Better Wi-Fi is channel planning, not more access points

There's a dead spot in the back office, so somebody bought another access point and mounted it near the problem. Coverage improved. Everything else got worse — video calls at the front of the building started stuttering, and a desk that used to be fine now drops.

That happens often enough to be a rule rather than bad luck, because Wi-Fi is not like ethernet.

Airtime is the resource, not signal

Every device on a given channel — yours and your neighbor's — shares one half-duplex medium and takes turns. A radio listens, waits for quiet, transmits. Only one transmission can succeed at a time within earshot.

So an access point is not a pipe you add capacity with. It is another participant competing for the same airtime. Put a third AP on channel 6 within range of two others already on channel 6 and you have tripled nothing; you have added a contender to a queue and made everyone wait longer between turns. Coverage went up, throughput went down.

Everything below is about giving each AP its own turn to speak.

The channels you actually have

On 2.4 GHz in the US you get channels 1 through 11, and at the standard 20 MHz width only 1, 6 and 11 don't overlap. Every other channel partly covers its neighbors, and partial overlap is worse than sharing — co-channel devices take turns, while overlapping ones simply raise each other's noise floor into interference nobody can decode. Never run 40 MHz on 2.4 GHz; there isn't room, and doing it occupies most of the band by yourself.

5 GHz has far more room, in two kinds. Channels 36 to 48 and 149 to 165 are yours to use freely. The block between them — 52 to 144 — is shared with radar, so an AP there must run a 60-second listening check before transmitting and must vacate the channel within 10 seconds if it detects a radar pattern. Near an airport, a coastline or a weather installation, that shows up as an access point going silent and every client on it dropping at once. Usable spectrum, but verify it behaves in your building before you depend on it.

Wider channels are a trade, not an upgrade

Doubling channel width doubles the theoretical rate and also doubles the noise you take in, costing about 3 dB of signal-to-noise every time. An 80 MHz channel starts roughly 6 dB worse off than a 20 MHz one at the same signal level, so it reaches less far and falls back to slower rates sooner. It also consumes four 20 MHz channels' worth of spectrum, which is spectrum the AP down the hall now can't have.

In a house with one AP and no neighbors, 80 MHz is free performance. In an office with four APs, or a strip of retail units sharing walls, 40 MHz beats it and behaves more consistently.

The move nobody makes: turn the power down

The instinct when coverage is poor is to raise transmit power. It rarely helps and usually hurts, for two reasons.

The link is two-way. Your AP might run 23 dBm; the phone in someone's pocket runs closer to 15 and has a worse antenna. Shouting louder only extends the range at which a client can hear the AP and associate with it, then discover its own frames aren't making it back. You've enlarged the region where the Wi-Fi appears connected and doesn't work.

And high power makes cells overlap. Clients pick an access point and are notoriously reluctant to leave it, so a device that walked past the front AP at full power stays glued to it three rooms away at a miserable rate — burning huge amounts of airtime to send very little data, slowing everyone else on that AP. Meanwhile every AP hears every other AP, and your careful channel assignment collapses back into one shared medium.

Good design is the opposite: smaller cells, lower power, more separation between same-channel radios. That's also the condition under which adding an access point finally does what people expect, because a new AP on a clean channel at modest power is genuinely new capacity.

What to do

Survey before you buy anything. On RouterOS's legacy wireless package, /interface wireless frequency-monitor wlan1 reports per-channel usage and noise floor; the current package scans with /interface/wifi/scan wifi1. Read from where people sit, not from the rack.
Assign 2.4 GHz strictly to 1, 6 and 11 at 20 MHz. Consider handing that band to IoT devices and putting everything that matters on 5 GHz.
Choose 40 MHz on 5 GHz for anywhere with more than two access points. Reserve 80 MHz for isolated sites.
Drop transmit power until neighboring cells stop overlapping heavily. Aim for around -67 dBm where clients live, and roughly 15 to 20 percent overlap between cells — enough to roam, not enough to linger.
Use one SSID with identical security and VLAN settings on every AP. Separate SSIDs per access point is the most common self-inflicted roaming problem there is.
Re-plan channels every time you add a radio. A plan for three APs is not a plan for four.

If the Wi-Fi has been fixed twice by adding hardware and is still bad, a survey will cost less than the third access point.