Price the outage first, then talk technology
The four things Wi-Fi 7 (802.11be) brings — 320 MHz channels, 4K-QAM, multi-link operation (MLO) and resource-unit puncturing — all come with conditions. 320 MHz only exists on clean 6 GHz spectrum; 4K-QAM needs a very high signal-to-noise ratio and works mainly at short range with no obstruction; and MLO's value is not peak rate but losing fewer packets during handover and retransmission.
So the first step is not choosing a model but answering one question: what does one outage cost here?
Those four answers point to four different designs. Each is taken in turn below.
Healthcare: continuity over peak rate
The first bottleneck in a hospital is never speed. It is continuity while moving: a nursing trolley goes from ward A to ward B through two fire doors and a lift lobby, and the client has to hand over along that route without dropping.
The second is a complicated mix of clients. On one floor three kinds run at once: nursing trolleys needing throughput with low loss, PACS image review pulling hundreds of megabytes at a time, and infusion pumps and monitors sitting on 2.4 GHz at low rate for hours. Their radio behaviour is entirely different.
Worth paying for
- Radio density, not per-AP power. One radio source per corridor — distributed antennas or a wall-plate AP — replaces covering through walls with covering up close. After two walls, throughput is usually under a third.
- The handover gain from MLO. Two links held at once mean the moment of handover does not drop everything — most directly useful for trolley terminals.
- Network separation. Internal, external and IoT traffic on separate SSIDs and VLANs — a compliance requirement, not a performance option.
Easiest to get wrong
Treating wards like meeting rooms and mounting one AP in the middle of the ceiling to cover six rooms through walls. The corridor tests full at acceptance, the rate at the bedside is halved, and the trolley reconnects the moment it moves.
Education: concurrency over single-client rate
The load curve at a school is extreme: five minutes before the bell, fifty or sixty devices in one classroom start associating and authenticating at once; during the lesson there is almost no traffic; before the end they all leave together. The peak is not bandwidth — it is concurrent association and authentication.
Which explains why many schools still stutter after moving to Wi-Fi 7: the APs changed, the authentication server did not. 802.1X queues at the peak, handsets sit on connecting for ten or fifteen seconds, and the user concludes the network is broken.
Worth paying for
- Coordinated scheduling across APs. Classrooms sit close together and adjacent APs interfere; dividing airtime cleanly beats turning up transmit power.
- Authentication capacity. Routinely left out of the budget, and the main source of complaints in the first week of term.
- Plan capacity by room, not by floor area. A standard classroom is sixty devices, a lecture theatre two hundred. One average of clients per AP cannot be applied to a whole building.
Easiest to get wrong
Estimating AP count from floor area. A library and a laboratory differ fivefold in device density, and an area-based count is wrong at both ends.
Hotels: experience and running cost first
What makes a hotel different: no IT team of its own, but complaints that land at the front desk. When a guest room network fails, the first thing to notice is not monitoring — it is a guest picking up the phone. That makes the priority less peak performance than few faults, and remote handling when they happen.
Room layout decides the method too: a bathroom and a load-bearing wall cut each room into two spaces, so a corridor AP has to cross two walls to reach the guest, and 5 GHz barely arrives at the bedside.
Worth paying for
- One radio source per room — a wall-plate AP. This turns coverage into a question with a known answer, one AP per room. The cost is in point count, the gain in complaint rate.
- Central management. With dozens or hundreds of points you need remote restarts, bulk configuration and self-healing at a single failure — otherwise every issue means sending someone.
- Separate guest internet from TV. IPTV and guest browsing on different VLANs, so one does not drag the other down.
Easiest to get wrong
Applying an office rule of so many rooms per AP to guest rooms. Wall attenuation and occupancy swings in a hotel bear no resemblance to open-plan offices: a weekend at 90% occupancy carries three times the airtime load of a weekday at 30%.
Manufacturing: determinism over average rate
The acceptance metric on a factory floor should not be average rate but the latency distribution: an AGV runs from A to B across three AP boundaries, and one handover over the threshold stops the vehicle. A high average with a bad p99 is a fail.
Interference is different too. Variable-frequency drives, microwave equipment, metal racking and crane rails make 2.4 GHz nearly unusable, while 5 GHz is cleaner but still reflects. In this situation — part of the subcarriers interfered with — MRU puncturing does more than more power.
Worth paying for
- MLO and time-sensitive scheduling. Parallel links plus reserved airtime for cyclic traffic turn best effort into a floor.
- Keep 5 GHz and 6 GHz for production. Office devices on 2.4 GHz or a separate SSID; production must not share with guests or desk traffic.
- Survey by workstation and track, not by floor area. The metal structure of a line changes the whole radio picture.
Easiest to get wrong
Taking an office AP onto the line and accepting it because signal strength looks fine. The line needs latency under threshold at every metre of the path — a different survey method entirely.
The four industries in one table, and three phases
The conclusions above, compressed into one table to check a specification against:
| Industry | First bottleneck | Worth paying for | Easiest to get wrong | Acceptance metric |
|---|---|---|---|---|
| Healthcare | Continuity while moving | Radio density + MLO + network separation | Dense mounting covering wards through walls | No disconnects moving between wards; rate at the bedside |
| Education | Concurrent association and authentication | Coordinated APs + authentication capacity | AP count estimated from floor area | Association success in the five minutes around the bell |
| Hotel | Room coverage and running cost | One radio source per room + central management | Corridor APs covering rooms through walls | Rate at the bedside; share of issues fixed remotely |
| Manufacturing | Latency determinism along the path | MLO + time-sensitive scheduling + reserved bands | Office APs moved straight onto the line | p99 latency and loss, not peak rate |
Three phases: do not replace everything at once
- Phase one — survey and baseline. Measure the existing network until where it drops, when it drops and for how long are data. Choosing hardware before this is ordering from impression.
- Phase two — a single pilot. One ward floor, one lecture theatre, one floor of rooms, one production line, run for two weeks. Set control metrics, or you cannot tell whether the improvement came from the equipment or from fewer people.
- Phase three — repeat in batches. Freeze the pilot into a configuration template, then repeat by area. Do not run three configurations in one building; operations pays for it six months later.
Three judgements that always hold
Whichever industry, these three hold:
- Coverage comes from density, not power. More transmit power only carries interference further; it does not fix attenuation through walls.
- Acceptance has to be reproducible. A metric you cannot write a test for ends up as feels much better.
- Fix what can be fixed before replacing what could be replaced. Many projects that were sure the wireless was no good needed authentication, channel planning and placement corrected — not new hardware.