluntary survey; official statistics in development.
8 October 2026 · Read the source (opens in a new tab)The Quarterly Journal of EconomicsGenerative AI at Work
Published study of 5,172 customer-support agents at one business. AI assistance increased issues resolved per hour by 15% on average, with substantial variation by worker experience. Human agents remained responsible for the conversation. This is not a study of autonomous agents or Dazza.
2025 · Brynjolfsson, Li & Raymond · Read the source (opens in a new tab)OECDGenerative AI and the SME Workforce
Representative late-2024 survey of more than 5,000 SMEs across seven countries. 31% reported using generative AI; among those using it, 65% reported improved employee performance. These are self-reported effects, not a measured 65% productivity increase.
5 November 2025 · Read the source (opens in a new tab)Project evidenceProduct, design system and implementation
Product details reflect the implementation reviewed on 3 November 2026. The case study draws on the marketing website, application, source documentation and supplied product imagery and motion studies. Delivery counts are not adoption or performance statistics. The motion clips contain example conversations.
Beyond the interfaces shown here, the build includes conversation-based subscriptions, unlimited team seats, Stripe usage metering, plan entitlements, a help centre, release notes and a browser-based brand studio with editable files, high-resolution raster and outlined SVG exports.
Read Dazza’s release notes (opens in a new tab)Wall & Fifth