What problem does it solve?
A broadband line can test fine while the Wi-Fi inside the home is poor. Neighboring networks and extenders crowd the same channels, too many devices compete for airtime, and background traffic eats the bandwidth a video call needs. To the customer the internet feels broken, and the traditional response is to wait for a call, run a diagnosis and sometimes send a technician. Aprecomm, describing its ACT Fibernet deployment, calls this gap between "the internet is working" and "the internet feels broken" the place where trust was eroding and leading to churn.
Sky Italy's Head of Broadband Product Delivery made a related point at a conference: with more than 80% of devices connecting over Wi-Fi, fixing Wi-Fi settles most connectivity problems, and speed tests no longer describe what customers experience. What matters is latency, jitter and whether a given application works on a given device.
- Sky Italy's Head of Broadband Product Delivery said at Network X that more than 80% of devices connect via Wi-Fi and about 90% of smartphone data travels over wireless networks, so resolving Wi-Fi issues addresses most customer connectivity problems.From Speed Tests to Real Results: How Sky Italy Cut Support Calls by 20% Using AI (2026)
How does it work?
- Measure. Software on the gateway or router, plus a cloud service, collects Wi-Fi telemetry such as channel use, signal strength, retries and latency, and scores the quality of experience for each device and application. ACT Fibernet's system uses more than 15 parameters per score.
- Detect. The model spots interference from neighboring networks, congested channels, devices stuck on a weak band and early signs of degradation, often before the customer notices.
- Act. The system changes the channel, steers a device to a better band or access point, or prioritizes traffic that is sensitive to delay, for example a video call or a remote worker's laptop on weekdays. Sky Italy's proofs of concept covered channel changes and band steering (found by scanning off channel) and traffic prioritization for home workers.
- Support. When a customer does call, the agent sees device history, recent channel changes and application performance, so the call starts from a diagnosis instead of questions.
- Learn across homes. Patterns that fix one household feed the model for similar homes, and the operator's own telemetry becomes the training data.
- Audience
- Back office
- Autonomy
- Autonomous
- Adoption
- Early adopters
- Channels
- API and system to system, Agent desktop
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Contact deflection | Too few to pool | 15% to 25% Not pooled: up to 30% | 2plus 1 up to | 1 organization, 1 vendor |
| Quality score uplift | Too few to pool | 20% | 1 | 1 vendor |
| Churn reduction | Too few to pool | Not pooled: up to 18.8% | 0plus 1 up to | 1 vendor |
Value drivers: Lower cost to serve, Customer experience, Revenue growth.
Indicative value
A broadband operator with 1 million home subscribers
USD 100,000 to USD 1 million
Support call cost avoided per year
How this is calculated
Formula: subscribers * wifiCallsPerSub * reduction * costPerCall. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Home broadband subscribers subscribers, subscribers | 1,000,000 | 1,000,000 | The reference operator. |
| Wi-Fi related support calls per subscriber per year wifiCallsPerSub, calls per subscriber per year | 0.2 | 0.5 | Editorial assumption, replace with your own call reasons. |
| Reduction in Wi-Fi related calls reduction, fraction of calls | 0.1 | 0.2 | Conservative against the evidence on this page. Sky Italy's Head of Broadband Product Delivery reported 15 to 20% in a two week proof of concept (per the Network X write up), and Aprecomm reports 25% for ACT Fibernet. Source |
| Fully loaded cost of one support call costPerCall, USD per call | 5 | 10 | Editorial assumption, replace with your own cost per contact. |
What it leaves out: Leaves out avoided technician visits, churn (Aprecomm for ACT Fibernet and Plume for Armstrong describe fewer departures, but Plume's 18.75% for Armstrong is a vendor figure with no baseline), router software and cloud costs, and any managed Wi-Fi revenue.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
ACT Fibernet
India · Telecommunications · 2026
ACT Fibernet, a large fixed line broadband provider in India, runs ACT SmartWi-Fi on its Zippy router operating system with Aprecomm's agentic AI. The software scans in home Wi-Fi, scores the quality of experience of each device from more than 15 parameters, and fixes interference and congestion by switching channels and steering devices between bands, with no action from the subscriber. The first launch covered 250,000 homes. Aprecomm's later account of the deployment reports 25% fewer inbound support calls, with the deployment complete in 1.35 million homes.
- Contact deflection: 25%, reported after about six months of deployment
"With the AI catching and resolving issues before subscribers noticed them, the volume of inbound support calls dropped by 25%."
Claimed by: vendor - Quality score uplift: 20%
"Device QoE scores improved by an average of 20% over the 1.35m homes where deployment was complete."
Claimed by: vendor
Sky Italy
Italy · Telecommunications · 2026
Sky Italy ran two proofs of concept on home Wi-Fi. In the first, Cujo AI agents scanned off channel to find interference from neighboring networks and third party extenders and then changed channels or moved devices to another frequency. In the second, the operator identified customers working from home on weekdays and prioritized their laptop traffic. The Head of Broadband Product Delivery presented the work at a conference, and the organizer's write up reports that customer service calls fell by 15 to 20% within two weeks.
- Contact deflection: at least 15%, within two weeks of the traffic prioritization proof of concept
"Sky Italy achieved a remarkable 15-20% reduction in customer service calls within just two weeks."
Claimed by: organization
Armstrong
United States · Telecommunications · 2022
Armstrong, a broadband service provider, was running an architecture that Plume's page says trailed retail Wi-Fi products while customer calls and truck rolls increased. It adopted Plume's cloud based experience management platform, which Plume describes as data centric and AI driven, to improve in home coverage, give customers control of their Wi-Fi and let support staff troubleshoot remotely. Plume's page, citing an Omdia case study, lists a drop of up to 30% in Wi-Fi related service calls and up to 18.75% lower churn.
- Contact deflection: up to 30%
"30% drop in WiFi-related service calls."
Claimed by: vendor - Churn reduction: up to 18.8%
"18.75% reduction in customer churn."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Wi-Fi telemetry per access point and device, collected at short intervals
- Support contacts tagged by reason, to find the Wi-Fi related share
- Router make, model and firmware inventory
- Customer consent and notice for collecting device level data in the home
Systems to integrate
- Gateway or router management system with remote configuration rights
- Cloud analytics platform for telemetry
- Contact center agent desktop and ticketing
- Customer app, if customers can see or control their Wi-Fi
Complexity: Medium
The build needs gateways or mesh pods that can report telemetry and accept remote channel and band changes, which depends on the router software the operator controls (ACT Fibernet runs its own router operating system). It then needs a cloud service, a support dashboard connected to the ticketing system and a consent basis for collecting household device data.
- 1
Baseline the Wi-Fi call share
Tag support contacts and technician visits by cause so you know how many are Wi-Fi quality, not a line fault, and set the measurement window before launch.
- 2
Start with channel and band steering
Channel switching and band steering are low risk and visible in telemetry. Run them in a subset of homes with a holdout group so call reduction is measured against homes without the service.
- 3
Add traffic prioritization carefully
Prioritizing a category such as remote work video calls needs a legal check against traffic management rules and a clear customer facing description.
- 4
Give agents the device history
Show channel changes, device scores and application performance on the agent desktop, so the first call resolves more.
- 5
Scale by device and home type
Widen to more homes after the holdout shows a result, then measure churn over a longer period.
Guardrails
- Collect Wi-Fi and device data only for optimization and care, with notice and an opt out where the law requires it
- Limit traffic prioritization to categories justified by technical needs and disclosed to customers
- Rate limits on how often a home's settings change, so the optimizer cannot disconnect devices repeatedly
- A remote switch to turn optimization off per home or per router model
KPIs to instrument
- Wi-Fi related contacts per thousand subscribers, against a holdout group
- Device quality of experience score distribution, including the worst decile
- Technician visits and router swaps for Wi-Fi complaints
- Churn of optimized homes against comparable homes
- Disconnects and complaints caused by automatic changes
Human in the loop
The model changes settings by itself within rules that network engineers set. Engineers review changes that cause disconnects, support teams can switch optimization off for a home, and product owners approve any new prioritization category.
Common failure modes
- Averages hide the unhappy homes
- A rise in the mean score can coexist with a group of homes that got worse. Track the worst decile and complaint rates per home.
- Automatic changes interrupt a call
- Switching a channel mid call drops the session. Limit changes during active traffic and cap how often they occur.
- Results measured in the wrong window
- Two weeks of data (Sky Italy's proof of concept) is too short to judge churn. Plan a longer read for retention.
- Router estate too mixed
- Old or unmanaged routers cannot be steered. Start with equipment you can configure and plan replacement for the rest.
What are the risks and rules?
EU AI Act
Minimal risk
Tuning radio channels and traffic priorities in a home network is not listed in Annex III. Point 2 covers safety components in critical digital infrastructure, and the in home LAN and gateway are not the public electronic communications network. Recital 55 limits safety components to systems that protect the physical integrity of that infrastructure, which channel tuning does not do. The system also does not interact with the customer as a conversational system. A chatbot or voice assistant that explains the changes to customers carries the Article 50 duty to say it is AI, unless that is obvious.
Rules that apply
Guidance
- BEREC Guidelines on the Implementation of the Open Internet Regulation (BEREC, Europe). Sets how traffic management may differ between categories of traffic in the EU, which matters if the optimizer prioritizes some applications over others. The telecom consumer rules apply to this use case through the Open Internet Regulation's transparency and traffic management rules, not through contract, billing or switching rules. Check the current version.
- General Data Protection Regulation (European Union, Europe). Device lists and traffic patterns from a household are personal data, so purpose limitation and minimization apply.
Controls to put in place
- Documented purpose and retention limits for in home telemetry
- Legal review of any traffic prioritization categories
- Holdout group that stays on the previous setup for measurement
- Change log of automatic settings per home, available to support staff
Frequently asked questions
- Does AI Wi-Fi optimization reduce support calls?
Operators and vendors report it does. Aprecomm says ACT Fibernet's inbound support calls dropped by 25% once the deployment was in place, and Sky Italy's head of broadband product delivery reported a 15 to 20% reduction within two weeks of a proof of concept. Both are the reporting parties' own figures and neither was independently audited. The Sky Italy write up ties the figure to the traffic prioritization step, which prioritized laptops of home workers on weekdays from 9 AM to 6 PM, and names Cujo AI agents only for the channel scanning work, so that step may follow fixed rules rather than a learned model.
- What does the AI change in the home?
Mainly the channel a device or access point uses, the band it connects on, and the priority of delay sensitive traffic. ACT Fibernet reports that over 80% of Smart TVs and over 70% of laptops were steered to better channels in its first launch.
- Is prioritizing some traffic allowed?
In the EU, traffic management must follow the Open Internet Regulation, and BEREC's guidelines say reasonable management should rest on objectively different technical requirements of categories of traffic. Get legal advice before you prioritize a category and describe it to customers.
How to cite this page
Blits.ai AI Use Case Library, "AI for home Wi-Fi experience optimization", last verified 14 October 2026, https://www.blits.ai/ai-use-cases/home-wifi-experience-optimization. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 14 October 2026: First published