AI for Work Nobody Wants to Do
AI workflows that read, sort, summarise, and move business work forward.
LIVE SIGNAL · AI WORKFLOWS · V1.0 · PROD
0103
In plain English
This case study is about using AI where it actually helps: reading messy documents, extracting useful details, searching internal knowledge, and turning scattered business information into cleaner CRM workflows. The goal was not to build a flashy chatbot. It was to reduce manual admin work, improve visibility, and help teams move faster with outputs that people can still review and trust.
Business value
- Less time spent on manual document review
- Messy files became cleaner business data
- Search became smarter by finding related meaning, not just matching exact words
- Important details were easier to spot before they got buried in daily operations
- Reviewable AI outputs before action
- Role
- Full Stack Owner / AI Workflow Engineer
- Design · Build · Ship
- Timeframe
- Recent production work
- Domain
- CRM/ERP Operations, AI Workflows, Internal Systems
- Category
- ai_automation
- Built AI assisted workflows for reading documents, emails, notes, and CRM records more efficiently
- Used OCR and structured extraction to turn messy files into cleaner, reviewable business data
- Applied embeddings to make internal knowledge searchable by meaning, not only exact keywords
- Designed the AI layer around human review, confidence checks, and practical business rules
- Focused on reducing manual admin work while keeping the CRM as the source of truth
- Built AI assisted workflows for reading documents, emails, notes, and CRM records more efficiently
- Used OCR and structured extraction to turn messy files into cleaner, reviewable business data
- Applied embeddings to make internal knowledge searchable by meaning, not only exact keywords
- Designed the AI layer around human review, confidence checks, and practical business rules
- Focused on reducing manual admin work while keeping the CRM as the source of truth
Where the Work Gets Messy
A lot of business work still starts in messy places: scanned documents, PDFs, email threads, old notes, uploaded files, and CRM/ERP records that are not always written in the same format.
I wanted to explore how AI could help with the boring but important parts of operations: reading documents, extracting useful details, finding related information, and preparing cleaner data for the team to review.
The goal was not to create another AI toy or a chatbot sitting outside the system. The goal was to connect AI into the actual CRM/ERP workflow so it could support daily work without replacing the people responsible for checking it.
Scattered Data, Slow Decisions
The problem was not that the business lacked data. The problem was that useful information was scattered across too many formats.
Some details lived inside documents. Some were buried in notes. Some were attached to emails. Some needed to be searched by meaning, not by exact wording. When teams are busy, that kind of information can be slow to find and easy to miss.
Manual checking also does not scale well. Reading documents, copying fields, comparing records, and preparing updates takes time. It is important work, but it is also repetitive work that can slow down the rest of the operation.
The problem was not a lack of data. It was that useful information was trapped in documents, notes, emails, and scattered CRM/ERP records.
Ownership
Everything I designed, built, and was accountable for.
Product & UX
- AI workflow architecture
- Human review workflow
Engineering
- CRM/ERP integration logic
- API and backend implementation
- Frontend workflow and review UI
Additional scope
- OCR extraction flow
- Structured data mapping
- Embedding and vector search design
- Business rules and confidence checks
- Database schema and data flow
Key decisions
The calls I made, what I rejected, and why: these are the tradeoffs that shaped the system.
Keep AI inside the CRM/ERP workflow
A separate AI tool that users would need to open, manage, and trust outside the main system
The CRM/ERP needed to remain the source of truth. AI was only useful if it helped the existing workflow move faster without creating another disconnected place for business data to live.
The AI Workflow Layer
I designed the workflow around a simple idea: AI should prepare the work, but people should still control the decision.
The system starts with document or text intake. OCR reads the file and turns the content into usable text. Structured extraction then pulls out important fields such as names, dates, references, document details, notes, and other business specific signals.
For search and knowledge retrieval, embeddings are used to convert content into meaning based representations. This allows the system to find related records, similar documents, or relevant internal information even when the wording is not exactly the same.
The AI layer sits inside the CRM/ERP workflow instead of replacing it. Outputs are designed to be reviewable, traceable, and supported by confidence checks and business rules. The CRM/ERP remains the source of truth, while AI helps reduce the manual effort around reading, sorting, searching, and preparing data.
AI only becomes useful when it is connected to the workflow, reviewable by people, and grounded in the system the business already trusts.
0104
review · validation · workflow feedback loop
Frontend
Backend
Database
Integrations
Also used
Cleaner Signals, Faster Workflows
- Less manual admin
- Smarter CRM/ERP search
- Reviewable AI outputs
- Cleaner business data
- Workflow first automation
The result was a more practical AI workflow focused on real operational value.
Documents became easier to process. Important fields became easier to extract. Internal information became easier to search. Teams could spend less time reading through scattered records and more time acting on the right information.
The biggest improvement was not just speed. It was clarity. AI helped turn messy inputs into cleaner business signals while still keeping human review and CRM/ERP rules at the centre of the workflow.
~60% less manual admin effort in document and CRM/ERP review workflows
0106
~60% less admin effort
Faster context finding
Cleaner structured data
More usable records
Clearer team handoff
Controlled AI workflow
This gave the business a cleaner way to handle information that usually gets buried inside documents, emails, notes, and CRM/ERP records. Instead of relying only on manual reading and copying, teams could review AI prepared fields, search related context faster, and move work forward with better visibility. The main value was not replacing people. It was reducing the slow admin layer around business operations so teams could spend more time checking, deciding, and acting.
“Rusty understands the difference between adding features and making software actually usable. He looks at how people work, finds the friction, and improves the system in a way that makes daily operations feel smoother.”
Operations Stakeholder
Internal Platform Team — name under NDA
What This Changed
This project reinforced one of my strongest beliefs about AI in business systems: the useful part is rarely the flashy part.
A chatbot can look impressive in a demo, but most operational value comes from smaller workflow improvements. Reading a document faster. Extracting the right fields. Finding related records. Summarising messy notes. Preparing cleaner information before a person reviews it.
That is where AI becomes valuable inside a CRM/ERP platform.
The technical side matters, but the workflow design matters more. OCR, embeddings, vector search, structured extraction, and AI summaries only become useful when they are connected to the way the business already works.
I also wanted the system to avoid blind automation. AI outputs should not silently change important records without review. They should be explainable enough for teams to trust, practical enough to save time, and controlled enough to fit real business operations.
For me, the best AI systems are not the loudest ones. They are the ones that quietly remove friction from work nobody wants to do, while keeping the business in control.
The best AI systems do not replace the workflow. They remove friction from the parts nobody wants to do manually.
Quiet AI. Cleaner workflows. Better business signals.
8+
3
5+
10+
Need AI that actually fits your business workflow?
I help businesses design practical AI workflows around CRM/ERP systems, documents, internal tools, and operational data. If your team is still buried in manual checking, scattered records, or repetitive admin work, I can help turn that into something cleaner, reviewable, and easier to run.
Discuss an AI Workflow