Where this goes next
A system that proves its own work
At Portcroft we have built a system that reads a defined set of sources, does the work, checks it back against where it came from, and holds it until a person says yes. It has run our own business since August 2026. What we point it at is the part that can change.
Running today. Everything in Insights came through here.
We built it for Portcroft first, and it has been running here ever since.
Today we draw on multiple sources, swept through the day. Every source we publish from has had its publisher's terms read and written down, and what each permits is enforced by the system rather than remembered: how much may be taken, how often, and whether the text may be kept at all. Sources that have not been cleared do not reach a draft. Language is no barrier to it, so French and German trade titles are read in the original rather than skipped, and whoever carried the story is credited by name. It works out what is worth writing about, drafts it, and checks the draft back against the article it came from, line by line. Nothing goes out until we have read it and said yes. After something is published it gets re-read against its source every week, because sources move and an insight that was right in March can be quietly wrong by July. Every run is recorded, the failures included, in the words of whatever refused.
Our system does not stop at the writing. What we approve is scheduled across the channels, published, and then measured. Where a post went out and no figures came back, we chase it rather than dropping it, so the numbers we work from cover everything we published and not only the parts that reported cleanly.
What Portcroft has actually built
Six capabilities, each of them straightforward on its own. Having all six in one system is what makes the output dependable enough to act on.
It watches sources you choose
A defined set, swept continuously and chosen for the categories the business buys in. Coverage you can name, and audit.
It decides against your rules
What counts as worth acting on is set by the business, written down, and applied the same way every time.
It checks its work against the source
The draft is held against what it came from, word for word where a number or a claim is involved.
It waits for a person
Nothing reaches the world on its own. Approval is a step, not a setting, and it cannot be skipped.
It keeps checking afterwards
Sources change. Anything already out is read again against its source, so something that quietly stopped being true gets found.
It writes down what happened
Every run, every refusal, in the words of whatever refused, and a run that did nothing says so.
The technology is the easier half
Knowing what the steps are, what has to be true at each one, and what ought to stop the whole thing before it goes any further, is the part that takes ten years rather than ten weeks.
We have spent that decade inside these processes, on the buying side, in manufacturing, with the suppliers and the specifications and the arguments about who signs off what. We know the shape of the process because we have run it. Engineers can be hired in a quarter. The decade cannot, which is why this is being built by people who have run procurement rather than by a company with more engineers.
So each step gets an engine of its own. It supports the step, and it gates it. Nothing moves on until the step before it holds, and when something does not hold, the engine says which step, on what evidence, and what would clear it. A step that does not hold stops there, rather than being carried forward into everything after it.
Proof is the part that is hard to build
Producing a report is the easy part. What decides its value is whether you can act on it straight away, with the checking already done for you.
That comes down to two things, and neither is about the writing. Can you see what a given line was drawn from, and are you told when that source moves. Get those right and the work can be trusted at a glance. Get them wrong and someone has to check it all by hand, which is the job you were trying to do something about in the first place.
This is the part we spent the time on. It is also the part that carries across to anything else we point it at, which is why it is worth more than the insights it currently writes.
Our system does not wait to be asked
A system that only answers questions is only ever as good as the questions somebody remembered to ask. Ours is built the other way round.
A second opinion you did not ask for
More than one reading of the same material, made separately, so a single opinion never stands unexamined. Where the readings disagree you hear about that, because a disagreement is worth knowing, and where one is contradicted outright the work is held rather than passed on.
It raises what nobody noticed
What is slipping, what has been sitting open too long, what is about to collide with something else. Each one arrives with the evidence behind it and what it costs to leave alone.
It carries the plan, not only the task
Work with dates on it, and a word when two things are about to collide or something has sat open too long. Coordination stops being something one person holds in their head and worries about on a Sunday evening.
It says when it does not know
Where the evidence is thin it says so, rather than filling the gap with something that reads well. Unmeasured is reported as unmeasured, never quietly as zero.
Built so nothing has to leave
In procurement, confidentiality is usually the condition rather than the preference. So it has to be settled when the system is built, not promised once it already exists.
It works where your documents already sit
Built to read the specifications and contract terms where they are, rather than requiring the lot to be shipped somewhere first and hoping for the best about where it ends up.
Nothing calls out
Our spend analysis workbook already works this way, on a local machine with no macros, no external calls and nothing transmitted. Client spend data is not uploaded anywhere to be analysed, and what we build next is held to that same line.
Nothing goes into public tools
Where AI supports the work it is on business terms rather than consumer accounts. Your material never touches free or publicly available tools, and it does not train anybody's model.
The picture stays yours
What gets built out of your documents belongs to you. It is worth very little to anybody else, and it goes when you say it goes.
The same shape turns up all over the business
Publishing was the first. Each of these is a process we have run ourselves, which is what makes it a credible next step.
- Origin, duty and landed costWhere something is made, what it costs to bring in, and what a change of origin does to the duty. The paperwork already exists and nobody enjoys reading it twice.
- Stock, life and spaceWhat is held and what it is worth, what is running out of life before anyone will need it, what has sat in the wrong place since it arrived, and what could be consolidated to give the space back.
- Production records and labourWhat a line actually made against what it was meant to make, where the hours went, and what would lift output or take cost out without anyone working harder.
The advisory work and the technology are one business
It is a fair question which one this is, so here is the answer plainly.
The advisory work is where the knowledge comes from. You cannot put a gate on a process you have never run, and every rule this system applies came out of doing the job rather than reading about it. The technology is what lets one senior person cover more ground without the work getting thinner, which is what usually goes wrong when a small firm grows.
Each half strengthens the other. The advisory work funds the build and tells it what to do, and the technology lets that advisory work reach further. The investment case rests on the two together.
What backing would do
Everything Portcroft builds comes out of the work itself. Data that will not line up, decisions waiting on somebody to go and find a number, and costs that are real enough but never plainly visible. The business is built to be funded out of what it earns, so backing would bring this forward rather than make it possible.
The advisory side
More work at once
Someone taking delivery days, so we can run larger engagements, or several at a time, without the senior attention being spread thin. It is the one thing most likely to cap what we can do next year.
Room to keep selling
Business development that carries on at full pace instead of stopping every time an engagement starts and short-term income needs attention.
Stronger models behind the work
The more capable models cost more to run, and the difference shows up in the quality of the finding rather than in the writing. Right now that ceiling is a cost decision.
The technology side
Tools that run where the data cannot leave
Spend data, supplier lists and contract terms, worked on without any of it going anywhere. Our analysis workbook already runs offline on a local machine. The rest should meet the same standard, because for most clients confidentiality is the condition rather than the preference.
Build it against a real library
A first manufacturer, a real set of specifications, and the messy shapes real documents turn up in. That is what turns a system that works into something that sells.
The paid tier, built properly
Portcroft Assistant is free and stays free as it is. A fuller version for procurement teams needs its own terms and liability position, built once and built right rather than in the gaps between billable work.
Come and see it running.
Whether you have a wall of specifications nobody can query, or you are interested in backing what comes next, we are happy to walk you through the whole thing and let you poke at it. If you would rather look before speaking to anyone, Insights is running on our technology page.