Almost no site is built badly on purpose. It is built in a hurry. There is a line start-up date, there is equipment waiting on the dock, and there is someone asking for the network to be up on Monday. The cable gets pulled, the connectors get crimped, the switch lights up, the machines respond, and everyone goes home convinced it turned out fine. And it did turn out fine: on day one, every site works.
The problem is that a site is not judged on day one. It is judged the day a link drops in the middle of production and someone has to find it, or the day new equipment arrives and twelve more cables have to go through a conduit that was already full. That is when you pay for what was not done. And you pay in the most expensive currency a plant has, which is an hour of unplanned downtime.
These are the six mistakes we see most often when we are called in to work on a site that already existed. None of them are exotic. All of them are avoidable.
1. No documentation, no labeling: the silent time bomb
This is the most common mistake and the most underestimated, because it has no visible consequence in the short term. An unlabeled site transmits exactly the same as a labeled one. Packets do not know whether the cable has a tag on it.
The consequence shows up on the day of the failure. A link degrades, a piece of equipment loses communication, and somebody has to stand in front of the rack and figure out which cable, out of the hundreds coming off the patch panel, corresponds to that equipment. With no labeling and no as-built drawings, diagnosis stops being an engineering task and becomes an archaeology task: unplug things and see what drops, trace the cable by hand along the tray, chase it with a tone generator. Every attempt burns time, and every test disconnection risks taking down something that was actually working.
The real cost is not the roll of labels nobody bought. It is the difference between a failure resolved in minutes because you know exactly where the link is, and a failure resolved in hours because nobody knows what goes where. In a plant running continuous production, that difference is measured in product that never came off the line.
Documenting means three concrete things: physical labels at both ends of every link, with identifiers that match; an as-built drawing that reflects what was actually installed, not what was originally designed; and a technical record with the test results. It is work you do once and it serves the entire useful life of the site. Without it, every new person who touches the rack starts from zero.
2. Not certifying the links: the test that proves nothing
There is a widespread confusion here that is worth taking apart. A continuity tester — the kind that tells you whether all eight conductors reach the other end in the right order — certifies nothing. It only tells you there is copper from one end to the other.
The problem is that a link can have perfect continuity and still be wrong. Twisted pair works because the twist cancels noise: the two conductors of a pair are twisted around each other at a specific rate so that interference picked up by one is cancelled by the interference picked up by the other. When the installer untwists the pair too far while terminating the connector, or over-tightens a zip tie and deforms the internal geometry of the cable, or bends it tighter than the minimum bend radius, that cancellation degrades. The result is crosstalk — the neighboring pair injects noise into the one you are using — and return loss from impedance mismatch.
A continuity tester detects none of that. The link passes the basic test and gets installed. And on day one it works, because at low traffic there is margin to spare and error correction absorbs the problem. The degradation appears when the link runs under real load: retransmissions climb, latency turns erratic, and the symptom that reaches the IT desk is not “the cable is bad”, it is “the machine is slow sometimes”. That “sometimes” is what makes people burn weeks chasing a ghost in software when the problem lives in the physical layer.
A link certifier measures the parameters that actually matter — crosstalk, insertion loss, return loss, delay — and compares them against the limits for the cable category that was installed. Either the link passes or it does not, and the result is on record. That certification is what turns a bundle of cable into a deliverable installation, and it is why in any serious structured cabling project certification is not an optional extra: it is part of the handover.
3. Sharing conduit with the electrical installation
This mistake is almost always made to save on civil work. There is already a tray run for power, and running the data cables through the same one avoids installing a second. It sounds reasonable. It is not.
The mechanism is electromagnetic induction. A conductor carrying alternating current generates a varying magnetic field around it. That field, crossing the data cable running alongside it, induces a parasitic current on the conductors. Twisted pair is designed to cancel that kind of noise, and it does so reasonably well when the source is far away. But field strength falls off with distance: right up against a power conductor, the induced noise can exceed what the twist is able to cancel, and from that point on it goes straight into the signal.
What makes it worse is that noise from the electrical installation is not constant. It is proportional to the current flowing. A site cabled alongside a feeder can run without problems while the load is low, and start failing when the full shift starts and that feeder carries the current it was designed for. The failure ends up correlated with the production rhythm, which is one of the hardest correlations to diagnose if nobody suspects the conduit.
The fix is not expensive if you think about it beforehand: separate conduits, with the appropriate separation distance, and crossings at ninety degrees when there is no way to avoid crossing — because a perpendicular crossing minimizes the coupling length between the two cables. What is expensive is discovering it afterwards, because correcting it means installing the conduit that was never installed, with the plant already running and the cables already pulled.
4. Ignoring motors and variable frequency drives
The previous point covers the electrical installation in general. Variable frequency drives deserve their own section, because they are an entirely different level of electromagnetic aggression.
A drive does not output a clean sine wave. It outputs a high-frequency switched waveform with very fast rising edges, and those edges are broadband emitters: they do not inject noise at one frequency, they inject it across a wide range. A large motor starting up also produces current transients that couple both by radiation and by conduction through the common ground. In an industrial environment with several drives running, the electromagnetic environment around the cable trays looks nothing like an office.
This is where you have to stop defending copper out of habit. Twisted pair has a limit: no matter how well shielded it is, it is still a metallic conductor sitting in a varying field. Optical fiber does not conduct electricity and transmits light through a dielectric. It is immune, by physics, to electromagnetic induction. There is no noise to couple into it.
So the decision is not “fiber because it is more modern”. It is: if the link has to cross a zone with drives, large motors, or welding equipment, fiber stops being a premium option and becomes the only medium that will run stably on that segment. Ignoring this produces the worst kind of failure: the intermittent one, the one that appears when a specific machine starts and disappears when it stops, and that can consume months of diagnosis before somebody connects cause with effect.
5. Expanding onto a disordered site instead of ordering it first
A disordered site has an unpleasant mathematical property: the disorder grows faster than the equipment. Every expansion made on top of an already tangled rack does not add one cable, it adds one cable plus the difficulty that cable contributes to every future move.
The dynamic is always the same. Because the rack is tangled, nobody wants to touch what is already there — the risk of taking down something in production is too high — so the new cable is run over the top of the existing mess. The next expansion finds a thicker mess and does the same thing. Layers form, like strata, and the cables in the lower layers get buried: they still work, but they are now untouchable. Nobody knows whether they are in use. Nobody dares remove them. And because they are never removed, the conduit fills up with dead cable that takes up space and dissipates heat without carrying a single bit.
The cost here is twofold. First, every future intervention takes longer and carries more risk, because moving one cable means dragging ten others. Second, and more serious, you reach a point where the only way to expand is to rebuild the rack completely — disconnect everything, remove the dead cable, re-run and reconnect — and that cannot be done hot: it requires a scheduled downtime window.
The practical rule is uncomfortable but it holds: when a site reaches the point where nobody wants to touch it, the next expansion should not be an expansion. It should be a reorganization with the expansion folded into it. You pay for a downtime window now, controlled and planned, instead of paying for an unplanned one later, when the rack collapses on its own.
6. Sizing the conduit with no room to spare
The last mistake belongs to the design stage, not the installation, and it is the one that costs the most civil work to correct.
When you size a tray or a duct for exactly the cables you are installing today, you are making an implicit decision: that this site will never grow. That is a bet nobody wins. Plants grow, lines get added, cameras and sensors and access points show up that were not in the original design.
On top of that, filling a conduit to the brim has an immediate effect, not just a future one. Tightly packed cables deform each other under pressure, and that deformation alters the geometry of the twisted pair — which brings us back to the mechanism from mistake two: degraded noise cancellation. And a dense cable bundle dissipates heat worse, which in turn increases signal attenuation in copper. A saturated conduit degrades the links it already carries, not just your ability to add new ones.
The economic consequence is the most brutal on this list. When the conduit is full, expansion stops being a cabling job and becomes civil work: breaking, running new conduit, patching, with the plant in operation and with all the permits and logistics that implies. It is a project an order of magnitude larger than what would have been needed if the tray had been sized with spare capacity from the start. Growth reserve is one of the best-returning investments in a site’s entire infrastructure, precisely because its marginal cost at design time is minimal and the cost of not having it is measured in construction.
The pattern behind all six
If you read the list again, all six mistakes share the same structure: none of them show up on handover day, and all of them charge you later, with interest. Labeling charges you during the failure. Certification charges you under load. Conduit separation charges you when production ramps up. Fiber charges you when the drive starts. Order charges you at the next expansion. Spare capacity charges you when you have to reopen the floor.
That is why a site should not be judged by whether it works on the day it is handed over — anyone can manage that — but by two questions: how fast can I diagnose a failure here, and how expensive will the next expansion be? A well-built site answers both well. A badly built one answers neither, even if today it is transmitting without a single error.
If you are about to build a new site, or you are looking at an existing one and you recognized more than two of these mistakes, it is worth reviewing it before you build on top of it. At De La Mar we handle the conduit, the structured cabling, and the civil and electrical infrastructure that supports it, in plants that are already running. If you want us to take a look at yours, get in touch and we will go through it properly.