Case Studies

What engineer-led AI actually looks like in production.

Representative engagements showing real workflows, real systems, and real results — from AI Discovery Sprint through production go-live.

Manufacturing · SAP ECC · Workflow Automation

PO Exception Handling — Specialty Chemical Distributor

180-seat specialty chemical distributor, Midwest · ~$120M revenue · SAP ECC 6.0 · Procurement team of 9 FTEs

Representative engagement — results based on typical AI Discovery Sprint + Pilot Implementation outcomes.

The Problem

Roughly 35% of inbound purchase orders required manual exception handling — mismatched unit-of-measure codes, vendor part number discrepancies, and freight terms that didn't map cleanly to SAP line items. Each exception averaged 22 minutes of analyst time, consuming ~23 staff-hours daily. A prior RPA initiative stalled after discovery revealed the ECC data model was too brittle for standard bot tooling.

What We Built

After a 3-week AI Discovery Sprint mapping the top 6 exception categories against SAP BAPI endpoints, we executed a Pilot Implementation: an AI layer that intercepts PO exceptions before they hit the analyst queue, cross-references vendor data, and either resolves autonomously or routes with a pre-filled correction proposal — integrated directly into the existing SAP workflow inbox. No UI changes for end users, no middleware replacement.

Results

62%exceptions auto-resolved
18.3 hrs/dayreclaimed for procurement team
$187Kannualized labor savings
11 weeksBlueprint to go-live
Logistics · Manhattan WMS · Document Intelligence

Freight Invoice Reconciliation — Regional 3PL

Regional 3PL, 4 DCs, Southeast · ~240 employees · Manhattan Associates WMS (on-prem, 2014) + QuickBooks Enterprise · 6-person billing team

Representative engagement — results based on typical AI Discovery Sprint + Pilot Implementation outcomes.

The Problem

Freight invoice reconciliation was almost entirely manual: each carrier invoice had to be matched against WMS load confirmations, then validated against a contracted rate card in Excel. Discrepancies — on ~28% of invoices — required back-and-forth with carriers and manual journal entries in QuickBooks. The process consumed ~26 staff-hours per week. A Power Automate pilot failed due to variability in carrier invoice formats (PDFs, EDI 210s, email attachments).

What We Built

The AI Discovery Sprint (4 weeks) identified document intake as the primary bottleneck. We executed a Pilot Implementation combining a document intelligence layer (handling PDF, EDI, and structured email extraction) with a reconciliation engine that queries the WMS API and rate card logic. Matched invoices post automatically to QuickBooks via SDK. Exceptions surface in a lightweight review UI with discrepancies pre-annotated and carrier contact pre-populated.

Results

71%invoices reconciled touchlessly
22 hrs/wkreclaimed in billing team
$94K/yrin recovered billing discrepancies
9 weeksBlueprint to go-live

Your Engagement

Every engagement starts with an AI Discovery Sprint.

2–3 weeks. Fixed fee. A prioritized roadmap with ROI estimates before a single line of production code is written.

We reply within 1 business day.