Notes on AI and operations.
Grounded, practical writing from the work itself: what actually helps a small or mid sized business run better, and where AI earns its place once the data and processes underneath are right. No hype.
LatestWhat a year of putting AI into SMB operations actually taught us
A grounded look back at what worked, what stalled, and what surprised us after a year of building AI into real business workflows for small and mid sized companies.
- ·2 min read
Document extraction is where most small businesses should start with AI
If you want a first AI project that pays for itself quickly and carries little risk, pulling structured data out of documents is almost always the right place to begin.
AIOperationsData - ·2 min read
Why your AI pilot stalled, and why it was probably not the model
Most AI pilots that quietly die do not fail because the model was weak. They fail for reasons that have nothing to do with AI at all.
AIOperations - ·2 min read
Retrieval over your own data beats a cleverer model
For most operational questions, giving the model access to your own documents matters far more than reaching for a bigger, smarter model.
AIDataOperations - ·2 min read
Human in the loop is a design decision, not an afterthought
Where a person reviews AI output is not a safety patch you add at the end. It is a core design choice that shapes whether the whole thing works.
AIOperations - ·3 min read
How we measure whether an AI feature is worth keeping
An AI feature should earn its place with honest numbers, not enthusiasm. Here is how we decide whether to keep one, fix it, or quietly retire it.
AIOperationsData - ·2 min read
Choosing a model is a smaller decision than you think
Teams agonise over which model to pick and underinvest in the things that actually determine whether an AI feature works. The choice matters less than the surroundings.
AIOperations - ·3 min read
The hidden cost of an AI feature is the review time
The model bill is the cost everyone watches. The cost that actually decides whether an AI feature pays off is the human time spent checking its output.
AIOperations - ·3 min read
Classifying inbound email without pretending it is magic
Sorting and routing a busy inbox is one of the most useful things AI can do in operations, as long as you design it for the cases it will get wrong.
AIOperationsAutomation - ·3 min read
Drafting and triage: where language models earn their place
Across the workflows we build, two uses pay off more reliably than any other. Drafting a first version and triaging what comes in. Here is why they work so well.
AIOperationsAutomation - ·3 min read
Guardrails for AI that touches customer facing work
When AI gets near customers, the stakes change. These are the practical guardrails we put in place before any AI output reaches the people who pay you.
AIOperations - ·2 min read
When not to use AI, from a team that sells AI work
We get paid to put AI into businesses, so it matters when we tell a client to walk away from it. Here is where AI does not belong.
AIOperations - ·2 min read
Clean data is still the whole game
Every few years a new tool promises to make data quality stop mattering. It never does. Here is why clean data is still the precondition for everything.
AIDataOperations - ·2 min read
Prompts are not a strategy
A clever prompt can feel like a breakthrough in the room. It is not the same as a system you can rely on every day. Here is the gap, and how to close it.
AIOperations - ·2 min read
What changed in AI tooling over the last six months
Setting aside the noise, a few things genuinely shifted in AI tooling lately. Here is what actually mattered for the businesses we work with.
AIOperations - ·3 min read
Turning a messy inbox into a structured queue
A shared inbox is where work goes to get lost. Turning it into a structured queue is one of the most reliable early wins for AI in operations.
AIOperationsAutomation - ·2 min read
The quiet wins: AI inside workflows nobody sees
The most valuable AI in a business is usually invisible. It lives inside back office workflows, doing unglamorous work that nobody notices until it stops.
AIOperationsAutomation - ·3 min read
Confidence scores, and why your team should see them
When an AI feature tells you how sure it is, and your team can see it, the whole system gets safer. Hiding that number is a mistake we see often.
AIOperations - ·3 min read
Building an AI feature you can switch off
An AI feature you cannot turn off is a risk, not an asset. Designing for the off switch from day one is what makes teams comfortable saying yes.
AIOperations - ·3 min read
Why we start AI work with a single workflow
The fastest way to get AI wrong is to start big. We start with one workflow, end to end, because that is where real learning and real trust come from.
AIOperations - ·3 min read
Extraction accuracy is a process problem, not a model problem
When AI pulls the wrong data out of a document, the instinct is to blame the model. Usually the real fix is in the process around it.
AIDataOperations - ·3 min read
How to brief a model on your business without overfitting
Give a model enough context to be useful, but not so much that it learns your exceptions as if they were rules. The skill is knowing which is which.
AIFoundationsOperations - ·2 min read
The cost of getting AI wrong in a regulated workflow
In regulated work the failure mode is not a bad answer, it is a confident wrong answer that nobody caught. The cost is measured in fines, lost licences and trust.
AIOperationsFoundations - ·3 min read
A practical checklist before you add AI to operations
Before a model touches your operations, a handful of plain questions will tell you whether you are ready or whether you are about to automate a mess.
AIOperationsFoundations - ·3 min read
What good AI adoption looks like on the floor
Adoption is not measured in licences bought. It is measured in whether the people doing the work reach for the tool without being told to.
AIOperationsAdoption - ·2 min read
Real time visibility is the foundation AI sits on
A model can only reason about what it can see. If your view of the business is a day old, so is every decision the AI makes on top of it.
AIFoundationsData - ·3 min read
One source of truth, and why everything else follows
When everyone agrees where the real answer lives, arguments end, reports reconcile and AI finally has something solid to stand on.
FoundationsDataOperations - ·3 min read
Spreadsheet sprawl is a symptom, not the disease
The shadow spreadsheets running your business are not the problem. They are smart people routing around a system that stopped serving them.
FoundationsOperationsData - ·3 min read
Build, buy, or configure: how we actually decide
Build, buy or configure is not a matter of taste. It comes down to whether the thing is genuinely yours and how fast it needs to change.
FoundationsOperations - ·3 min read
Where low code fits, and where it does not
Low code is a sharp tool for a specific job. Knowing where it shines and where it strains is the difference between leverage and a tangle.
FoundationsOperations - ·3 min read
The three foundations: clean data, sharp processes, real adoption
Almost everything we believe about making technology work, AI included, comes back to three foundations. Get them right and the rest is detail.
FoundationsDataAdoptionAI - ·3 min read
Integration is where most operations quietly break
Most operational pain is not inside any one system. It lives in the gaps between them, where data is copied by hand and nobody owns the seam.
OperationsIntegrationFoundations - ·2 min read
Automating the handoff between two teams
The riskiest moment in most processes is when work passes from one team to another. Get that handoff right and a lot of friction simply disappears.
OperationsFoundations - ·3 min read
Reporting you can trust starts with the data model
A report is only as honest as the structure underneath it. If two people get two numbers for the same question, the problem is rarely the dashboard.
OperationsDataFoundations - ·2 min read
The Discovery Sprint: why we start with a working prototype
Long requirements documents describe a system nobody has used yet. We would rather spend 2 to 4 weeks building something you can actually click.
OperationsFoundations - ·3 min read
Data migration is a business project, not an IT chore
Moving data into a new system looks like a technical task. The hard decisions are about the business, and treating it as plumbing is how projects quietly fail.
OperationsDataFoundations - ·3 min read
Vendor lock in and the total cost of a system
The sticker price of a system is the easy number. The cost that decides whether you regret it is the one nobody quotes you up front.
OperationsFoundations - ·3 min read
Scaling operations without scaling headcount
Growth usually arrives as more work, and the default answer is more people. Often the real constraint is not capacity but how the work is structured.
OperationsFoundations - ·3 min read
What one screen instead of nine actually does
Counting the systems a person opens to do one task is a fast way to find waste. The cost of the gaps between screens is larger than it looks.
OperationsFoundations - ·2 min read
Maritime operations and the paperwork problem
In maritime businesses the work is physical but the operation runs on documents. When the paperwork is scattered, the whole operation runs on memory.
OperationsMaritimeFoundations - ·3 min read
Professional services: turning billable time into clean data
In a professional services firm, time is the product. When the record of it is messy, you are guessing at your own margins.
OperationsProfessional ServicesData - ·3 min read
Manufacturing: from work orders to live visibility
On many shop floors the work order is the plan, but the truth of where a job actually is lives in people's heads. Closing that gap changes everything.
OperationsManufacturingFoundations - ·3 min read
Field service when the work happens offline
Field teams rarely work where the signal is good. The systems they rely on have to keep working when the bars drop to zero, and reconcile cleanly the moment they reconnect.
OperationsField Service - ·3 min read
Data hygiene is a habit, not a project
Cleaning your data once and declaring victory guarantees you will be doing it again in eighteen months. The businesses that stay clean build the discipline into how they work every day.
OperationsData - ·3 min read
Why your systems do not talk, and what it costs you
When your tools cannot pass information to each other, your people become the integration. The cost shows up as wasted hours, stale numbers, and decisions made on data that is already wrong.
OperationsIntegration - ·2 min read
Adoption is the feature that makes the others work
A system nobody uses is worth nothing, no matter how well it is built. Adoption is not training bolted on at the end. It is something you design for from the first conversation.
OperationsAdoption - ·2 min read
Process mapping before software, every time
Building software before you understand the process just makes the mess run faster. We map how work actually flows first, because that is where most of the value, and most of the surprises, live.
OperationsFoundations - ·3 min read
The real cost of manual data entry
Manual data entry looks cheap because the cost is spread thin and never invoiced. Add up the hours, the errors, and the work people stop doing, and it is one of the most expensive habits a business keeps.
OperationsData - ·3 min read
Picking the right stack for the problem, not the trend
The best technology choice is the one that fits the problem, the team, and the timeline, not the one trending this quarter. Chasing the new thing usually costs more than it saves.
OperationsFoundations - ·3 min read
Why we stay after the build
Software is not finished when it ships. It is finished when it stops fitting the business, which is never. We stay because the value of a system is in the years after launch, not the launch.
OperationsPartnership - ·3 min read
Operations debt, and how it quietly accumulates
Every workaround you keep is a small loan against the future. Individually harmless, collectively crippling, operations debt builds in the dark until the day the business cannot move.
OperationsFoundations - ·3 min read
Modernising without ripping everything out
Modernisation does not have to mean a risky big bang rebuild. The smarter path is usually to keep what works, replace what hurts, and move in steps the business can absorb.
OperationsFoundations
