AI can build almost anything now. Knowing what's worth building is the hard part.
For 20 years I worked as the client, the one who decides what gets built. Today I build the solutions myself. I find the constraint, tie it to a number, and move it.
Constraint → KPI → BuildDiagnosis, not just toolsBuilds + drives adoption
The shift
Building is cheap now. Knowing what's worth building isn't.
Tools change every year. What doesn't expire is judgment: naming the real problem, moving the right number, and getting the organization to actually use it.
“Most AI help is a pharmacist. It hands you what you asked for. I'm the doctor: I find what's actually wrong, and fix it.”
Diagnose → prescribe → build → prove
The method
Constraint → KPI → Build
01
Constraint
Find the one constraint that, once fixed, makes the business faster, earns more, or stops a bleed.
02
KPI
Name the number we will move before building. It becomes the project's North Star, and the proof it worked.
03
Build
Build the solution. I build it myself, so it doesn't stop at a pretty deck. Then I drive adoption.
Anyone can say “I built an AI agent”. Almost nobody can say “I moved THIS number by THIS much, for THIS business”.
The value menu
Pick the number that hurts.
What value AI actually creates, by industry. Pick an industry. Don't present every number, ask which one hurts, and run Constraint → KPI → Build on that one.
AI in energy: fewer outages, lower losses, faster steering and compliance.
SAIDI
Reliability and customer experience.
Predictive maintenance
Grid losses
Direct effect on the bottom line.
ML loss detection
Customer NPS
Better relationship, less friction.
AI customer service
Fossil-free share
Climate targets and strategic shift.
Forecast + dispatch
EBITDA margin
Overall company profitability.
OPEX automation
NIS2 response time
Regulatory and operational risk.
Incident triage
Time-to-close
Faster steering and compliance.
Automated reporting
AI in manufacturing: raise throughput and yield while cutting downtime, scrap, and inspection.
OEE
Availability x performance x quality.
Predictive maintenance
Unplanned downtime
Unexpected stoppages kill output.
Failure prediction
First pass yield
Share good without rework.
Process optimization
Scrap rate
Material lost to defects.
Vision defect detection
Throughput
Good units per period vs plan.
Bottleneck detection
On-time delivery
Orders shipped by need date.
Demand forecasting
AI in construction & real estate: catch schedule and design risk early, cut rework, speed documents.
RFI turnaround
Days to resolve information requests.
NLP retrieval + drafts
Schedule slippage
Days late vs baseline milestones.
Schedule-risk agents
Rework cost
Spend redoing defective work.
Deviation vision detection
Change-order cycle
Speed to price and approve changes.
Document extraction
Energy intensity
Operational energy per m².
Predictive HVAC control
Occupancy rate
Leased vs empty, drives NOI.
Demand forecasting
AI in retail: personalizes, prices, and forecasts for more revenue per visitor.
Conversion rate
Share of visitors who buy.
Personalized recommendations
Cart abandonment
Around 70% of carts abandoned.
Proactive chat + offers
Average order value
Revenue per order.
AI cross-sell
Customer lifetime value
Long-run value per customer.
Churn prediction + retention
Inventory turnover
Stock efficiency vs availability.
Demand forecasting
Return rate
Returns erode margin and logistics.
Fit/size prediction
AI in logistics: forecasts demand and predicts disruptions for on-time delivery with leaner stock.
OTIF
Orders complete and on time.
Route optimization + ETA
Forecast accuracy
Demand prediction vs actual.
ML demand forecasting
Perfect order rate
Error-free end-to-end fulfillment.
Anomaly detection
Days of inventory
Capital tied in stock.
Demand-driven replenishment
Fill rate
Demand met from stock on hand.
Forecast-aligned placement
Freight cost/unit
Efficiency of logistics spend.
Load consolidation
AI in healthcare: predicts risk and automates ops for better outcomes and throughput.
Patient no-show
Missed appointments waste capacity.
Risk prediction + reminders
Readmission rate
Costly, penalized quality metric.
Risk stratification on EHR
Length of stay
Bed-days per admission.
LOS prediction + discharge
Diagnostic accuracy
Correct, early detection.
Imaging models
Time-to-diagnosis
Speed to catch deterioration.
Real-time early warning
Drug discovery cycle
Speed from target to candidate.
Generative molecule design
AI in finance & insurance: detects fraud, prices risk, automates claims and onboarding at scale.
Fraud detection rate
Share of fraud caught.
ML anomaly detection
False-positive rate
Legit cases wrongly flagged.
Precision scoring
Loss ratio
Claims vs premiums, core profit.
AI risk pricing
Claims cycle time
Days from FNOL to settlement.
Automated claims triage
KYC/onboarding time
Days to verify an account.
Document + ID automation
Compliance accuracy
Error rate in AML checks.
Automated screening
AI in SaaS: retains users, deflects support, personalizes onboarding for more revenue.
Churn rate
Share of customers lost per period.
Predictive early warning
Net revenue retention
Expansion vs contraction.
Health scoring + upsell
Activation rate
New users reach value fast.
AI-guided onboarding
Ticket deflection
Cases resolved without a human.
Agentic support
CAC
Cost to acquire a customer.
AI lead scoring
Time-to-resolution
Speed to close support cases.
Automated triage
AI at the agency: cuts production hours, speeds turnaround, lifts win rate and margin.
Billable utilization
Share of hours on billable work.
Automated admin offload
Campaign turnaround
Speed from brief to launch.
AI drafting + generation
Cost per asset
Production cost per deliverable.
Generative content
Pitch win rate
Share of proposals won.
AI research + first drafts
Hours per deliverable
Labor cost per output unit.
AI drafts + editing
Client retention
Accounts kept year-over-year.
Faster delivery + insight
Proof, not CV
The question isn't how long. It's what you built.
AIOS
I built my own AI operating system: email, calendar, CRM, contact and job pipelines, all agent-driven. Proof I ship, not just advise.
Qvantum
Automated manual reporting with Claude Code. Constraint gone, reporting time down. Constraint → KPI → Build for real.
Telling Technology
Building in public. Content and demos that make AI legible to non-technical audiences. A public track record most consultants lack.
Shipped AI systems, real companies, anonymized demos. See all cases →
The channel's own systems, built in the open. See all builds →
Same method, two entry points. Both start with one call, no sales pitch.
01
Direct with your company
We take a number that hurts, find the constraint, and I build the solution and drive adoption. Constraint → KPI → Build.
02
As the operator behind your agency
You keep the client, I'm the AI operator behind the account. A new, high-margin service to bring to accounts you already have. White-label or side by side.
