Implementation experience, first-hand: what works, what doesn't, and why.
We estimated a project at 49 hours and spent 129. The estimation error is almost never about hours — it is about undeclared scope creep and anchoring.
In 2016 we automated proposal generation and got results companies now expect from AI. Here is what actually drove that outcome — and it wasn't the writing.
Work that never reaches your system of record doesn't exist for the business. What error data on spreadsheets and data quality reveal, and how to fix it.
A month after launch, if you can't pull up numbers on completed scenarios and failures, you launched a demo. Here are the three numbers that matter, and why.
Digitization projects rarely fail on architecture — they fail in the final ten percent of the path to the user. A breakdown of adoption mechanics and how to measure it.
If no process rule changed, a project reports interface coverage instead of time saved. The warning signs to check for, backed by healthcare workload data.
Some tasks companies buy language models for are better solved with rules — cheaper and more predictable. Here's where that line falls and who owns the error.
A statement of work manages expectations, not the build itself. The out-of-scope list beats a solution description, backed by real requirements volatility data.
Calculating by hours saved almost always overstates the effect. The proper method: count eliminated operations and the failure rate, step by step.
Resistance concentrates wherever a system takes control away from people, not workload. What change management should read from it, and why the term misleads.
An estimation stage needs a defined input, an owner, a documented output, and a re-estimation trigger. The data show how question format changes outcomes.
A deviation that reaches a weekly report is already too late to fix. How events differ from metrics, and how to move a process onto event-driven management.
An aggregate effect number can't be attributed to one change. How to decompose a result into assembly, delivery and events — and why skipping this makes it unrepeatable.
The costliest process defects skip every internal report — only the client sees them, and stays silent instead of complaining. How to assign an observer.
Rolling out ERP or CRM on top of an unfixed process locks the mess into the configuration. A breakdown of three rollouts: what to settle before the project starts.
A pilot has no process owner, no system of record integration, and no deadline to sunset the old way. That design, not the model, decides the outcome.
A results range this wide isn't measurement error — it depends on whether the automated step was the line's bottleneck. The mechanics and global robotics data.
In a team split across sites, it isn't communication that breaks — it's the shared definition of done. Data and analysis on distributed team readiness.
What can't be recalled in full doesn't get followed. Why process length is a measurable predictor of compliance, and how to shorten it without losing content.
A monitoring loop's value comes not from the check but from routing results to the decision maker. A breakdown of the loop and payment-discipline data.
Banning an incomplete result costs less than downstream inspection by an order of magnitude. The math, the data on record quality, and where the method breaks.
The spread between bids isn't about job complexity — it's about how much uncertainty a vendor accepts. A look at the data behind estimate accuracy by class.
Active users is a metric that grows from curiosity alone. Real adoption is measured by the completed scenario rate — here's how to calculate it in practice.
Adoption spreads through the people colleagues turn to informally, not a training schedule. Here's the mechanics, plus data on opinion leaders' measured impact.
An exit is planned around the sequence of handing over processes, not selling assets. A breakdown of the mechanics from two of my own exits in eleven months.
A foreign partner in China pays for fitting the structure, not for market access. The mechanics: negative list, data transfer thresholds and investment statistics.
Exit terms for a China joint venture get negotiated at deal structuring or never at all. A practical look at arbitration, FX procedure, and deal timelines.
A business incubator in an emerging market delivers access to decision makers, not funding. A delegation leader's account, plus current investment data.
A free zone competes on registration speed and licence cost, not on when your business starts operating. Procedures, tax rules and official statistics reviewed.
Rate savings pay off only when you have your own requirements owner on your side. A breakdown of the mechanics and the measured cost of cross-site coordination.
The first post: why I'm starting a blog, and what it will cover.
