Orders, data, systems and exceptions.
Four connected areas where repetitive work builds up in a distribution business. Each starts with understanding the workflow. Not every business needs all four, and not every workflow should be automated.
[01]Discipline
Sales & order operations. Less re-keying between inbox and ERP.
Sales offices spend hours keying emailed orders into the ERP. Purchase orders are our first focus and the use case we're validating now. RFQ routing, draft quotations under approved rules and order acknowledgements are possible extensions, once the core workflow is proven.
You probably need this if
- Customer POs arrive as PDFs and spreadsheets and are keyed in by hand.
- RFQs are forwarded around until someone picks them up.
- Order queries pull people away from processing orders.
How we approach it
- 01Map how orders arrive and who handles each step
- 02Define the fields, checks and approvals that matter
- 03Build intake and review in a test environment first
- 04Hand over drafts or exports, never unapproved orders
Where it applies
- Purchase-order intake and processing
- RFQ classification and routing, later
- Draft acknowledgements, staff-approved
- Cleaner sales-admin handoffs
Try the model / Sales & order operations
Purchase-order intake
A synthetic demonstration of our first use case. Each step has a handler, a trigger and a definition of done.
Step 01 / 06
PO received
- Handled by
- Workflow
- Starts when
- A PDF or Excel PO reaches the authorised sales inbox
- Done when
- Sender, attachment type and a unique document reference are recorded
[02]Discipline
Document & data workflows. Inconsistent documents, consistent data.
Customers and suppliers send information in their own formats. We turn supported documents into structured, checked data. Fixed rules come first; AI-assisted extraction is used only where layouts vary, and it never invents a missing value.
You probably need this if
- Every customer's PO looks different.
- Customer part numbers have to be matched to your SKUs by hand.
- Missing or duplicated details are found too late.
How we approach it
- 01Collect representative, anonymised samples
- 02Prefer fixed rules; use AI only where it adds value
- 03Check required fields, codes, units and duplicates
- 04Send anything uncertain to a person, not a guess
Where it applies
- Extraction from PDFs and spreadsheets
- SKU and product-code matching
- Missing-field and duplicate checks
- Approved structured exports
Try the model / Document & data workflows
Build a rule
Choose a trigger and an action, and decide whether a person approves it first. A front-end demonstration with synthetic data.
When a customer PO arrives, then extract it into a structured preview, once a person has approved it.
[03]Discipline
Systems & integrations. Connections that are verified, not assumed.
Validated information should be entered once. Every ERP and OMS is different, so we check the edition, APIs or import routes, permissions and test access before anything is built. Where a safe connection isn't possible, an approved export may still help.
You probably need this if
- The same details are typed into two or three systems.
- Exports and imports are run by hand, then checked by hand.
- Nobody is sure what the ERP can accept from outside.
How we approach it
- 01Audit the systems involved and how they're really used
- 02Verify APIs, permissions and a sandbox or test company
- 03Connect in test first, with failures handled explicitly
- 04Document what connects, and what deliberately doesn't
Where it applies
- Email-to-workflow connections
- ERP or OMS integration, where feasible
- Structured imports and exports
- Fewer repeated entries between tools
Try the model / Systems & integrations
Where does the information go?
Choose a part of an illustrative order workflow to see what it holds and exchanges. What connects in practice depends on your systems and access.
- Holds
- Customer emails, POs and RFQs as they arrive
- Passes on
- New attachments into intake, with the original message kept for reference
- Receives
- Requests for missing details, sent only once a person approves them
[04]Discipline
Exceptions & approvals. Monitored, recoverable and owned.
Automation shouldn't remove people from decisions that matter. We design how exceptions reach the right person, which steps wait for approval, and how every run is logged and monitored, with a manual fallback when something fails.
You probably need this if
- Exceptions sit in a shared inbox until someone notices.
- Nobody can see whether a process ran, or where it stopped.
- When something breaks, there's no clear way back.
How we approach it
- 01Decide which outcomes need a person, and who
- 02Route each exception with the reason it was raised
- 03Log every run, check, correction and approval
- 04Monitor failures and keep a manual fallback
Where it applies
- Exception routing and discrepancy alerts
- Human approval steps
- Audit logging and status visibility
- Error handling and manual fallback
Try the model / Exceptions & approvals
Who owns the exception?
Switch between exceptions left in a shared inbox and exceptions routed to a named owner. An illustrative scenario.
- Unknown customer product codeShared inbox
- Price differs from the price listShared inbox
- Required field missing on the POShared inbox
- Two possible SKUs for one lineShared inbox
- Possible duplicate purchase orderShared inbox
- Order above the approval limitShared inbox
[05]Where to start
Not sure where to start? Start with the bottleneck.
- We keep re-entering the same order data.Start with01 Sales & order operations
- Every customer's documents look different.Start with02 Document & data workflows
- Our systems don't share what we need.Start with03 Systems & integrations
- Exceptions get stuck between people.Start with04 Exceptions & approvals
[09]Next step
Let's find the work worth automating.
Start with a conversation about the repetitive work in your operation: where it sits, what it costs and what your systems allow. You'll leave with a clearer view of what's worth doing, whether or not we work together.