FWAIP Case Study

Flywheel AI Partners, a business unit of ShooflyAI

Turning 19 Years of Legacy Data Into AI-Ready Customer Intelligence

How we helped Global1 Resources unlock two decades of hidden customer data and built a governed, reusable foundation for AI-powered operations and revenue programs.

0+ hours of manual work eliminated through intelligent automation

19 yrs of contact history recovered from unstructured invoice data

⚡

2.5 Weeks vs 6+ months of manual

01 / Snapshot

A Goldmine of Data Locked Away

Global1 Resources has been in business for almost two decades. Every sale runs through QuickBooks Desktop, and for 19 of those years the owner had been typing customer emails and phone numbers into the invoice description field instead of the structured contact fields.

They knew there was significant value locked in that history, but it was effectively unusable:

They couldn't easily reactivate dormant customers
They couldn't reliably export a clean contact list
They wanted future AI agents (quoting, outreach, account management) to be built on solid data, not half-manual workarounds

This is a familiar pattern: a company sitting on years of valuable behavioral and transactional data that isn't yet organized into a clean, governed, revenue-driving intelligence layer.

02 / The Problem

Stuck Between Two Bad Options

When we first met, Global1 was stuck between two paths, neither of them good.

Do it manually. Open invoices one by one and copy/paste emails and phone numbers into customer records.
Walk away from 19 years of history. Accept that the data is "there," but practically unreachable.

Key Constraints

Data in the wrong place: Contact details lived inside free-text invoice descriptions, not in standard fields that reporting, exports, and downstream systems (CRM, analytics, AI) are built around.
Export limits: Standard exports truncated descriptions and often skipped the lines that contained contact info.
Scale: Roughly 20,000 records × ~90 seconds of manual work each = ~500 hours of low-value effort.

Records to Process

~90s

Per Record (Manual)

Hours of Manual Work

Years of History

They'd already considered asking a developer to "hack something together" or bringing in an intern to grind through copy/paste work. Neither approach would produce a durable single source of truth the whole business could trust. Just another fragmented list.

03 / Objectives

Three Clear Goals

Together we set three clear objectives:

Collapse Manual Effort

Reduce ~500 hours of manual cleanup into a few weeks using automation instead of people.

Recover Contact History

Recover as much of the 19-year contact history as technically possible from QuickBooks Desktop.

Establish a Single Source of Truth

Return the cleaned data in an owned, governed format that could feed any future CRM, AI system, or reporting tool. This becomes the reference point when anyone asks "how many reachable customers do we have?"

04 / Our Role

Fractional AI Product & Data Team

We stepped in as a fractional AI product and data team, not a one-off script vendor or black-box tool.

Implemented a practical export strategy within the constraints of QuickBooks Desktop
Designed automation to parse, clean, and remap 19 years of invoice-level data
Structured the output so future AI agents could use it without re-doing the foundation work
Documented the process for repeatable, governed use going forward

The mandate was simple: turn messy, legacy data into a clean, governed, AI-ready asset Global1 actually owns.

05 / Approach

From Feasibility to Delivery

5.1 Proving Feasibility

We started with a quick feasibility pass:

Tested multiple export paths to access full invoice description text
Verified that emails and phone numbers were present in the underlying data, even if standard reports didn't expose them

Result: the data was there; the default exports just weren't surfacing it.

5.2 Finding the Right Export Path

Through follow-up working sessions, we explored additional export methods, including the legacy IIF (Intuit Interchange Format).

Key discovery: The IIF export preserved the full memo/description content, including the phone and email lines that standard CSV/Excel exports were dropping.

This confirmed we weren't dealing with data loss. We had an extraction and modeling problem that could be solved once, then reused.

We confirmed this by:

Searching for specific customers in the IIF file
Matching exported phone and email lines back to their historical invoices

This proved:

The problem wasn't data loss, it was an extraction problem
It required the correct extraction strategy
A fully automated cleanup was realistic

5.3 Cleaning, Structuring, and Remapping

With a reliable export secured, we scoped and delivered the cleanup.

Manual cleanup estimate ~500 hours

Automated approach Delivered in 2.5 weeks

Technical Steps

Ingest exported data into a processing environment

Parse invoice descriptions using consistent patterns

Normalize and dedupe contact details while preserving linkage to original records

Prepare a cleaned file for updating official contact fields

Structure output for feeding CRM or AI systems

Document the process for safe, repeatable ongoing use

We treated this as the first layer of an intelligence stack: a repeatable, governed process that creates a single source of truth, not a one-time rescue.

06 / Outcomes

Measurable Results

~500 hours of manual work avoided

Cleanup completed in weeks, not months.

19 years of contact history recovered

Previously trapped data now structured, governed, and trustworthy.

Single source of truth for outreach and AI

Clean contact dataset lives in the right fields and feeds every tool the same definitions. No more "which list is right?" debates.

AI-ready foundation

Quoting assistants, automated outreach, and other AI workflows now run on a consistent view of the customer instead of fragmented lists.

07 / Why This Pattern Matters

For Data-Rich, Campaign-Driven Businesses

Messy Foundations → Structured Truth

Many long-lived businesses accumulate critical data in the wrong fields, flat tables, or legacy schemas that no one designed for analytics. Before any meaningful AI or insights layer can exist, that information must be extracted, cleaned, and modeled into entities and relationships the business actually cares about: customers, contacts, transactions, engagement history.

From "Data Everywhere" to "One Version of the Truth"

The goal is a single, reliable view of activity and performance so that two teams asking the same question get the same number. Whether that means customer contact data, engagement funnels, or purchase behavior, everyone pulls from one governed source.

Intelligence Layer ≠ Science Experiment

Real leverage comes from structuring data around high-value questions and building the intelligence layer that can answer those questions instantly, accurately, and consistently. No more reconciling spreadsheets or debating whose export is correct.

For Global1 Resources, this meant turning 19 years of invoices into an AI-ready contact graph. For other data-rich, campaign-driven businesses, the same approach turns fragmented behavioral and transactional data into a governed foundation for AI agents and insights products that actually move revenue.