Why office Wi-Fi fails at capacity, not coverage

When Wi-Fi is bad in an office, the first instinct is almost always to add access points. Sometimes that is right. More often it makes the problem worse, and understanding why is the difference between a network that scales and one that gets a new budget line every eighteen months.

The medium is shared

Wi-Fi is half-duplex and contention-based at the channel-access level. Devices and access points compete for transmission opportunities, and modern features such as OFDMA and MU-MIMO can serve multiple clients within one opportunity. Those features make shared airtime more efficient, but they do not turn two access points on the same channel into two independent pools of capacity.

The practical consequence: two access points on the same channel that can hear each other share the same contention domain. Adding the second access point can improve coverage and client distribution, but it does not create a second independent pool of airtime on that channel. Add a third and the same shared-airtime constraint remains.

This is why a floor with eighteen access points can perform worse than the same floor with nine. Coverage was never the constraint.

What the symptoms actually look like

Contention problems are easy to misread because they do not look like weak signal:

• Signal strength is good everywhere, but throughput collapses when the room fills.
• Video calls degrade in meeting rooms and recover in empty corridors.
• Latency is spiky rather than uniformly high: the median is fine, the 95th percentile is terrible.
• Moving a laptop three metres fixes it, which people interpret as a coverage problem.

Weak signal looks different: it is consistent, location-bound, and it does not care how many people are in the room.

The four levers

Channel width. Wider channels mean higher peak throughput and fewer available channels. In a dense office, that trade is often wrong. Reducing channel width increases the number of non-overlapping channels available to the plan, which can be worth far more at capacity than a peak figure a single client will rarely reach. In high-density designs, narrower channels are often the safer default unless the available spectrum and client mix justify wider ones.

Transmit power. The instinct is to turn it up. In density you turn it down. Lower power produces smaller cells, smaller cells mean fewer clients per contention domain, and fewer clients per domain is the entire objective. High power also produces a subtler failure: access points can transmit at higher power than the client radios in laptops and phones, so a client can hear an access point perfectly while the access point struggles to hear the client. The client sees a strong signal, refuses to roam, and sits at the edge of a cell transmitting slowly, consuming airtime that everyone else is waiting for.

Minimum data rate. Leaving very low basic rates enabled keeps some management and broadcast traffic on air for longer, and edge clients can consume disproportionate airtime at low data rates. Raising the minimum basic rate removes the slowest rates from the cell and can encourage clients at the edge to move to a closer access point. It also reduces the effective cell size without touching transmit power.

Band planning. 2.4 GHz has limited usable spectrum and is shared with a wide range of devices, so dense office designs should avoid relying on it for capacity. 5 GHz remains important because client support is broad and it provides substantially more channel space. 6 GHz adds another large block of spectrum where the client mix and local regulations support it. The right design uses each band for what it can reliably carry rather than treating any one of them as the answer.

On automatic radio management

Most controller platforms can select channels and power automatically. Those systems are useful, with one condition: give them sensible boundaries.

Given the full channel list and the full power range, these algorithms will optimize for the conditions they observe, including transient interference and a building that is empty at 2am. Left unbounded, they will find a channel plan that is defensible in isolation and wrong for the room at 10am on a Tuesday. Constrain the channel list to the ones you planned for, constrain the power range to the cell size you designed, and let the algorithm work inside those walls.

What a design actually needs

A predictive survey using the floor plan and wall materials is the right starting point, and it is only a starting point. Predictive models do not know about the glass partition that went in after the drawings, the metal-backed whiteboards, or the neighboring tenant's network on the other side of a plasterboard wall. Once the network is live, the design is validated and fine-tuned by monitoring actual client behavior and network performance under real use. The difference is that the tuning starts from a planned deployment rather than from access points placed by eye and the hope that it will work.

The measure of success is not a coverage heat map. It is whether the support queue for Wi-Fi empties and stays empty.

Explore this expertise: Network & Infrastructure

All technical notes