Supply chain systems
Process automation · Applied AI

D/L

David
Lemons

Business systems & automation

I sit with the people doing the work, find where the process breaks, and build the fix.

David Lemons

In short

I'm a senior business systems analyst at Viasat. I build the forecasting apps, automations, and AI agents that supply chain Operations runs on, from procurement through logistics.

Most of my work starts the same way. Someone is rebuilding a spreadsheet every week, answering the same question for the tenth time, or finding out about a shortage after it's too late. I watch how the work actually happens, map it, and build something that takes the busywork off their plate.

Delivered since January 2026

7Tools shipped in six months
1,500+Hours a year returned across Operations
$3.7MProjected 3-year value
$330KA year in licensing and support avoided

Patent
pending

Published international patent application · 2026 · Named inventor

Satellite service unit allocation

When satellite terminals are scarce, which regions and subscription services get them first? I co-invented a multi-variable allocation engine that answers that with math instead of gut feel.

Patent drawing: a distribution manager with region, prioritization, and inventory managers allocating a pool of satellite terminals across regional subscription services
Fig. 3 from the published application: a distribution manager allocating a shared terminal inventory across regions and subscription services

The problem

Constrained hardware has to be split across global markets, each with different demand, pricing, and commitments. That call used to be made by judgment in spreadsheets, and the trade-offs were hard to see or defend.

The invention

The engine builds every feasible allocation scenario, scores each one on long-term value across regions and subscription services, and ranks them. The winning plan is written back into the planning system, so the decision and the system of record never drift apart.

Read the published application

Selected
work

Seven tools built for Operations at Viasat, all live within the last year, plus one in build. Open any of them for the value delivered and a map of how it works.

01

Software Revision Automation

Commercial aviation configuration planning

01 / Outcome

$56K

3-year value: 144 hours a year back, and pegging that took hours now runs in minutes

View project

02

Statistical Forecasting Platform

In-house demand forecasting app

02 / Outcome

$1M+

3-year cost avoided versus SAP IBP licensing and support ($330K a year)

View project

03

Logistics AI Agent

LLM self-service in Slack

03 / Outcome

$392K

3-year value, with 843 support hours freed in year one and growing

View project

04

Ship Readiness Report

Clear-to-ship automation

04 / Outcome

$100K

3-year value: 260 hours a year of manual reporting gone, checks now daily

View project

05

Shortfall Early Warning

60-day risk dashboard

05 / Outcome

60 days

of shortfall warning on every order, plus 130 hours a year saved

View project

06

Backlog Software Reconciliation

Order-to-configuration matching

06 / Outcome

1 to 2 mo

earlier fixes for missing software definitions, before orders are built

View project

07

Month-End Inventory Reconciliation

Multi-site consolidation

07 / Outcome

Day to 5 min

month-end close, returning 95 hours a year with every override logged

View project

08

Inventory Budget Planner

In build, live next fiscal year

08 / Target

2 to 3%

targeted cut in inventory spend, with every quarter tracked against one approved budget

In progress

How
I work

Old-school industrial engineering with modern tools. Frank Gilbreth timed bricklayers to find wasted motion. I do the same thing with spreadsheets, ERP screens, and Slack threads.

  1. 01

    Sit with it

    Watch the work the way it really happens, not the way the SOP says it does.

  2. 02

    Map it

    Draw the process so everyone agrees on what's broken before anyone writes code.

  3. 03

    Build it

    Ship the smallest tool that fixes it, on top of the systems people already use.

  4. 04

    Stay for adoption

    Train, fix, and iterate until the new way is just the way.

My story

I grew up in San Diego. Right after high school I joined the Navy, where I spent more than five years as a Hospital Corpsman taking care of sailors and Marines. That included a deployment in support of Operation Inherent Resolve, training hundreds of junior corpsmen, and a Navy and Marine Corps Achievement Medal. The Navy taught me to stay calm when things break and to fix the system, not just the symptom.

After I got out, I went back to school. I earned my bachelor's at Cal State San Marcos, then my MBA in Management Information Systems, with an emphasis in data analytics, at Cal State LA.

That led me to Viasat as a business systems analyst, working with the MES and ERP systems behind our operations and manufacturing teams. As AI got good enough to matter, I started pairing it with my coding background to automate the busywork and build in-house tools that replace software we used to pay for. These days that's most of my job: finding the expensive, manual, or rented thing and building something better that we own.

Path

  1. Viasat · Senior Business Systems Analyst2022 to now
  2. Cal State LA · MBA, Information Systems2020 to 2022
  3. Cal State San Marcos · Bachelor's2017 to 2019
  4. US Navy · Hospital Corpsman2012 to 2017

Build

  • Python
  • SQL
  • VBA
  • Streamlit
  • Next.js
  • Supabase
  • Docker

Systems

  • Oracle R12
  • OBIEE
  • SAP S/4HANA
  • SAP IBP
  • MES
  • Starburst
  • PostgreSQL

Planning and process

  • S&OP
  • MRP
  • Inventory optimization
  • SAP Signavio
  • Change management

Applied AI

  • LLM integration
  • Agentic workflows
  • Kore.ai
  • Claude Code
← All work

Project 01 of 07

Software Revision Automation

Commercial aviation configuration planning

Role
Sole developer, co-designed with the program manager and QA
Status
Live, runs every monthly planning cycle
Stack
Excel · VBA · ERP and BI exports · Power BI

Value delivered

Money

$56K

3-year value ($18K a year)

Time

144 hrs

a year of manual work removed

Speed

Minutes

to re-run allocations, down from hours

Risk

0

single-person dependencies left, with wrong-software risk cut

Before

Every planning cycle, a program manager pulled up to eight reports and worked out, aircraft by aircraft, which software load and FAA 8130 airworthiness tag each shipment needed. Then a buyer rebuilt the supplier pegging file by hand, reapplying lead times, exclusions, and customer priorities from memory and old email threads. Bad upstream data usually didn't show up until the hardware was ready to ship.

Now

A rules engine that ingests all eight sources, assigns the software revision and 8130 tag for every order, then pegs supplier POs to customer orders with codified rules for lead times, customer priority, exclusions, and returns. When it can't verify a match it flags the row instead of guessing, so data problems surface at planning, not at the dock.

Process map

Before · manual
Pull 8 reportsMatch orders by handLook up software loadAssign 8130 tagRebuild pegging fileReapply priorities from email

Inputs · 7

  • Customer demand model
  • Open sales orders and backlog
  • Open POs and ship notices
  • Outside processing POs
  • Software configuration matrix
  • On-hand inventory
  • Lead times and priority rules

Automated logic

  1. 01Normalize and match keys
  2. 02Assign software revision
  3. 03Assign 8130 airworthiness tag
  4. 04Rank by customer priority
  5. 05Peg supplier POs to orders
  6. 06Flag rows it can't verify

Outputs

  • Dated planning workbook
  • Pegging plan for the supplier
  • Exceptions list for buyers
  • Data-quality flags for demand owners

In the field · October 2026

A routine monthly run flagged airline customer orders the tool couldn't match to the demand system. Instead of guessing, it left them blank and flagged them. Tracing those flags turned up four separate problems upstream:

  1. 01Orders stuck in an early status, so they were never assigned to a service site
  2. 02A BOM error that kept confirmed orders from driving any component demand
  3. 03A broken backlog sync that silently dropped orders from the demand model
  4. 04Shipsets accidentally converted to the wrong aircraft program after a schedule update

All four were fixed before the next demand refresh. Without the flags, each one would have shown up later as a missed ship date, an expedite, or parts bought for the wrong program.

Why it matters

  • Hours of work each cycle now runs in minutes, so allocations refresh whenever schedules, configurations, or shortages change
  • Sends the contract manufacturer accurate priorities and keeps customer commitments on track
  • Catches demand-data problems at planning instead of at shipment, avoiding expedites, missed ship dates, and parts bought against bad data
  • Removes a single-person dependency and lowers the risk of an aircraft shipping with the wrong software or airworthiness tag
← All work

Project 02 of 07

Statistical Forecasting Platform

In-house demand forecasting app

Role
Architect and lead developer
Status
V1 live and hosted with SSO, signed off by the business
Stack
Python · Streamlit · PostgreSQL

Value delivered

Money

$1M+

3-year cost avoided

Money

$330K

a year in licensing and support not spent

Accuracy

-14.5 pts

WAPE on a volatile series (41.8% to 27.3%)

Quality

First

measured, model-based forecast for two business units

Before

Planners forecast from customer signals and judgment in spreadsheets, and accuracy was rarely measured. The off-the-shelf option was an SAP IBP subscription at $200K+ a year, plus people to run it.

Now

A web app that produces statistical baseline forecasts (exponential smoothing, Holt-Winters) and rebuilds the planning grid people already know: demand, supply, and netting by product and month, with a BOM matrix, product families, and scenario modeling.

Process map

Before · manual
Collect customer signalsForecast by judgmentRebuild planning gridVersion by file name

Inputs · 4

  • Demand history
  • Customer demand signals
  • BOM and product families
  • Supply and on-hand

Automated logic

  1. 01Clean and aggregate history
  2. 02Fit statistical models
  3. 03Score accuracy (WAPE)
  4. 04Net demand against supply
  5. 05Model scenarios

Outputs

  • Baseline forecast by part
  • Monthly planning grid
  • Scenario comparisons

Why it matters

  • Avoids SAP IBP licensing plus the support staff it would need
  • Gives two business units their first measured, model-based forecast
  • Fixed a systematic bias that took a volatile series from 41.8% to 27.3% WAPE
← All work

Project 03 of 07

Logistics AI Agent

LLM self-service in Slack

Role
Product owner and builder
Status
Phase 1 passed UAT, Phase 2 expanding coverage
Stack
Kore.ai · LLM · Slack · Knowledge base

Value delivered

Money

$392K

3-year value

Time

843 hrs

of support time freed in year one, growing with coverage

Service

24/7

instant answers in Slack

Scale

Flat

effort as inquiry volume grows, no added headcount

Before

The global logistics team answered every routine question directly. Tracking, proof of delivery, shipping terms. About 281 hours of the team's time every month.

Now

A self-service agent in Slack, trained on a logistics knowledge base I documented from the team's real questions. Routine questions get an instant answer at any hour, and the team keeps the hard ones.

Process map

Before · manual
Question comes inWait in the queueExpert looks it upAnswer by hand

Inputs · 3

  • Historical inquiries
  • Logistics SOPs
  • Shipping terms and FAQs

Automated logic

  1. 01Rank top question types
  2. 02Document the answers
  3. 03Train intents
  4. 04Answer in Slack
  5. 05Review misses and expand

Outputs

  • Instant answers, 24/7
  • A written knowledge base
  • Team time back for complex work

Why it matters

  • About 843 hours a year freed at Phase 1 deflection rates, growing as coverage expands
  • Tracking, proof of delivery, and shipping-term questions answered instantly
  • Knowledge that lived in people's heads is now written down
← All work

Project 04 of 07

Ship Readiness Report

Clear-to-ship automation

Role
Sole developer
Status
Live, now on v3.7
Stack
Excel · VBA · BI exports

Value delivered

Money

$100K

3-year value

Time

260 hrs

a year of manual reporting removed

Speed

Daily

ship-readiness check, up from weekly

Accuracy

1 view

one source of truth for logistics and planning

Before

Analysts pulled several BI reports (orders, holds, warehouse on-hand) and stitched them together by hand every week to see what could ship.

Now

One run consolidates orders, holds, and serial-level warehouse inventory into a single clear-to-ship view, including lot-controlled parts and two on-hand sources.

Process map

Before · manual
Export ordersExport holdsExport warehouse stockStitch with lookupsRepeat weekly

Inputs · 4

  • Open orders
  • Active holds
  • Serial-level warehouse inventory
  • Lot-controlled stock

Automated logic

  1. 01Join orders, holds, and stock
  2. 02Net inventory by order
  3. 03Grade each line ship-ready or not

Outputs

  • Clear-to-ship view
  • Short and on-hold lines
  • One shared source of truth

Why it matters

  • About 260 hours a year of manual data gathering gone
  • Runs in minutes, so the check happens daily instead of weekly
  • Logistics and planning work from the same view, which speeds shipment decisions
← All work

Project 05 of 07

Shortfall Early Warning

60-day risk dashboard

Role
Sole developer
Status
Live
Stack
Excel · VBA · BI exports

Value delivered

Money

$50K

3-year labor value, before expedite savings

Time

130 hrs

a year of checks and escalations removed

Risk

60 days

of forward visibility on every order

Accuracy

Every order

graded red, yellow, or green

Before

There was no forward view. Inventory was checked right before shipping, and shortfalls went out by email at the last minute.

Now

A dashboard that grades shortfall risk on every order shipping in the next 60 days, by part and customer, as red, yellow, or green.

Process map

Before · manual
Check stock at ship timeFind the gapEmail escalationPay to expedite

Inputs · 3

  • Orders shipping in 60 days
  • On-hand inventory
  • Open supply

Automated logic

  1. 01Roll demand forward
  2. 02Net against supply
  3. 03Grade red / yellow / green

Outputs

  • Risk by part and customer
  • Early action list

Why it matters

  • About 130 hours a year of pre-shipment checks and escalations removed
  • Shortfalls show up weeks earlier, when a supplier call can still fix them
  • Protects delivery commitments to airline customers
← All work

Project 06 of 07

Backlog Software Reconciliation

Order-to-configuration matching

Role
Sole developer
Status
Live, run weekly
Stack
Excel · VBA · BI and SharePoint exports

Value delivered

Risk

1 to 2 mo

earlier detection of missing software definitions

Money

$21K

3-year labor value, before avoided rework and returns

Time

54 hrs

a year of manual matching removed

Accuracy

100%

of the backlog checked, not spot-checked

Before

The order backlog was spot-checked by hand against a software configuration matrix. Gaps often went unnoticed for weeks, sometimes until ship time.

Now

Matches the whole open backlog against the configuration matrix and gives every gap a disposition: remove, review, or define.

Process map

Before · manual
Export backlogOpen the matrixSpot-check by eyeChase gaps in Slack

Inputs · 2

  • Open order backlog
  • Software configuration matrix

Automated logic

  1. 01Match every order and config
  2. 02Classify gaps
  3. 03Assign remove / review / define

Outputs

  • Gap list with actions
  • Clean data for serial-level automation later

Why it matters

  • About 54 hours a year of manual matching removed
  • Missing definitions get fixed before orders are built, not after
  • Lowers the risk of shipping the wrong software, avoiding rework and returns
← All work

Project 07 of 07

Month-End Inventory Reconciliation

Multi-site consolidation

Role
Sole developer
Status
Live, monthly
Stack
Excel · VBA

Value delivered

Speed

5 min

down from a full workday every month

Time

95 hrs

a year of month-end work removed

Money

$37K

3-year value

Accuracy

100%

of ERP overrides logged for audit

Before

A planner merged four or five source files with lookups and compared ERP balances against third-party physical counts part by part. It took a full day every month.

Now

Merges cost accounting, international site, and third-party count data, then reconciles ERP against physical counts and logs every override.

Process map

Before · manual
Open 5 filesBuild lookupsCompare part by partFix overrides by hand

Inputs · 4

  • Cost accounting data
  • International site data
  • Third-party physical counts
  • ERP balances

Automated logic

  1. 01Normalize part numbers
  2. 02Merge sources
  3. 03Reconcile ERP vs. physical
  4. 04Log every override

Outputs

  • Same-day inventory baseline
  • Override audit log

Why it matters

  • About 95 hours a year of month-end effort removed
  • The inventory baseline is ready the same day
  • Every override is logged, which helps accuracy and audits, and anyone can run it

Let's
connect

Open to supply chain systems, operations automation, and applied AI roles. Remote from San Diego.

Email

dlemons.sd@gmail.com

LinkedIn

linkedin.com/in/davidlemonsmba

© 2026 David Lemons · San Diego, CA