Case Studies

From Dashboards and Data Delays to Instant Answers: Building Liquidity Services' First AI Data Agent

Connecting Microsoft Copilot directly to marketplace data gave leadership board-ready answers in seconds, not days

Overview

Liquidity Services operates a portfolio of B2B and B2C reverse supply chain marketplaces, including GovDeals, AllSurplus, Liquidation.com, and Retail Rush. While the company already had a strong reporting analytics foundation with BigQuery and a suite of reporting dashboards, leadership wanted to take the next step in their AI journey and enable employees to interact directly with anonymized enterprise data using natural language.

Liquidity Services worked with Seer to build a custom Model Context Protocol (MCP) server that would connect BigQuery to their Microsoft Copilot Studio environment. With the custom MCP, business users can now ask questions, explore trends, and generate insights without relying on technical resources to write queries or build reports. The strategic investment helped Liquidity Services improve self-service analytics and create a scalable framework for future agentic AI use cases.

Client

Liquidity Services

Industry

[The Impact]
3 mins
to answer an ad hoc question (down from 3-4 hours)
3 sources
in one conversation (GA4, Google Ads, Search Console)
1 query
surfaced peak traffic days that marketing had been missing

The Challenge

Liquidity Services had already invested significantly in its analytics ecosystem, including BigQuery and a suite of Data Studio dashboards. While these tools offered visibility into marketplace performance, they required technical resources to write custom queries. Leadership wanted a more intuitive way for business users to ask questions, explore trends, and investigate topics like buyer behavior, bot traffic, and channel performance.

Liquidity Services sought to leverage Microsoft Copilot as a natural-language interface for enterprise data. That setup would let employees query live marketplace data using conversational AI inside a tool they already used daily, while maintaining appropriate security and governance.

The company enlisted Seer to help design and build the framework to connect Copilot Studio and BigQuery. The catch: there was no established blueprint for this type of implementation. The project required real-time problem solving across Google Cloud, Microsoft Azure, and Copilot Studio, and Model Context Protocol (MCP).

The Strategy

Seer scoped this as a hands-on build, not a configuration exercise, and made sure Liquidity Services' team could run with the tool once it was live.

  • Custom MCP server build: Built a purpose-made MCP server as an Azure Function, since no existing connector supported Copilot Studio to BigQuery communication out of the box. This included writing the server code, wiring up BigQuery client initialization, and standing up Azure Key Vault for credential storage.
  • Cross-platform authentication: Solved the hard problem of authenticating Copilot Studio against Google Cloud via OAuth 2.0. The Seer team sourced the right authorization and token URLs, defined BigQuery read-only scopes, and worked through Copilot Studio's inconsistent handling of MCP versus custom-connector modes, which repeatedly redirected the team to unrelated support documentation.
  • Cost monitoring and guardrails: Built GCP billing budgets and threshold alerts with the team, walking through real billing data together to see what individual queries and days of usage actually cost. That visibility gave leadership the confidence to expand access without worrying about runaway spend, and it now informs how the team sizes next year's token budget.
  • Grounding the model, not just connecting it: Once the connection was stable, Seer built knowledge-base instructions and data-quality guardrails inside the agent. Rules like "never guess, always query, never estimate a number" keep the tool behaving like a deterministic reporting layer rather than a chatbot prone to hallucination. The agent also closes each response with a follow-up prompt, which nudges users from a single answer toward a line of inquiry.
  • Enablement over handoff: Seer stayed hands-on after the MCP went live, working alongside the Liquidity Services team in live sessions as they tested new questions, refined the knowledge base, and rolled the tool out to more of the team. Seer pulled in its own MCP specialists, including an engineer who had prototyped a similar server on Seer's data, and worked directly with Liquidity Services' IT and Azure resources in live troubleshooting sessions rather than one-way handoffs.

Ask in Copilot, get answers in minutes from BigQuery

Knowledge base

Grounding rules and definitions

Connected sources

GA4, Google Ads, Search Console

Copilot Studio

Microsoft 365

MCP server

Azure Function

OAuth 2.0

Read-only

BigQuery

Data warehouse

What the team gets back

 

Verified data

Pulled straight from the source

 

Deeper analysis

Insights you can drill into

 

Better decisions

Marketing planning and optimization

seerinteractive

The Results

The connection works, and the numbers hold up. After several months of iterative building and troubleshooting, Liquidity Services' Copilot environment can reliably query anonymized live BigQuery data and return answers pulled straight from the warehouse, validated against existing GA4 numbers for accuracy. The agent now spans three connected sources, GA4, Google Ads, and Google Search Console, so a single question can cross paid, organic, and on-site behavior. It also maintains conversation context across sessions, something the team's prior Copilot Studio setup could not do.

Marketing stopped filing data requests and started building strategy. In a knowledge-sharing session in August 2026, the marketing team walked through what they had done with the tool on their own:

  • Improved campaign timing: A marketplace marketing leader built a full seller campaign plan from queries alone. They compared month-to-date performance against the prior month, pulled traffic sources and state-level geography, and surfaced that the account's two highest-traffic days were Sunday and Monday. Campaign timing had been missing both. The team adjusted its messaging schedule as a result.
  • More efficient campaign builds: The same session produced campaign concepts, a mapped email sequence, and a set of blog topics from follow-up prompts against the same data.
  • New keyword opportunities: A paid search leader cross-referenced Google Ads and Search Console to compare paid against organic, identify where the two reinforced each other, and pull a list of strong organic terms not yet running in paid. Landing page selection and audience segmentation questions ran against GA4. Negative keyword list expansion, which was previously tedious and manual, now runs in minutes.
  • Refined navigation parameters: Marketing and analytics together wrote a query to decode numerically encoded side-navigation parameters on Liquidation.com, producing usage data on specific nav elements that had been impractical to dashboard. That analysis is now informing navigation ranking recommendations to the IT team.
  • Regional traffic analysis: Event marketing is using the agent to trace regional traffic through to registrations, bids, and watches by source and medium.

 

Cost is visible, so access can grow. With budget alerts and real usage data in place, the team can see what the tool costs and set sensible thresholds, which is what makes expanding access beyond the initial pilot group a manageable decision rather than an open-ended one. Sales enablement is next in line.

The pilot changed the roadmap. Adoption data is now shaping what gets built next: reusable agent skills for workflows people repeat, and scheduled insights delivered to inboxes to replace manual weekly reporting that currently requires pulling from several systems by hand. Liquidity Services is hiring dedicated Copilot development support to take that on.

Looking to build a custom MCP? We can help.

Let's Talk

S-stack

"Building on the BigQuery data foundation they helped establish, Seer introduced an MCP and agentic AI strategy that aligned closely with our vision for AI-enabled decision support. Liquidity Services recognized the potential business impact and invested in the initiative to accelerate its AI roadmap."

- Srikanta Rao, Senior Director of Analytics, Liquidity Services

[Up Next]

Fortune 50 Company Cuts Content Production Time by 50% While Staying On-Brand with New Seer-Crafted AI Content Model (SCAI)

Check it out
SEO / GEO

From Traffic Declines to Increased Brand Authority: How Trex Turned AI Overviews Into a Traffic Driver

Check it out

About Seer Interactive: 

Seer Interactive is at the forefront of Generative Engine Optimization (GEO). We run enterprise-scale AI search experiments and turn them into proven strategies that deliver AI visibility and business impact. We help enterprise brands navigate the shift between traditional SEO to AI visibility, building strategies that win across search engines, LLMs, and wherever your customer is looking for answers.

Our edge? Decades of thinking about how people get answers to problems, across social, SEO, Paid, and UX, now applied to AI search through live GEO experiments that deliver measurable visibility. Wil Reynolds and the Seer team have been leading the conversation on AI, SEO, and GEO since the beginning.

Seer partners with industries from SaaS and e-commerce to healthcare and banking , helping brands like American Family Insurance and Capital One adapt, move faster, and capture visibility in an AI-first world.