I lead the design of AiDA (Ansarada’s AI agent) as a Senior Product Designer, which is now being used in 25% of all live subscriptions in Ansarada.
I launched AiDA with our team in October 2025, evolving her from being a simple pre-prompt based chatbot to now, an increasingly nuanced AI assistant who provides rich insights to customers inside the Ansarada platform.
How can we leverage AI to help customers reduce or eliminate administrative work in a high-stakes M&A deal?
Human error is extremely costly in M&A. How can we leverage AI to reduce this possibility and risk?
My design process
Establish a shared problem
Our team and I only had an extremely tight runway of 6 weeks to design and launch AiDA. We knew what we had to do but there was no shared understanding of what LLM's could do back in October 2025, nor a clear vision of what our MVP would be.
To address this gap, I researched and catalogued all of the different response types best-in-class LLM’s could generate, mapping those outputs to data sources we could connect in Ansarada.
I translated my observations into a capability map, plotting how user prompts in AiDA would translate into outputs. I leveraged this artefact to guide our team and helped establish must-haves and nice-to-haves for AiDA's launch and roadmap.
Designing in parallel with Engineering
By defining the scope and AiDA's capabilities, the next challenge was iterating designs quickly for engineering to start technical estimations and development.
I led daily design reviews with engineering, product and senior leadership to gain clarity and alignment.
Together, we stress-tested interactions, validated the feasibility of designs and locked down constraints as a collective. This ensured everyone was clear on what was achievable and so we could be flexible and make ad-hoc adjustments to deliver on time.
Importantly, by running together with Engineer and senior stakeholders and establishing clear deliverables - it created space for myself and the Product Manager to tackle onboarding and activation.
Driving quality of response
To ensure we launched on time, our team compromised on AiDA’s initial capabilities. AiDA at launch, could only provide answers based on a set of pre-written and controlled prompts.
However this could not become the status quo. AiDA had to become a dynamic and intelligent assistant grounded with context and not just a basic chatbot, which could not scale or deliver sufficient value.
To move towards this direction, I extended myself and participated in quality assurance rituals and processes, testing AiDA and refining her underlying system prompt. By embedding myself into the QA process, it gave me the opportunity to influence stakeholders on what quality meant and the vision.
My persistence paid off, because by December 2025 (Release 2), our engineers discovered they could enhance AiDA with RAG (Retrieval-Augmented Generation) and here, I created another artefact to help our team prioritise the key databases we had to unlock to help customers reduce administrative work and deal-risk.
To have influenced this outcome, I have had to be consistently involved in the AI quality assurance and system prompt design process, establishing myself as one of the quality gates for reliability, tone, and trust.
Outcome
Most recently I just designed and launched AiDA to a platform level, enabling customers to ask AiDA across all of their subscriptions from one place.
