Running UX and CxD Workshops
Role: Conversation Designer & UX Writer
Awin: Building Ava from the ground up
Awin · Since January 2025
LLM chatbot design AI strategy UX writing Prompt design
Product:
Ava - Awin's virtual assistant
Scope
0 → global deployment
My role:
Content Designer - Chatbot
The challenge
I joined Awin at the start of its transition toward a more tech and AI-led approach to customer support. The primary task: build a production-ready LLM-powered chatbot from scratch. My role sat at the intersection of engineering and UX - connecting the two teams, advocating for safe and appropriate AI use, and ensuring that the product being built would genuinely serve Awin's users before it reached them.
Roadmap so far
Year 1: Foundations
Building Ava's identity, guardrails, and strategic baseline
Led workshops to define Ava's persona and tone of voice. Stress-tested guardrails, mapped risks early, and refined the core prompt to deliver a safe and effective user experience. Reached out to Marketing and Customer Success to understand team pain points and ensure Ava was built around real user and business needs.
Late 2025: Measured launch
UK and US deployment: Pushing back on premature global rollout
Advised the wider team to restrict the initial launch to the UK and US despite pressure to go worldwide. Test data indicated Ava wasn't yet robust enough to handle multiple languages, and a multilingual AI strategy hadn't been developed. This recommendation protected the product's quality and long-term credibility.
Q1 2026: Global rollout
Successful multi-market testing and worldwide English deployment
Following successful testing across multiple markets, Ava was rolled out worldwide in English. The roll-out happened gradually in order to mitigate any risks. At the end of Q1 2026, Ava was available across all markets. The foundation for multilingual support has been laid and will follow the overall roadmap for this year.
Results Q1 2026
13,000+
Monthly active users - up 71% from 2025
7,500 hrs
Autonomous support delivered in 2026, freeing teams for complex cases
94% → 97%
Resolution rate improvement from 2025
25% → 3.1%
Escalation rate reduction within Q1 2026 alone
Key projects so far
Response quality - making Ava sound like Ava
My main task from day 1 at Awin has been to develop Ava as a user experience. As such, I have spearheaded the efforts to make Ava's responses more concise, conversational, and consistent with her established persona. I collaborated with other content designers on Awin's wider style guide and then used this to develop a dedicated appendix specifically for Ava. This covered how to design for LLM output and how to evaluate the chatbot's responses against quality and persona standards.
Gathering CSAT without causing unnecessary user friction
Getting feedback from users when engaging with a chatbot is never an easy task. People tend to only leave feedback when they're really frustrated at which point it starts to affect trust and the likelihood of them returning.
Having tackled this challenge across many chatbot projects, I know how difficult it is to implement a feedback feature that works for users. The original feedback process for Awin was cumbersome and complex, so I decided to overhaul the experience completely for Ava. Working closely with the dev team, I identified two low-friction moments in the journey: after a period of inactivity, and when the user closes the chatbot.
The feature went live at the start of 2026. In the first quarter, we achieved a consistent CSAT score of 3.3 out of 5. I have since conducted qualitative audits to better understand where user friction exists and how we can continue to improve Ava's performance.
Handover redesign: from friction to flow
After Ava went live globally, the team conducted research that showed the existing handover process was too complex and frustrating for Awin's users. Basically, they were being asked to click on a link that would take them to a form, which they then had to fill in from scratch which would then be reviewed by a Support Team colleague. This often led to the user either getting frustrated (which they would take out on the colleague once they got back to them), or they would just abandon their attempt to get help altogether.
As this is an all too familiar pattern for many chatbots, I knew we needed to upgrade this process to minimise frustration and friction. Along with stakeholders I led a full redesign focused on reducing user effort at the most critical transition point in the journey. We employed AI to summarise and pre-populate the form, so that the user just had to review, potentially edit, and then submit the ticket. It also meant that the Support now received tickets that were more uniform and, as such, easier to scan, which reduced case handling time.
Before
A complex, multi-step handover process that placed too much burden on the user at an already frustrating moment.
After
A streamlined experience using AI to summarise and pre-populate the user's issue. Users simply review, edit if needed, and submit. Effort reduced, trust maintained.
The approach
Strategic advocacy
I continue to act as the voice for safe, data-led, and user-centred AI - pushing back on speed-to-market pressure when the product wasn't ready, and ensuring decisions were grounded in data.
Persona & tone
Facilitated workshops to establish Ava's identity from the ground up, creating documentation that gave the whole team a shared standard for how our LLM-powered chatbot should sound and behave.
Cross-functional bridge
I sit as the key connector for engineering and UX. I have established myself as the voice of design within the engineering team, while also considering the requirements - and occasional constraints for the chatbot that the dev team has identified.
Stakeholder research
I proactively engaged Marketing and Customer Success to understand their needs, expectations - and even concerns when it came to deploying a chatbot for Awin. This helped me tremendously in ensuring Ava was designed around genuine user and business needs, rather than just assumptions.