Accenture


AI Pricing Assistant


This project required us to work closely with a global team where we in the early stages of generative ai being used by businesses, created a curated chat bot for custom software.


This engagement has been adapted for client confidentiality, but I’ve shared below what I can publicly post. If you’d like to hear more or dive deeper into the work, feel free to reach out.

 

Role:

Product Designer


TEAM:

Lauren, Natalie - Design

Alex, Portia, Jeanette, Tom - Developers


Tools:

Figma, Figma Dev Mode, React Aria, Tailwind, HTML, CSS, Tailwind


Timeline:

March 2023 - May 2023

Project Overview:


When a client prepared to launch new pricing software, they needed more than a chatbot, rather, they needed a trusted assistant. In just six weeks, I helped design a generative AI solution that delivered accurate, source-backed answers, built user trust, and laid the foundation for scaling AI into sensitive pricing workflows.

The challenge



Accuracy, trust, and speed in a high-stakes rollout


When a confidential client prepared to launch new pricing software, they knew adoption hinged on trust. Users were expected to navigate complex, unfamiliar workflows—but without help, they risked errors with long-lasting financial consequences.


The business wanted a built-in assistant powered by generative AI, but the stakes were high:

  • Zero tolerance for hallucinations: in the pricing space, even small mistakes have outsized impacts.
  • Unreleased workflows: we were designing in parallel with a product still under development.
  • Tight deadlines: the AI assistant had to be conceived, tested, and delivered in just six weeks to secure funding for the product’s release.


And we had one more layer of complexity: a globally distributed team, which meant async collaboration across time zones and 5 a.m. calls to keep momentum moving.

Discovery



Finding the pain points, framing the risks


We started by embedding ourselves with pricing practitioners, support agents, and leadership. Through workshops, contextual inquiries, and documentation audits, we mapped out the reality:

  • Practitioners needed quick, reliable answers without derailing their workflow.
  • Support agents required clear escalation when the AI couldn’t resolve ambiguity.
  • Leadership demanded traceable, source-backed responses to uphold pricing discipline.


From this, three non-negotiables emerged: accuracy, transparency, and escalation. If we couldn’t solve for these, the assistant wouldn’t earn user trust—or leadership sign-off.

Define & Ideation



Designing for trust


With constraints clear, we explored multiple interaction models. Should the assistant appear inline, in a side panel, or in a modal? How would it indicate confidence? When should it hand off to humans? Through prototyping in Figma and rapid feedback loops with SMEs and early users, we defined a system with three pillars:

  • Contextual guidance—inline suggestions at the moment of need.
  • Persistent access—a side panel assistant for deeper queries.
  • Trust signals—confidence indicators, source citations, and explicit escalation paths.


On the backend, we adopted a retrieval-first approach: curating a knowledge base that prioritized verifiable answers over model creativity. In a world where hallucinations could cost millions, curation was the real innovation.

The solution



A co-pilot for pricing workflows


The result was a validated prototype for an AI assistant that:

  • Delivered concise, source-backed answers within the pricing workflow.
  • Escalated gracefully to experts when confidence dropped.
  • Automated repetitive setup tasks, reducing manual effort by ~90%.
  • Empowered admins with curation and governance tools to maintain trust at scale.
Impact



Reliability breeds adoption


Early pilots showed clear benefits:

  • Higher reliability and transparency compared to unguided AI tools.
  • Increased user confidence, reducing dependence on manual support.
  • Scalability—a pattern for extending AI assistance across other pricing processes.


Most importantly, the assistant helped the business secure executive buy-in for the larger product launch. Trust wasn’t just a design principle—it was the lever that unlocked adoption.

Next Steps



Scaling governance and reach


With the assistant validated, the roadmap focused on:

  • Building admin tooling for dataset versioning and governance.
  • Instrumenting accuracy, support load, and user satisfaction metrics.
  • Running A/B pilots to fine-tune confidence thresholds.
  • Scaling onboarding and localization for global rollout.


This wasn’t just a project delivered on a six-week deadline—it was the start of a broader AI strategy for the client, proving that generative AI can be a trusted partner in even the most sensitive business contexts