Blog
From decision to deployment
Practical writing on AI, data, and automation, from the people who build it.
The Report Looks Fine. Why Does the Company Find the Problem Only After the Fact?
Auditors found 133,863 delayed mailpieces. The system showed 37,392. How to check whether the information a decision needed was available in time, before you order another report.
The Team Works Fast. Why Is the Customer Still Waiting?
Every department does its job quickly, and the customer still waits for days. The gaps between activities are usually to blame. How to find the one gap that, once closed, actually speeds up delivery.
Every Department Did Its Job. Why Isn't the Case Moving?
The sale is closed, the CRM data is complete, and the implementation team doesn't have the case on its list. What triggers the next step, who picks it up and how you can tell. A step handover card in six points.
Automation Works Until an Unusual Case Shows Up
The order is complete, and still nobody can say which of two non-interchangeable products to pick. What the automation must not guess, what can carry on and where the basis for a decision comes from. An exception card in six points.
The AI Works. When Can You Stop Doing the Job by Hand?
The AI reads delivery confirmations correctly, yet purchasing still checks every email. What has to be proven before the manual check is retired, how you'll notice quality slipping, and who takes over when the automation has to be scaled back.
Don't Automate Your Most Important Process First. Pick One the Company Can Learn On
Your most important process is rarely the best first automation project. How to pick one with the right balance of impact, cost and risk.
Does Automation Pay Off? How to Build the Business Case Before You Invest in AI
Recovered hours aren't money yet. How to compare two futures of the same process, calculate the real benefit, and check whether the project clears your decision threshold.
Do You Really Need an LLM? Picking the Right Technology for Process Automation
An LLM isn't simpler or harder than rules by definition. Using receipt analysis as an example, we show how to break a process into problems and match the technology to each one.
From Excel to AI. Six Ways Companies Work With Data
Missing history, unused data, hand-merged exports, a warehouse, BI, forecasts and AI. Six ways of working with data, and a simple test for which change would actually improve decisions.
You Have Data and Reports. Why Don't They Lead to Better Decisions?
Access, preparation, shared meaning, use. Four places where data stops helping a decision, and a small experiment that checks whether a change actually works.
The AI System "Works". But How Often Is It Right?
Ten correct answers are the result of ten attempts, not a guarantee. How to define a correct result, what 95% accuracy can hide, and what to ask before accepting an LLM-based system.
99% AI Accuracy Can Be Too Little. Or Too Much.
The team wants budget to push AI accuracy from 95% to 99%. Before you pay for extra percentage points, find out what the number measures, which mistakes will remain and what they cost.
The AI Passed 1,000 Tests. Is It Ready to Go Live?
The report shows 1,000 runs and zero errors. Before you approve the launch, find out where the cases came from, who decided what counts as correct and how many tests covered what matters most.
What Does It Really Cost to Run AI? What's Easy to Miss Before You Build
The quote for building the app and the forecast for model usage both look reasonable. Before you approve the budget, check the seven places where an AI system's costs arise, and what will push the bill up as you scale.
What Can AI Do Without Human Approval? How to Set the Limits of Automation
A customer reports a late delivery and a damaged product. The AI can refund the delivery fee, but is it allowed to close the case? How to separate the right to perform one action from the right to settle the whole request.
Who Is Responsible for AI Errors? How to Monitor the System and Respond to Problems
The application is up, the integrations respond, and the AI is closing complaints it never fully handled. How to watch the system's decisions, review samples of real cases and agree who acts when something goes wrong.
How to Choose an AI Implementation Partner. What to Check Before You Sign
A vendor's AI system turns a sample inquiry into a finished quote in seconds. Before the conversation moves on to price and timeline, check whether anyone has asked why the process takes so long today and what really needs to change.
Prompt Caching: How to Cut LLM Costs by up to 90%
Prompt caching can cut LLM input costs by 90% and time to first token by up to 80%. How prefix caching and semantic caching work, and when they pay off.
An ERP Is Not the Same as Process Automation
An ERP can work perfectly well while people still move data by hand between email, PDFs, Excel and other systems. How to find the places where a process still runs on people.
7 Manual Processes in Manufacturing Worth Checking for Automation
Customer orders, supplier confirmations, invoices, production reporting, quality documentation, shipping and reporting. Seven places in a manufacturing company where repetitive work worth automating tends to hide.