What problem does it solve?
Telecom stores sell technical products whose offers keep changing: new devices, discounts and trade in values, and plans whose eligibility depends on the customer's contract. At T-Mobile, a daily promotions report sent to retail representatives had become complex and hard to search, trade in values sat in other systems, and device details meant a visit to manufacturers' websites, so representatives often left the customer conversation to find an answer. At Bouygues Telecom, contact centre reps sifted through more than 500 articles, sometimes skimming up to 12 pages for one inquiry, and turned to supervisors or colleagues when unsure, which gave customers inconsistent answers.
Many customers, meanwhile, doubt their own technical knowledge and may feel embarrassed to ask a person for help, as UneeQ notes in its Deutsche Telekom case study; unsure buyers tend to pick the cheapest option and can be disappointed. A static screen in a store can show a brochure, but it cannot answer a question about the customer's own situation.
How does it work?
- Answer the associate in seconds. On a tablet or the store system, the associate asks in plain language about a promotion, a device comparison or a policy, and gets a summarised answer from approved content and live data, with the source.
- Bring the customer's context. After the customer is identified, the assistant shows their plan, contract end date, open cases and eligible offers, so the conversation starts informed.
- Build comparisons to show. It assembles device and plan comparisons that the associate can show or send to the customer.
- Serve customers at the screen. On a kiosk or a digital human screen, customers ask questions, compare options and check eligibility themselves, and the assistant calls an associate or books a slot when they want to buy.
- Keep every channel consistent. The same knowledge serves the contact centre, online sales and stores, so a customer hears the same answer everywhere.
- Audience
- Employee facing
- Autonomy
- Assist
- Adoption
- Early adopters
- Channels
- Kiosk and branch, Internal tools, Agent desktop, Web chat
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 |
|---|---|---|---|---|
| Users served | Not pooled | 6000 to 83,000 | 2 | 2 vendor |
| Accuracy | Too few to pool | 95% | 1 | 1 vendor |
Value drivers: Employee productivity, Revenue growth, Customer experience.
Indicative value
An operator with 300 stores
USD 1.1 million to USD 9.1 million
Associate time released per year
How this is calculated
Formula: stores * lookupsPerDay * openDays * hoursSaved * hourlyCost. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Stores stores, stores | 300 | 300 | The reference operator. |
| Product and account lookups per store per day lookupsPerDay, lookups per store per day | 20 | 40 | Editorial assumption, replace with your own store activity. |
| Trading days per year openDays, days per year | 300 | 360 | Editorial assumption. |
| Associate time saved per lookup hoursSaved, hours per lookup | 0.03 | 0.07 | Editorial assumption of about two to four minutes per lookup. Salesforce reports that Bouygues Telecom's Iris answers in seconds a task that once took minutes of manual searching. Source |
| Fully loaded associate cost hourlyCost, USD per hour | 20 | 30 | Editorial assumption, replace with your own cost. |
What it leaves out: Values only the associate time saved on lookups. It leaves out higher conversion and basket size from better informed conversations, shorter queues, faster onboarding of new staff, and the cost of the AI, devices, kiosks and integrations. Released time is only a saving if staffing or sales change as a result.
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.
Bouygues Telecom
France · Telecommunications · 2026
Bouygues Telecom built Iris, an employee facing agent on Salesforce Agentforce, over a rewritten knowledge base of more than 500 articles and a unified customer profile. Its 6,000 service reps ask Iris questions about billing, technical issues and promotions during calls, and the same answers are now available to associates in 500 retail stores. Before launch, Bouygues Telecom required Iris to exceed 90% accuracy or outperform a supervisor; Salesforce reports 95% on day one. Reps stay in control of what the customer hears.
- Accuracy: 95%, at launch, against a 90% threshold
"Iris cleared it on day one, reaching 95% accuracy."
Claimed by: vendor - Users served: 6000, contact centre service reps using Iris
"Today, 90% of the 6,000 service reps who use Iris rate it four or five stars — a clear signal of trust in the answers it delivers."
Claimed by: vendor
T-Mobile
United States · Telecommunications · 2025
T-Mobile built PromoGenius on Power Apps to give retail and call centre representatives one place for current promotions, discounts and trade in values, used on iPads on the shop floor. An agent built in Copilot Studio reads more than 20 device makers' websites, answers technical questions in natural language during a customer conversation and builds comparison tables that can be shown to the customer. Microsoft reports over 83,000 unique users and 500,000 launches a month for the app, which supports all T-Mobile retail stores and call centres.
- Users served: at least 83,000, unique users, retail and call centre staff
"The app, called PromoGenius, is the second most popular app at T-Mobile, supporting all T-Mobile retail outlets and call centers, with over 83,000 unique users and 500,000 launches a month."
Claimed by: vendor
Deutsche Telekom
Germany · Telecommunications · 2024
Deutsche Telekom uses a family of UneeQ digital humans that speak German and guide customers to products: Selena explains home broadband options after asking about the customer's home and lifestyle, Max answers customer questions on the Telekom website and app, and Mia works at events such as MWC. The vendor presents them as a way to give hesitant customers confidence in technical purchases. The case study shows conversion, cart abandonment and rating tiles without readable figures, so no metric is recorded. The case study does not describe screens in Telekom shops; the kiosk channel reflects Mia's use at events.
No outcome disclosed.
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- One current source for promotions, prices and trade in values with start and end dates
- Device specifications from manufacturers or a product data feed
- Knowledge articles rewritten for retrieval, with owners and review dates
- Store and customer data access rules per role
Systems to integrate
- Product catalogue and promotions system
- CRM and customer account data
- Point of sale and order systems
- Store tablets, kiosks or digital human screens
- Queue management or appointment booking in store
Complexity: Medium
An associate copilot over a clean knowledge base is quick to deliver. Account context and live promotions need CRM and catalogue integration, and a customer facing digital human adds hardware, speech in a noisy store, accessibility and privacy design.
- 1
Clean the knowledge first
Rewrite and standardise the articles associates use, as Bouygues Telecom did with Salesforce for its more than 500 articles, and give each an owner and review date.
- 2
Set an accuracy bar before launch
Agree a minimum accuracy on a test set of real associate questions and launch only when the assistant meets it. Bouygues Telecom required its agent to exceed 90% accuracy or outperform a supervisor.
- 3
Start with associates, then customers
Associates can judge and correct answers; customers cannot. Launch the copilot first and add a customer facing kiosk or digital human once answers are reliable.
- 4
Design the kiosk for a real shop floor
Test speech in store noise, offer touch and text alternatives, avoid showing personal data on a public screen, and make calling an associate one tap.
- 5
Share content across channels
Use one knowledge source for stores, contact centre and online sales, so an update reaches every channel at once.
Guardrails
- Answers only from approved content and live tools, with the source shown to associates
- Prices and promotions only from the catalogue, with end dates enforced
- No personal account data on a public screen without identification and privacy screening
- AI disclosure on every customer facing screen and digital human
- No emotion recognition or camera analysis of customers without a lawful basis and clear notice
KPIs to instrument
- Answer accuracy on a weekly checked sample of associate questions
- Weekly active associates as a share of store staff
- Time to answer common questions, before and after
- Conversion and basket size in stores with and without the assistant
- Kiosk conversations handed to associates and resulting sales
Human in the loop
Associates decide what to tell and sell to the customer and remain responsible for the sale. Content owners approve every article and promotion the assistant uses, and store managers report wrong answers, which are reviewed weekly.
Common failure modes
- Outdated promotions
- Associates quote an ended offer from the assistant. Keep promotions in a tool with end dates, not in documents.
- A gimmick instead of a tool
- A digital human draws attention but cannot answer real questions. Measure conversations that lead to help or a sale, not visits.
- Privacy on the shop floor
- Account details appear on a screen others can see. Design the screen and identification flow for a public space.
- Different answers per channel
- The store says one thing and the app another. Use one knowledge source for all channels.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A digital human or kiosk that talks to customers must be designed so that they are told they are interacting with an AI system (Article 50(1)). An associate copilot over product content is not listed in Annex III and is minimal risk. Inferring the emotions of employees at work is prohibited (Article 5(1)(f)); emotion recognition of customers by camera is high risk under Annex III point 1(c), biometric categorisation by sensitive or protected attributes is high risk under Annex III point 1(b), and categorisation that infers race, political opinions, religion or sexual orientation is prohibited under Article 5(1)(g). Both need separate legal review.
Rules that apply
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be informed that they are interacting with an AI system unless this is obvious from the context, which includes digital humans on store screens.
- Customers to get clearer broadband information (Ofcom, Europe). UK providers must give clear information about the broadband technology before a customer buys, including in person, so in store assistants must use the same approved descriptions.
Controls to put in place
- AI disclosure on every customer facing screen
- Content ownership and review dates for all articles and promotions
- Role based access to customer data in the store
- Accuracy tests before launch and after every content or model change
- Privacy impact assessment for kiosks and any camera or microphone in store
Frequently asked questions
- How accurate does a store assistant need to be?
- Set the bar before launch. Bouygues Telecom required its Iris agent to exceed 90% accuracy or outperform a supervisor before it entered the contact centre; Salesforce reports it reached 95% on day one, and Iris has since been extended to 500 retail stores.
- Should we start with a digital human or an associate copilot?
- Starting with associates carries less risk, because they can catch mistakes; the documented deployments at T-Mobile (PromoGenius, over 83,000 unique users among retail and call centre staff) and Bouygues Telecom (Iris) are both associate tools. Customer facing digital humans, such as Deutsche Telekom's Selena, who explains home broadband options, and Max, who answers questions on the Telekom website and app, guide customers to products; in a store, plan a clear handover to an associate when the customer is ready to buy.
- What privacy issues come with kiosks?
- Screens in a public space should not show account details without identification and privacy design, microphones and cameras need a clear legal basis and notice, and inferring the emotions of employees at work is prohibited under Article 5 of the EU AI Act.
How to cite this page
Blits.ai AI Use Case Library, "AI assistant for telecom retail stores, from associate copilot to digital human kiosk", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/retail-store-and-kiosk-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published