Industrial AI practice  ·  Gulf  ·  Arabic and English Reply within one working day
One Industry AI
Menu

Case studies

35 records, organised by the problem

Delivered for

Work by the people in this practice.

  • Saudi Aramco, Saudi Arabia
  • NEOM, Saudi Arabia
  • Red Sea Global Saudi Arabia
  • Saudi Data and AI Authority Saudi Arabia
  • Zakat, Tax and Customs Authority Saudi Arabia
  • Ministry of Interior Saudi Arabia
  • Ministry of Commerce, Saudi Arabia
  • Ministry of Media, Saudi Arabia
  • Ministry of Transportation Saudi Arabia
  • Ministry of Municipal, Rural Affairs and Housing Saudi Arabia
  • Royal Court Saudi Arabia
  • Saudi Industrial Development Fund Saudi Arabia
  • Saudi Railway Company, Saudi Arabia
  • King Fahd Causeway Authority Saudi Arabia
  • King Khalid University Saudi Arabia
  • Princess Nourah bint Abdulrahman University, Saudi Arabia
  • Riyadh Municipality Saudi Arabia
  • Jeddah Municipality Saudi Arabia
  • Eastern Province Municipality, Saudi Arabia
  • Tabuk Municipality Saudi Arabia
  • National Center for Government Resources Systems Saudi Arabia
  • Tahakom Saudi Arabia
  • Yale University, United States
01

The same four problems, in different uniforms

A telecom invoice problem and an industrial one share the same structure, so evidence from one can apply to the other.

Behind this practice sit 85+ documented engagements. They are the delivery record of the engineers and delivery leads who make up this practice, and they reduce to four problem shapes. That is what makes proof from one industry usable in another: a telecom's invoice problem is an industrial invoice problem, because it has the same structure.

Of those, 35 are published here. The rest is delivered work in sectors with no industrial analogue - media content production, religious-services content, consumer recognition, public-safety biometrics - or work held for private circulation.

A human reads it and types it in again

Invoices, contracts, scanned paper, submittals

6 documented deliveries

The answer exists and nobody can find it

Procedures, standards, specifications

13 documented deliveries

Too much arrives to check it all, so you sample

Inspections, calls, feedback, camera hours

7 documented deliveries

The same judgment, made differently every time

Risk scoring, prioritisation, approvals

9 documented deliveries

Figures are from the practice's own delivery records for the engagement named, measured against the process that preceded it.

02

How proof from one industry becomes proof for yours

Four problem shapes, and a telecom record reads as an industrial one because the shape is the same.

We built it there. The problem there had this shape. You have the same shape in your own function. Reading a contract to verify a figure is the same engineering problem wherever the contract sits.

Where it was built The shape of the problem Where the same shape sits

Healthcare, Telecommunications, Cross-industry, Financial Services, Professional Services

A human reads it and types it in again 6 deliveries

scanned archives, finance & procurement, bilingual paperwork, finance & reporting, HR

Transportation, Government, Energy, Logistics, Telecommunications

The answer exists and nobody can find it 13 deliveries

Ports & terminals, confidential archives, customer service, Oilfield services / petrochem, Logistics & 3PL

Manufacturing, Energy, Logistics, Insurance, Telecommunications

Too much arrives to check it all, so you sample 7 deliveries

Manufacturing / inspection & QC, Oilfield services / utilities, operations reporting, Ports & industrial zones, customer service & QA

Healthcare, Retail, Financial Services, Government, Energy

The same judgment, made differently every time 9 deliveries

capacity planning, Industrial distribution, shared services, risk & audit, Utilities / O&M providers

03.1

A human reads it and types it in again

Invoices, contracts, scanned paper, submittals

The document arrives, somebody reads it, and somebody types the same figures into the system that will be audited. The typing is not the problem. The reading is.

Partner settlements verified against contract terms

100%

from ~15% sampled

Handling time per record

~2.5 minutes

from 18 minutes

Average validation time per invoice

45 seconds

from 12 minutes

A Gulf public health ministry

Client engagement

Document extraction · AI Document Intelligence

Teams needed to convert paper and scanned records into structured data. Health records on paper cannot be queried, audited, or linked to anything - the information exists but is unavailable to every downstream system that needs it.

Where the same shape sits: Any - scanned archives. Built for Healthcare / Government, and the problem there had the shape above.

Measure Before After
Handling time per record 18 minutes ~2.5 minutes
Extraction accuracy no prior measure 96% clean digital forms / 84% poor scans / 61% handwritten entries
Fields routed to human review with their source region highlighted no prior measure 14%
Before 18 minutes After ~2.5 minutes

Read the full record →

Also in this shape, 5 records

03.2

The answer exists and nobody can find it

Procedures, standards, specifications

The document is on a drive somewhere and it is correct. The person who needs it does not know the words it uses for their situation, so they ask a colleague instead.

Visitors completing a service action in the same session as their question

68%

from 27%

Correct statutory provision in the top three results

93%

from 62%

Everyday-language queries returning a usable provision

89%

from 34%

A Gulf border-crossing authority

Client engagement

Conversational AI / digital customer service · Service & Conversation AI

Travelers needed immediate answers on crossing times, vehicle-insurance validity, and digital services, while service teams needed a scalable alternative to repetitive contacts.

Where the same shape sits: Ports & terminals. Built for Transportation / Border Operations, and the problem there had the shape above.

Measure Before After
Response time for common traveller questions † down 55%
Routine demand handled without entering the contact-center queue no prior measure 35%
Digital self-service completion † up 28%

† Measured against the prior level of the same measure.

35%

35%

Routine demand handled without entering the contact-center queue

The ring is the whole of that measure. The grey arc is the other 65%.

Read the full record →

Also in this shape, 12 records

03.3

Too much arrives to check it all, so you sample

Inspections, calls, feedback, camera hours

A percentage gets reviewed and the rest is assumed to look like it. Sometimes it does. The interesting question is what the unreviewed remainder has been hiding.

Calls receiving a quality review

100%

from ~2% sampled

Word error rate on Gulf-dialect calls

12.7%

from 31%

Lag from a review appearing to a grouped, scored issue reaching the product team

under 24 hours

from ~3 weeks

A beverage bottling operation, three high-speed lines

Single client engagement

Computer Vision (quality inspection) · Inspection & Safety Vision AI

End-of-line quality relied on manual spot checks at line speeds above 20,000 units/hour, so human sampling caught only a fraction of defects. Defective pallets occasionally reached distributors, and defect spikes were only discovered hours later through downstream complaints.

Where the same shape sits: Manufacturing / inspection & QC. Built for Manufacturing, and the problem there had the shape above.

Measure Before After
Defect escape rate to distribution † down 84%
Detection latency hours under 2 minutes, enabling same-shift correction
Customer returns tied to quality † down 61% over the following quarter

† Measured against the prior level of the same measure.

Prior level 0 100 200 down 84%

Measured against the client's own prior process, indexed to 100. The source publishes the size of the movement, not the absolute figure it moved from.

Read the full record →

Every vendor wanted to sell us a suite catching everything - but three defects were 90% of our real cost, and solving those first is what made the business case obvious.

Plant Quality Manager

Also in this shape, 6 records

03.4

The same judgment, made differently every time

Risk scoring, prioritisation, approvals

Two people look at the same case and route it two different ways, both defensibly. Nobody is wrong, and the outcome still depends on who was on shift.

Network no-show rate

9.1%

from 14.6%

Misrouting

1.4%

from 9%

Forecast accuracy (WAPE)

19% on piloted categories

from ~42% error

A five-site outpatient hospital network

Single client engagement

Core data science / forecasting & decision support · Forecasting & Planning AI

Clinics ran a flat 8% no-show assumption and overbooked uniformly, while real no-show rates swung from 4% to 22% by specialty. The result was patients waiting up to 90 minutes in some clinics while specialists sat idle in others.

Where the same shape sits: Any - capacity planning. Built for Healthcare, and the problem there had the shape above.

Measure Before After
Network no-show rate 14.6% 9.1%
Patient wait time in piloted specialties † down 31%
Specialist chair utilisation † +11 points, recovering ~240 slots per month per site

† Measured against the prior level of the same measure.

14.6% 9.1% 0% 100%

Read the full record →

The win wasn't a fancier algorithm - it was that we stopped treating every specialty the same, and our schedulers trusted the tool because they could see the reasoning.

Director of Ambulatory Operations

Also in this shape, 8 records

04

Every record, in one table

Each record appears with its strongest figure, grouped by problem shape.

Record Shape Sector Strongest result
A Gulf public health ministry A human reads it and types it in again Healthcare / Government 18 minutes → ~2.5 minutes Handling time per record
A Gulf telecom operator's finance function - Contract intelligence A human reads it and types it in again Telecommunications / Finance ~15% sampled → 100% Partner settlements verified against contract terms
A Gulf telecom operator's finance function - Multimodal invoice validation A human reads it and types it in again Telecommunications / Finance 12 minutes → 45 seconds Average validation time per invoice
Pulling sustainability figures out of reports A human reads it and types it in again Financial Services / Sustainability down 65% ESG report-review time
Reading Arabic documents field by field A human reads it and types it in again Cross-industry / Enterprise down 60% Manual document-review time
Screening applications against a real vacancy A human reads it and types it in again Professional Services / HR ~9 hours → 25 minutes Time to a ranked shortlist from a 300-CV posting
A Gulf border-crossing authority The answer exists and nobody can find it Transportation / Border Operations 35% Routine demand handled without entering the contact-center queue
A Gulf judicial body The answer exists and nobody can find it Government / Judiciary 21 minutes → ~3 minutes Handling time per document
A Gulf national cybersecurity body The answer exists and nobody can find it Government / Cybersecurity 27% → 68% Visitors completing a service action in the same session as their question
A Gulf petrochemicals operator The answer exists and nobody can find it Energy / Petrochemicals 2.9 minutes → 31 seconds Average time to a location, contact or facility answer
A Gulf postal operator The answer exists and nobody can find it Logistics / Postal Services 3.8 minutes → 25 seconds Average time to a shipment status answer
A Gulf social-insurance authority The answer exists and nobody can find it Government / Social Insurance 62% → 93% Correct statutory provision in the top three results
A Gulf tax authority - Arabic NLP, agent assistance The answer exists and nobody can find it Government / Tax and Customs down 65% Agent search time for tax guidance
A Gulf tax authority - Enterprise generative AI, analytics platform The answer exists and nobody can find it Government / Tax and Customs down 55% Document-processing time across supported workflows
A Gulf telecom operator - Conversational voice AI The answer exists and nobody can find it Telecommunications 6.2 minutes → 94 seconds Average time to complete an authenticated account or billing enquiry
A Gulf telecom operator's HR function The answer exists and nobody can find it Telecommunications / HR 74% Recurring HR questions answered in-app without a ticket
A Gulf tourism and real-estate developer The answer exists and nobody can find it Tourism / Real Estate Development down 60% HR response time
Answering a question across bilingual documents The answer exists and nobody can find it Cross-industry / Enterprise down 70% Information-retrieval time
One workspace across a split knowledge estate The answer exists and nobody can find it Enterprise Technology ~12 minutes → under 90 seconds Average knowledge-discovery time for indexed enterprise content
A Gulf energy operator Too much arrives to check it all, so you sample Energy / Industrial Security up 45% Effective patrol coverage
A Gulf fuel-retail operator Too much arrives to check it all, so you sample Energy / Retail ~3 weeks → under 24 hours Lag from a review appearing to a grouped, scored issue reaching the product team
A Gulf industrial-zone operator Too much arrives to check it all, so you sample Logistics / Industrial Zones ~15 minutes → under 30 seconds Time to locate a specific event in the footage
A Gulf insurer Too much arrives to check it all, so you sample Insurance 31% → 12.7% Word error rate on Gulf-dialect calls
A Gulf telecom operator - Call intelligence Too much arrives to check it all, so you sample Telecommunications ~2% sampled → 100% Calls receiving a quality review
A beverage bottling operation, three high-speed lines Too much arrives to check it all, so you sample Manufacturing down 84% Defect escape rate to distribution
Finding where a process actually stalls Too much arrives to check it all, so you sample Cross-industry / Enterprise Operations down 30% Cycle time in the targeted processes
A five-site outpatient hospital network The same judgment, made differently every time Healthcare 14.6% → 9.1% Network no-show rate
A mid-market online fashion and homeware retailer, ~14,000 SKUs The same judgment, made differently every time Retail / E-Commerce down 27% End-of-season clearance inventory
A mid-sized regional property & casualty insurer, ~1.2M active policies The same judgment, made differently every time Financial Services / Insurance 9% → 1.4% Misrouting
A national economic-development agency running an SME grant program The same judgment, made differently every time Government / Public Sector down 68% Officer time spent on eligibility screening
Allocating stock across sites that compete for it The same judgment, made differently every time Retail / Supply Chain down 20% Inventory-allocation cost
Catching transactions that do not fit the pattern The same judgment, made differently every time Financial Services / Payments up 25% High-risk transaction detection
Forecasting demand, and what to stock against it The same judgment, made differently every time Retail / Manufacturing down 25% Forecast error
Predicting which asset fails next The same judgment, made differently every time Energy / Industrial Operations down 30% Unplanned downtime
Stopping an assistant answering what it should not The same judgment, made differently every time Cross-industry / Enterprise down 45% Unsafe or off-policy responses

All 35 records as one document (PDF)

Start here

Which of the four is yours?

Tell us the documents and the monthly volume and we will send the two closest records, with the figures behind each and what it would take to repeat them on your material.

Book the audit All 35 records as one document (PDF)

You get a reply within one working day, from the engineer who would do the work.