5 minutes read

Designing AI Co‑Pilot for Vendor Management

Ai Co-Pilot
SAAS
Vendor Management

AI-powered chat agent that collapses five separate SaaS modules into a single conversational interface, enabling property managers to work without leaving the chat.

Read Case Study

Drag me

Screenshot of a chat interface on the left discussing pricing details for granite countertop installation and labor costs, alongside a project scope summary on the right detailing Sunshine Springs 1x1 project in Jacksonville with budget, dates, and itemized work including smoke detector and interior paint.
Screenshot of a dark-themed AI chat interface on the left showing a conversation about pricing details for granite countertop installation and labor costs, and a project scope of work details on the right for Sunshine Springs 1x1 including project info, budget, dates, and a materials list with quantities and prices.
01 — Overview

The Challenge

Wreno is a B2B SaaS vendor management platform built for the real estate industry. Property managers, owners, and operators use Wreno to find, vet, onboard, and manage third-party vendors for everything from routine maintenance to large-scale renovation projects.

The platform serves a network of over 400,000 vendors across the United States, handling the entire lifecycle of vendor relationships: compliance credentialing (with AI-powered verification of W9s, COIs, and trade licenses in as little as 2 minutes), scope of work creation, RFP distribution, bid comparison, contract management, and project tracking.

I joined Wreno as a Product Designer and was responsible for designing both the core SaaS platform and the AI CoPilot: a conversational agent that reimagines how property managers interact with the entire vendor management workflow.

02 — The Problem

Vendor procurement was broken

Property managers managing hundreds of properties faced a painful daily reality: when a resident submits a work order, the clock starts ticking. But finding the right vendor meant scrolling through search engines, calling individual contractors, chasing down insurance certificates, and gathering quotes into disconnected spreadsheets.

There was no centralized way to compare pricing against market rates, verify vendor credentials instantly, or make confident decisions under time pressure. Every delayed maintenance job eroded resident satisfaction and increased holding costs.

1.

Fragmented Workflows

Creating a scope of work, finding vendors, sending RFPs, comparing bids, and tracking projects all lived in separate modules. Property managers context-switched constantly between screens to complete a single end-to-end task.

2.

Manual, Repetitive Processes

Every new project required building scopes from scratch, manually searching vendor databases, and comparing bid documents side by side. The same patterns repeated across hundreds of units with no automation.

3.

No Portfolio-Level Visibility

Managers overseeing multiple properties across markets had no quick way to get a bird's-eye view of budgets, pending bids, or project status without clicking through individual records one by one.

4.

Decision Fatigue in Bid Evaluation

Comparing vendor bids involved evaluating cost, timeline, certifications, warranties, and references across multiple vendors simultaneously. Without intelligent comparison tools, the best vendor didn't always win.

Key Insight

Property managers spend 60% of their time on vendor coordination, not managing properties.

Wireframe flowchart for a Scope of Work (SOW) setup process, showing screens for starting SOW, choosing editing methods, adding rooms and items, selecting vendors, and finalizing details.
03 — The IDEA

One conversation to replace four modules

The Wreno CoPilot is an AI-powered conversational agent that allows property managers to accomplish their entire vendor management workflow through natural language and structured UI components embedded within a chat interface.

Instead of navigating between separate screens for scope creation, vendor search, bid management, project tracking, and budgeting, the CoPilot guides users through each workflow conversationally. It asks one question at a time, confirms changes in real time, and surfaces structured data exactly when the user needs it: tables, comparisons, document previews, vendor cards, and analytics.

The core design challenge was finding the right balance between conversational flexibility (letting users type anything) and guided interaction (providing structured choices when decisions are discrete). Too much free text creates uncertainty about what the system can do; too many buttons defeats the purpose of a conversational interface.

+ New Chat
S
Today
Ready to get started? Let us know how we can help you today.
Create SOW
Send Bid
Find Vendors
Monitor Status
⚡ 90
Write a prompt for Wreno CoPilot
04 — Interaction Patterns

The design system behind the conversation

Designing an AI chat agent isn't just about the words. It's about knowing when to use buttons versus free text, when to show a table versus a single answer, and when to switch from chat-only to a split-panel layout. These patterns formed the backbone of every CoPilot interaction.

Hybrid Input Model
Structured button choices for discrete decisions (template vs. scratch, pricing source). Free text for open-ended input (edit instructions, budget questions). The CoPilot dynamically switches between modes based on what the moment requires.
CoPilot
How would you like to price the granite countertop?
My Catalog Ask Wreno Ask Vendor
Structured → Free text
Add granite countertop in Kitchen and remove carpet from bedroom 1.
Done! Budget updated to $607.06
Contextual Memory
The CoPilot maintains context across the conversation. Stacked filters, follow-up questions, and "also do this" instructions all work because the system remembers what's already been established in the thread.
Filter Stack
Show me projects starting in February
Showing 4 projects
February ×
Show me the ones requiring bid approval
Narrowed to 2 projects
February × Bids Pending ×
Credit-Based Usage
A visible credit counter (starting at 90) in the input bar sets expectations about usage limits. The counter decrements with each interaction, making the cost of each query transparent and predictable.
Usage Tracking
⚡ 67
Write a prompt...
90
Start
→
89
SOW
→
85
Edits
→
78
Vendors
→
67
Award
Split-Panel Preview
When editing documents (SOWs, contracts), the layout shifts to chat on the left and a live-updating document on the right with download capability. Changes reflect in real time as the user gives instructions.
Chat
Template loaded: Sunshine Springs 1x1
Remove carpet from bedroom 1
Done! Line item removed.
Add granite countertop in Kitchen
Added. Budget: $607.06
SOW Preview ↧ PDF
Smoke detector$64.53
Interior paint 150sf$288.00
Granite countertopTBD
GFCI outlet$56.48
Single-handle faucet$198.05
Total$607.06
Progressive Disclosure Tables
Data surfaces at the right level of detail for the question asked. Budget queries get one number first. Project lists get filterable tables. Bid comparisons get multi-column evaluation grids with downloadable exports.
Response Density
#
Budget Query
"Total for Florida?" → $5,322,490
1 VALUE
☰
Project List
Filterable table with dates, status, CTAs
TABLE
▦
Bid Comparison
Cost, timeline, warranty, certs across vendors
GRID
Vendor Cards with Feedback
Vendor recommendations include detailed profile cards showing service type, compliance status, company background, and contact info. Thumbs up/down controls train the system while giving users agency over results.
Vendor Matches
Etico Construction Group
▲
▼
Verified General Contractor Jacksonville, FL W9 · COI · License
ProJax Maintenance Co.
▲
▼
Verified Maintenance & Repair Jacksonville, FL W9 · COI
Star Plumbing Solutions
▲
▼
Pending Plumbing
05 — In Action

A conversation that builds a project

Here's a compressed view of how the SOW creation flow unfolds in practice, from the first template selection through vendor award:

+ New Chat
CoPilot
Great! Do you want to use a template or create from scratch?
Use Template Create from Scratch
Which template would you like to use?
TemplateMarketProperty
Sunshine Springs 1x1JacksonvilleSunshine SpringsSelect
ORL Renovation StdOrlandoAll PropertiesSelect
Natural Language Edit
Add granite countertop installation in Kitchen and remove carpet from bedroom 1.
Done! Budget updated to $607.06. What pricing would you like for the granite countertop?
My Catalog Ask Wreno Ask the Vendor
Ask the vendor to bid on it.
Bid Comparison
CoPilot
Here are the top 2 bids for your project:
CriteriaEtico ConstructionProJax Maintenance
Total Cost$1,112.06$1,239.06
Granite Install$505.00$632.50
Timeline2–3 days1–2 days
Warranty30-day labor + materials60-day labor
Award Etico Award ProJax
Vendor Awarded
Etico Construction Group, Inc.
Vendor notified and will receive your contact information. Check on your project status at any time.
Review Project View Vendor Contact
⚡ 67
Write a prompt for Wreno CoPilot
06 — Prototype

SOW Creation

This flow verifies that a Statement of Work (SOW) can be generated from a template.

  • Use Template / Create from Scratch trigger correct flows
  • Project metadata and budget populate correctly with items grouped by category.
  • Line item table loads correctly, bot confirms selection, credits drop from 85.
Screenshot of Wreno interface displaying a chat-driven scope of work creation for Sunshine Springs 1x1, showing template selection, chat commands, and a detailed scope of work table with budget and itemized tasks.
06 — Prototype

Submitting a Bid

This flow validates comparing vendor bids and awarding the project.

  • Compare Bids shows vendor comparison table with key criteria.
  • Vendor actions: Etico, ProJax, or Find More Vendors.
  • Award confirmation appears,  Vendor Contact cards show, credits drop to 67.
Chat interface showing bid comparison table between Etico Construction Group, Inc. and ProJax Maintenance with project costs, timeline, warranties, and vendor notes, along with options to accept bids and review project or contact vendor.
06 — Prototype

Editing SOW in Chat

This flow tests editing the SOW using natural language in chat and verifying updates in the table.

  • Chat edit adds granite countertop and removes bedroom carpet.
  • Bot prompts pricing options: My Catalog, Ask Wreno, Ask the Vendor.
  • Photo + specs attach successfully, timestamps show, credits drop to 81.
Screenshot of Wreno app displaying a chat about editing a renovation scope of work and a detailed product list for Bedroom 1 and Kitchen including granite countertop installation with photo of a modern kitchen featuring granite countertops.
07 — Lessons Learned

What this project taught me

A design system is only as good as its adoption. While working on the platform, I needed to make the system effortless to use correctly and hard to use incorrectly.

1.

Conversation Is Not Always the Answer

The hardest design decisions weren't about what to put in the chat, but what to pull out of it. Some interactions (like reviewing a 20-line SOW) need a dedicated panel, not a chat bubble. The split-panel pattern emerged from recognizing when conversation hits its limits.

2.

Trust Requires Transparency

Users won't trust an AI agent that changes their data behind a black box. The live-updating document preview, the credit counter, and the explicit confirmation at every stage were all designed to make the system's actions visible and reversible.

3.

Design the Boundaries, Not Just the Happy Path

Knowing when to show buttons versus accept free text, when to confirm versus auto-proceed, and when to drill down versus surface everything. The quality of an AI agent is defined by its constraints, not its capabilities.

4.

Build a Contribution Model Early

I should have established a clear contribution and governance process from the start — how to propose new components, how changes get reviewed, and how tokens get updated.
Case Studies

View More Work

SaaS

public Safety

User interface of emergency response software showing active shooter and kidnapping in progress, with unit status and dispatch comments.
Computer Aided Diptach (CAD) for 911

A browser-based CAD system that replaced decades-old dispatch software for 911.

View project

Design System

Web3

AI

Dashboard interface showing visitor analytics with charts for visitors, wallets, transactions, sessions, and page visits, including acquisition data and last seen wallet activity.
Design system for Web3 analytics

Design system with an intelligent chart library that enables teams to make decisions they trust.

View project