.png)
Research report
AI Does Not Remove the Bottleneck
Find your real workflow constraints before buying AI tools.
The shop floor already has the data. Nobody can read it.
MES integration, OT/IT data platforms and reporting that connects the line to the board.
Where we help
The work we are asked to do in manufacturing.
01
MES and line-system integration
Getting data out of systems that were never designed to share it.
02
OT/IT data platforms
One model that survives contact with ISA-95 and with the plant network.
03
Predictive maintenance
Models built on your sensor history, evaluated against what actually failed.
04
Operational reporting
Numbers the plant manager and the board both trust, because they come from one source.
05
Legacy plant modernisation
Replacing systems older than the engineers who maintain them, without stopping production.
The numbers behind our manufacturing work
Industry 4.0 pilots usually fail on the OT side of the OT/IT boundary. The data model that works in a warehouse does not survive contact with a line running ISA-95, and the people who understand the line are rarely in the room where the platform is chosen. We put them in the room, and we design the model around what the line actually emits rather than what the platform would prefer.
23
where our software runs on the floor
+6 pts OEE
what it moved
15
years working with operational systems
Selected work
Engagements in manufacturing, described rather than named.
An industrial hygiene consultancyTurning lab data into review-ready reportsLab PDFs read automatically, exposure limits checked against the right jurisdiction, and report prose generated with the regulation cited.Under 1 hrto produce a report that used to take six to tenRead the case
A US medical equipment resellerBuilding a sales channel the business could ownA storefront of their own beside a high-volume eBay business, with inventory kept in step both ways and double sales resolved automatically.5 minfor stock to reflect across both sales channelsRead the case
An enterprise strategy software ventureInfrastructure first, features second.Strategic scorecarding, assessments and innovation pipelines as separately licensable modules, on infrastructure built before feature one.3 productslicensable separately on one multi-tenant platformRead the caseQuestions we get asked
Do you understand the OT side?
Enough to know it is where these projects fail. A data model that works in a warehouse does not survive contact with a line running ISA-95, and the people who own the line have justified reasons for saying no to things. We start on the OT side of the boundary because that is where the risk sits.
Can you connect our MES to the rest of the business?
That is the usual brief. The engineering is manageable; the difficulty is that plant systems and business systems disagree about what a batch, a unit and a shift actually are. We reconcile those definitions before moving data, because otherwise the reporting is confidently wrong.
Will this interfere with production?
No, and the architecture makes that guarantee rather than the process. We read from the line rather than into it, work through a historian or a read-only interface where one exists, and anything writing back to a control system is out of scope unless you specifically ask for it.
Can you do predictive maintenance?
Where the sensor history is good enough to support it, which is the first thing we check. Prediction needs failures in the record to learn from, and a line that has been well maintained often has very few. Where the data cannot support it, condition monitoring delivers most of the value without the modelling.
What about older equipment with no connectivity?
It is the normal case. Options run from a gateway reading an existing protocol, through a historian already collecting more than anyone realises, to added sensing where the value justifies it. We survey what is genuinely reachable before designing anything on top.
How do you handle multiple sites?
By expecting them to differ, because they always do. The same nominal process runs differently at each plant, and a platform assuming uniformity fails at the second site. We build for a common core with local variation, and prove it on the site with the most awkward setup.
How we deliver it
.png)
.png)
