Problems businesses live with, until someone shows them they do not have to.
Six engagements with small and medium businesses in Kuwait. Each starts from a rule the owner had to meet and a desk that could not meet it, and ends with a system that reads the Arabic, keeps the deadline and leaves the rest of the operation as it was. Where artificial intelligence did the work, the page says so. Where it did not, the page says that too.
Engagement details are anonymised and composited to protect client confidentiality.
Explore the workflow.
Choose a familiar problem. Follow what comes in, how it is handled, and what your team can use.
The notice that arrived in the inbox nobody opened
Every ministry now writes to one official inbox. Nobody owned it.
Government notices turned into tasks, each with its deadline and the sentence it came from.
Follow the steps
Illustrative workflowSelect a step to see what happens.
Start with the original notice.
A read-only mailbox connection brings the message into the workflow. The original stays available throughout.
- Source
- Official mailbox
- Content
- Arabic letter or notice
- Access
- Read-only connection
Turn the letter into proposed fields.
AI reads the sender, licence, requested action and deadline. The source sentence sits beside the extracted deadline.
- Identify
- Sender and licence
- Extract
- Action and deadline
- Trace
- Original sentence
Check the date against the letter.
The owner can compare the proposed deadline with the exact sentence. Draft replies are reviewed and sent by a person.
- Compare
- Deadline with source
- Review
- Draft Arabic reply
- Send
- A person decides
Give the notice an owner and a next step.
The task list and bilingual WhatsApp digest carry the work forward, with escalation when nothing is marked done.
- Task
- Action with deadline
- Digest
- Manager and partner
- Follow-up
- Unfinished tasks escalate
Where AI fits
AI did the reading
AI reads the Arabic and extracts proposed tasks and deadlines. Each deadline stays linked to its source sentence for a person to check.
What stays in place
The official mailbox, ministry portals and accounting spreadsheet. People still submit and send replies.
Explore the example output
- SahelWritten reply to complaint 2026/7041 (Sahel)
- Licence concerned
- Commercial licence 2024/18731
- Deadline
- 15 days from 19 Oct 2026 → 3 Nov 2026
- Kuwait MunicipalityActivity reclassification notice
- Licence concerned
- Commercial licence 2023/40912
- Deadline
- 30 days from 21 Oct 2026 → 20 Nov 2026
- Public Authority for ManpowerLabour-demand estimate requested
- Licence concerned
- Commercial licence 2022/07215
- Deadline
- 10 days from 22 Oct 2026 → 1 Nov 2026
Illustrative workflow
- with its deadline and the sentence it came from
- acknowledges from his phone
- has not been needed since
One inactive licence froze every hire
A dormant licence on the family file blocked every new work permit.
A licence register that flags what the ministry would flag, first.
Follow the steps
Illustrative workflowSelect a step to see what happens.
Bring the scattered paperwork together.
Start with the licences and scans the mandoub already has, including their linked files, activities and renewal dates.
- Source
- Licence scans
- Details
- Activity and address
- Dates
- Renewal deadlines
Check how the records fit together.
A shared register connects each licence to its file. Rules flag inactive licences, address mismatches and unlisted activities.
- Status
- Inactive licence on active file
- Address
- Compare linked records
- Activity
- Check against licence
Keep a person close to the records.
A person confirms the fields read from each scan. Ministry-visit outcomes are entered into the register by the team.
- Confirm
- Fields against the scan
- Update
- Status after ministry visits
- Acknowledge
- Renewal alerts
Make the next renewal visible.
A calendar sends Arabic and English alerts to the operator and signatory, with a record of who acknowledged each one.
- When
- Next renewal date
- Who
- Operator and signatory
- Record
- Alert acknowledgement
Where AI fits
No AI needed
The core is a register and consistency rules. Licence scans are read into fields, with each field confirmed by a person.
What stays in place
The paper file, ministry portals and mandoub's work. Status is entered after visits; this example has no portal integration.
Explore the example output
- Trading licence 2019/22604Spare parts, tyres · MOCIRenews 14 Feb 2027Active
- Workshop licence 2021/08817Vehicle maintenance · MunicipalityRenews 9 Dec 2026Alert sent · 30 days out
- Trading licence 2017/51130General trading · MOCILapsedInactive · blocks the file
Illustrative workflow
- closed rather than renewed
- before an inspector asked
- a printed list instead of a memory
The inspector asked for the Arabic invoice
English invoices, three price lists, and an inspector at the counter.
Bilingual invoices without changing the accountant's software.
Follow the steps
Illustrative workflowSelect a step to see what happens.
Start where the accountant already works.
The existing accounting export feeds an Arabic invoicing layer. The accountant keeps the same software and routine.
- Source
- Existing accounting software
- Input
- Invoice export
- Prices
- One master price list
Use AI for one focused job.
A model drafts Arabic product names using trade terminology. This is the first-pass translation step in the workflow.
- Input
- Product names
- AI task
- First-pass Arabic translation
- Status
- Awaiting owner approval
Approve the names before using them.
The owner reviews the translations once. The approved names can then be reused by the invoice template.
- Check
- Product names and terminology
- Decision
- Owner approval
- Reuse
- Approved Arabic names
Keep the invoice and prices together.
Templates produce and archive numbered bilingual invoices. The same price list feeds shelf labels, the catalogue and invoices.
- Invoice
- Numbered and bilingual
- Pricing
- Shelf, catalogue and invoice
- Record
- Archived invoice
Where AI fits
AI did one job
AI drafts Arabic product-name translations for owner approval. Invoice generation and the shared price list use templates and integration.
What stays in place
The accounting software, accountant's workflow, bank and customer-facing WhatsApp catalogue.
Explore the example output
- Item 1 · Cement, 50 kg bagqty 20 · KWD 31.000
- Item 2 · Steel rebar, 12 mmqty 40 · KWD 58.400
- Item 3 · Tile adhesive, 25 kgqty 10 · KWD 17.500
Illustrative workflow
- a matching shelf price
- did not change a single habit
- changes a price in one place
The order lived in WhatsApp, and the invoice did not exist
Four voice notes, a payment screenshot, and a law that now expects an invoice.
Voice notes and dialect chat turned into a numbered Arabic invoice.
Follow the steps
Illustrative workflowSelect a step to see what happens.
Begin with the conversation.
The workflow uses the WhatsApp thread, including text and transcribed voice notes, where the order already lives.
- Channel
- WhatsApp Business
- Content
- Text and voice transcripts
- Language
- Informal Arabic dialect
Give the conversation some structure.
AI extracts the items, quantities, price, delivery slot and payment status into a proposed order for the seller.
- Products
- Items and quantities
- Delivery
- Requested slot
- Payment
- Price and status
Let the seller confirm the order.
The structured order is presented for the seller to confirm. The invoice is generated from that confirmed order.
- Review
- Proposed order details
- Confirm
- Seller decision
- Next
- Generate the invoice
Leave a record beyond the chat.
A numbered bilingual invoice goes back into the conversation, alongside an order record with a matched payment reference.
- Invoice
- Sent back into chat
- Payment
- Reference matched to order
- Record
- Export and complaint tag
Where AI fits
AI did the reading
AI turns informal Arabic chat and transcribed voice notes into proposed order fields. The seller confirms the order before an invoice is produced.
What stays in place
Instagram, WhatsApp, the bank app and the driver. Customers keep using the shop they already know.
Explore the example output
- 0:42Transcribed · Kuwaiti dialect
- 0:18
- Payment link screenshot
- Order 0418 confirmedItems3 traysDelivery slotThu · 18:00 to 20:00PaymentHalf paid · reference matched
- Invoice 0418Numbered · Arabic · sent to the chat
Illustrative workflow
- an Arabic invoice before the driver leaves
- Payments are matched to orders
- from one phone
The tender wanted the same twelve documents again, in Arabic, by Thursday
A hundred pages of Arabic conditions, read on a phone at midnight.
Arabic tender extraction with page references, and a bid pack on demand.
Follow the steps
Illustrative workflowSelect a step to see what happens.
Start with the tender document.
The Arabic PDF provides the conditions. Supporting company documents are maintained separately in a versioned dossier.
- Source
- Arabic tender PDF
- Supporting files
- Company dossier
- Versions
- Dated documents
Make the requirements easier to find.
AI extracts requirements, dates, amounts and attachments. Each item points back to the page where it was found.
- Extract
- Requirements and dates
- Include
- Amounts and attachments
- Trace
- Source page reference
Verify the line in the original PDF.
The owner uses the page reference to check each extracted item against the document. The extraction is presented as source information.
- Open
- Referenced PDF page
- Compare
- Extracted item and source
- Pricing
- Estimator's spreadsheet
Assemble the documents for submission.
The maintained dossier produces a bid pack with document versions and expiry warnings. A person uploads the pack to the portal.
- Documents
- Version date on each file
- Check
- Expiry before submission
- Submit
- Person uploads to portal
Where AI fits
AI did the reading
AI extracts requirements from Arabic tender documents with page references. The company dossier, expiry checks and bid pack use ordinary workflow software.
What stays in place
The estimator's pricing spreadsheet, accountant and tender portals. A person still uploads the bid.
Explore the example output
- Bid bond2% of bid value, valid 90 daysPage 7
- Classification requiredCategory 4 or higher, electromechanicalPage 12
- Submission deadlineSun 14 Dec 2026, 13:00Page 31
Illustrative workflow
- reads a checklist first and the tender second
- Expiry warnings arrive before the bid window
- in an afternoon
The ratio changed every time a permit was issued
Sixty permits, one Excel sheet, and a national-employment ratio nobody could see.
A workforce register that shows the Kuwaiti percentage before the letter arrives.
Follow the steps
Illustrative workflowSelect a step to see what happens.
Use the records already in Excel.
The existing workforce sheet becomes the starting point for a register covering nationality, roles, permits and expiry dates.
- Source
- Existing Excel sheet
- People
- Nationality and actual role
- Dates
- Permit and residency expiry
Turn the records into a clear position.
Calculations show the workforce ratio against the sector percentage. Rules collect upcoming renewals and fees by month.
- Calculate
- Current workforce ratio
- Compare
- Sector percentage
- Collect
- Renewals and fees
See the effect before the next hire.
The owner can inspect a hiring scenario alongside the current position. The register supports planning using the real workforce data.
- Inspect
- Current workforce position
- Explore
- Effect of the next hire
- Decide
- Owner reviews the scenario
Make the next months easier to plan.
A renewal calendar and draft labour-demand estimate bring the workforce information into day-to-day planning.
- Calendar
- Monthly renewals and fees
- Estimate
- Draft labour demand
- View
- Current workforce position
Where AI fits
No AI needed
No AI is needed. Structured data, calculations and rules provide the register, hiring what-if and renewal calendar.
What stays in place
Payroll, the mandoub, PAM and Ashal portals, and PIFSS. The register reports the actual workforce position.
Explore the example output
Illustrative workflow
- the ratio and next quarter's renewal fees on one screen
- checked against reserved titles
- drafted from the register instead of from memory
Something here looks familiar?
Tell us where the work gets stuck. We can start with one workflow and work out what would make it simpler.