Start with the process.
Not the industry.
Good opportunities usually have the same shape: lots of repetition, lots of information, decisions, hand-offs, or several systems that have to agree.
Pick an area. Watch the process.
Nine areas, nine processes. Notice that it is the same drawing every time — that is the point. Once you can read one, you can read all of them.
Lead to opportunity
A person steps in for one thing: strategic lead or low confidence.
Lead to opportunity: Form, then Validate, then Enrich company, then Check duplicates, then Score, then Assign owner, then Create CRM record, then Prepare follow-up, then Schedule, then Update CRM. One case travels the whole chain. The AI does every step and stops at Score, where a person approves — strategic lead or low confidence. The case ends as: Opportunity.
The AI updates the CRM
opportunity
Campaign operations
A person steps in for one thing: approve creative and budget.
Campaign operations: Brief, then Audience, then Assets, then UTMs & naming, then Campaign setup, then QA, then Launch, then Monitor, then Report, then Recommend next action. One case travels the whole chain. The AI does every step and stops at QA, where a person approves — approve creative and budget. The case ends as: Campaign live.
The AI recommends what is next
campaign live
New customer to active customer
A person steps in for one thing: missing information — ask the customer.
New customer to active customer: Form, then Read documents, then Validate data, then Create CRM account, then Create ERP record, then Create project, then Billing setup, then Provision access, then Send welcome, then Notify owner. One case travels the whole chain. The AI does every step and stops at Validate data, where a person approves — missing information — ask the customer. The case ends as: Customer active.
The AI notifies the owner
customer active
Product and order operations
A person steps in for one thing: order exception or refund.
Product and order operations: Order, then Ecommerce, then PIM, then Stock check, then ERP, then Logistics, then CRM, then Notify customer. One case travels the whole chain. The AI does every step and stops at Stock check, where a person approves — order exception or refund. The case ends as: Order fulfilled.
The AI tells the customer
order fulfilled
Invoice to ERP
A person steps in for one thing: approve the payment.
Invoice to ERP: Email, then Read invoice, then Extract fields, then Find supplier, then Match PO, then Check rules, then Approval, then Post to ERP, then Archive, then Report exception. One case travels the whole chain. The AI does every step and stops at Approval, where a person approves — approve the payment. The case ends as: Posted.
The AI reports what did not match
posted
Request to resolution
A person steps in for one thing: sensitive, unusual or uncertain.
Request to resolution: Message, then Understand issue, then Find customer, then Retrieve history, then Check policy, then Take action, then Answer, then Update CRM, then Close. One case travels the whole chain. The AI does every step and stops at Check policy, where a person approves — sensitive, unusual or uncertain. The case ends as: Resolved.
The AI closes it
resolved
Ticket triage and system sync
A person steps in for one thing: conflicting records.
Ticket triage and system sync: Event, then Classify, then Enrich, then Check data quality, then Update master data, then Sync ERP & CRM, then Report. One case travels the whole chain. The AI does every step and stops at Check data quality, where a person approves — conflicting records. The case ends as: Systems agree.
The AI reports
systems agree
Content production
A person steps in for one thing: approve the content.
Content production: Brief, then Research, then Draft, then QA, then Metadata, then Approval, then CMS, then Measure. One case travels the whole chain. The AI does every step and stops at Approval, where a person approves — approve the content. The case ends as: Published.
The AI measures how it did
published
Access request to closed ticket
A person steps in for one thing: approve the access.
Access request to closed ticket: Request, then Identify user, then Check policy, then Approval, then Grant access, then Log it, then Notify, then Close. One case travels the whole chain. The AI does every step and stops at Approval, where a person approves — approve the access. The case ends as: Access granted.
The AI closes the ticket
access granted
The process can start anywhere.
A process does not need to start on a website. Calls, forms, scans, photos, barcodes and other physical events can start or continue a digital one. No API does not always mean no automation.
Online
- Form
- CRM
- Website
- API
- Chat
- Database
- App
Offline
- Call
- Document
- Photo
- Store
- Warehouse
- Meeting
- Machine
- Field visit
Same workflow.
01
A phone call
A customer phones. The call becomes a transcript, the transcript becomes an intent, the intent becomes a case and an action in the CRM.
Voice becomes a case: Call, then Transcribe, then Understand, then Create case, then Act. One case travels the whole chain. The AI does every step and stops at Understand, where a person approves — unclear request. The case ends as: Case open.
The AI acts on it
case open
02
A paper form
An invoice is photographed. The document is recognised, the fields are extracted, the supplier is matched and it lands in the ERP.
Paper becomes data: Scan, then Read it, then Extract fields, then Validate, then Post to system. One case travels the whole chain. The AI does every step and stops at Validate, where a person approves — unreadable field. The case ends as: In the system.
The AI posts it to the system
in the system
03
A field team
Someone on site says what they see. It becomes a maintenance ticket, with a location attached and the right person notified.
A photo becomes an issue: Photo, then Detect issue, then Classify, then Create work order, then Assign, then Notify. One case travels the whole chain. The AI does every step and stops at Classify, where a person approves — safety risk. The case ends as: Work order out.
The AI tells the team
work order out
04
Legacy software
No API socket appears. The workflow clicks, types, searches and saves in the application itself, then carries on into a modern CRM.
No API, still automated: Open the app, then Find the record, then Type the values, then Save, then Check it saved. One case travels the whole chain. The AI does every step and stops at Check it saved, where a person approves — it did not save. The case ends as: Record updated.
The AI checks it saved
record updated
Show the work — every figure, with its source
- Enterprise AI adoption is most visible in information-heavy functions — IT and knowledge management, marketing and sales, customer service, finance, HR and supply chain.
- Lead enrichment, deduplication, qualification, routing and sequencing are established sales-automation use cases; finance, procurement and order processing are common because they combine structured systems with repeated document and validation work.
- Customer onboarding is one of the end-to-end processes currently highlighted by enterprise automation vendors as a candidate for agentic orchestration.
- Intelligent document processing and OCR are established ways of converting documents and scanned material into machine-readable information; computer-use technologies can bridge workflows into systems with no modern integration.
- Every step chain on this page is written from the brief's own process descriptions. They are illustrative of the shape of the work, not a quotation of any one client's process.
Start with one process.
Show us how it works today. We will find what should stay human and what can be automated.