Selling with AI: Why product knowledge and judgment matter

Artificial intelligence rarely appears in retail as a spectacular robot. More often it is embedded in sales forecasts, order suggestions, product descriptions, search functions, or software that flags unusual inventory. For employees, a more practical question than "Will AI replace my job?" therefore arises: Which decisions can a system prepare, where does it need product knowledge, and when must a person correct it?

Austrian retail therefore offers a particularly relevant field of work. According to the AMS JobBarometer, 65,004 job ads in 2025 were for the occupational subgroup retail salesperson (m/f); the outlook for 2026 to 2028 is assessed as stable. At the same time, Statistics Austria reported in June 2026 that already 30 percent of Austrian companies with ten or more employees use AI technologies. Employment does not automatically disappear as a result. It changes where sales, merchandise management, and digital processes come together.

The short answer for employees

Routine tasks are prepared more thoroughly: AI can provide drafts, forecasts, sorting, and initial responses.
Responsibility remains human: Prices, orders, complaints, and customer inquiries require transparent approvals.
Product knowledge becomes more valuable: Only those who know the assortment, season, target group and store can detect implausible suggestions.
Digital competence becomes more concrete: Programming is not required for every role, but good briefing, reviewing and documenting are.
Advice remains a differentiator: In cases of complex products, uncertainty and complaints, empathy, experience and decisive decisions matter.

Why AI is taking off in retail right now

Austrian company data show a clear dynamic: according to Statistik Austria, the share of AI rose from 9 percent in 2021 to 20 percent in 2024 and 30 percent in 2025. Of the companies not using AI, 77 percent had not dealt with it at all. Those who had considered using it and then abandoned the idea cited, among other things, a lack of internal expertise, data protection concerns and legal uncertainties.

For employees this mix is important. On the one hand, systems are becoming more common. On the other hand, many businesses still lack a robust process for selection, training and oversight. Those who know practice on the sales floor and can critically assess technical proposals fill precisely this gap. This applies both to large store networks and to regional specialist shops.

The WKO cites marketing, back office, customer service, inventory management, and sales forecasting among the AI applications in retail. It refers to studies suggesting that generative AI can save up to 20 percent of working time. This is not a guarantee for every branch. Whether time is actually freed up depends on data quality, integration, error rates, and the necessary controls.

Seven retail tasks that are changing concretely

1. Sales forecasts and order suggestions

A system can combine past sales, season, promotions, and inventory levels. This results in a suggestion of what quantities to order. However, the specialist checks local specifics: construction sites in front of the branch, a city festival, weather changes, delivery problems, or customers who are statistically barely visible. The benefit lies not in blind ordering, but in a faster initial assessment.

2. Inventory and replenishment

AI can mark conspicuous discrepancies, for example, if an item should be in stock according to the system but is missing from the shelf. Employees then clarify the cause: incorrect booking, damaged goods, theft, misplacement, or delayed delivery. Good inventory work thus becomes less of a pure search task and more of a root cause analysis and clean correction.

3. Product descriptions and online assortment

Drafts for online shops, newsletters, or social media can be created from structured product data. Before approval, dimensions, material, scope of delivery, price, availability, and legally relevant information must be correct. Invented accessories or an incorrect product feature can trigger complaints. Retail knowledge is therefore the quality control between the text generator and the customer.

4. Search and product recommendations

Digital recommendations can guide customers to suitable variants. However, they do not automatically replace personal advice. For products that require explanation, someone must check whether a recommendation fits the need, budget, and intended use. Good staff also explain transparently why they change a suggestion or advise against an automated recommendation.

5. Customer service and initial responses

An assistant system can prepare standard answers to questions about opening hours, return procedures, or delivery status. Special cases remain delicate: damaged goods, conflicting payment data, goodwill gestures, warranty issues, or emotionally charged complaints. Clear handover rules are required here. The AI may structure the case; the binding decision is made by a responsible person.

6. Price and promotion control

Software can detect discrepancies between the merchandise management system, the online shop, and shelf labels. Still, a person must assess which information applies and how an error is resolved with customers. Especially for short-term promotions, a traceable approval chain is essential. A quick suggestion is worthless if no one takes responsibility for implementation.

7. Store analysis and daily planning

Patterns can be identified from footfall, sales, returns, and delivery data. Managers can use this to set priorities for goods receiving, customer assistance, or replenishment. Individual performance evaluation must be strictly separated from this. Teams should know which data are processed, what they are used for, and who may see the results. AI is not a free pass for non‑transparent surveillance.

What remains human on the sales floor

The WKO emphasizes personal, competent advice as a trust factor in current trends in brick-and-mortar retail. This matches practice: customers often state their needs incompletely, compare conflicting information, or change their priorities during the conversation. A system may recognize a pattern. An experienced salesperson additionally recognizes uncertainty, misunderstandings, and the moment when less choice is more helpful.

Human strengths are especially apparent in:

complex or safety-critical products,
complaints and goodwill decisions,
conflicting inventory or price information,
individual combinations outside typical buying patterns,
accessibility, language barriers, and special support needs,
Situations where trust is more important than speed.

Which skills really matter now

Product and process knowledge

Anyone who understands how ordering, goods receipt, shelving, checkout, returns, and the online shop are connected can look for errors in the right place. AI competence without process knowledge remains superficial.

Formulate good work assignments

A useful request specifies the goal, target audience, available data, constraints, and desired format. "Write a product text" is weak. Better is: "Create an objective draft from these approved product attributes; add nothing and mark any missing information."

Verify results for plausibility

Employees should check figures against the inventory management system or the cash report, not just against their gut feeling. For texts, checklists for facts, tone, completeness, and legally required information help. Verifying plausibility also means rejecting a suggestion.

Handle data responsibly

Customer, payment, personnel, or unpublished company data should not be entered into freely available tools without review. What matters are the approved application, the specific purpose, and internal rules. When in doubt, clarify before entering data, not after.

Mastering handovers and exceptions

Teams need a clear answer to three questions: When may a suggestion proceed automatically? When does a second person review it? When is it handed over to branch management, the relevant department, or customer service? Good exception handling prevents time savings from turning into later error correction.

Inform customers clearly

Employees don't have to give technical lectures. However, they should be able to explain whether a recommendation was prepared automatically, which information has been checked, and who makes the binding decision. That creates more trust than a seemingly all-knowing system.

AI competence is not a one-off training

The AI service center of the RTR points out that Article 4 of the AI Act has required measures for adequate AI competence since 2 February 2025. The scope and depth depend on technical knowledge, experience, training, and the deployment context. For retail, this means: a person who only reviews approved text drafts requires different training than an executive who selects forecasts or procures systems.

As part of the EU project Skills4Retail, the WKO has already offered its own modules on AI in retail, e-commerce, customer experience, analytics, and ethics. This shows how continuing education can be set up in a practical way. A helpful learning cycle consists of a short introduction, a guided use case, documented errors, and regular refresher training.

Anyone who needs a general introduction can additionally consult the jobspot.at-guide to "AI competence at work use. For a broader foundation, the article on is recommended digital skills in the job market.

A secure seven-step workflow for retail businesses

Limit the use case: Choose a concrete task, e.g., product text draft or inventory discrepancy.
Classify data: Clarify which information may be used and which is off-limits.
Assign responsibility: One person decides on approval, correction, and cancellation.
Set review criteria: Define facts, figures, mandatory information, tone, and potential customer impact.
Test on a small scale: Start with a few items, a single store, or for a limited time.
Document errors: Capture not just time savings, but also rework and types of errors.
Retrain the team:Update rules based on real cases and discuss exceptions together.

How employees can make AI experience visible in job applications

“I can work with AI” remains too vague. Experience becomes meaningful when the task, control, and result are mentioned together. Examples:

“Created product text drafts from approved master data and checked them against a fact-check list before publication.”
“Plausibility-checked order suggestions with local promotional and delivery data and documented deviations.”
“Pre-structured standard customer service inquiries; consistently forwarded complaints and warranty cases to the responsible departments.”

In a job interview, it is worth asking a follow-up question: “Which AI applications are already approved in the branch, who checks the results, and how are employees trained?” The answer reveals more about a company's digital maturity than a general promise of innovation. Those coming from a different industry can use the guide toLateral entry and career change structuring transferable experiences.

Three typical practical cases

Case 1: The order proposal ignores a local event

A branch expects significantly higher demand because of an event. The system does not know the special case and proposes the usual quantity. The responsible specialist documents the reason, adjusts the order with justification and reports the result back. AI prepares; local knowledge decides.

Case 2: A product description invents a property

The draft sounds convincing but mentions a material that does not appear in the master data. The text is not 'roughly' corrected but stopped. Only after clarification with purchasing or the supplier is it published. Speed is less important here than reliable product information.

Case 3: A complaint does not fit the standard response

The assistant system recognizes the subject as a return and suggests a text. In fact, the issue concerns damaged goods and possible follow-up costs. The sales associate forwards the case with a clear summary to the responsible department and does not promise a solution for which they have no authorization.

Common mistakes when using AI in retail

Measuring time savings but ignoring correction effort.
Publishing unverified product information.
Entering customer data into non-approved tools.
Presenting automatic recommendations as binding decisions.
Training teams only after rolling out a system.
Treating store-level knowledge as resistance rather than as necessary quality control.
Not defining a clear handover for complaints, warranties, or pricing errors.

FAQ on AI in retail

Will cashier and sales jobs disappear because of AI?

Individual routines can be automated. However, the AMS still reports a very high volume of job postings for retail salespeople and a steady trend for 2026 to 2028. Roles are changing mainly through more digital process work, increased oversight, and cross-channel advising.

Does every salesperson need to be able to program?

No. For many roles, approved tools, clear work instructions, fact-checking, data protection and exception handling are more important. Technical specialist knowledge becomes relevant where systems are selected, integrated or evaluated.

Am I allowed to input customer data into an AI tool?

Not without clear company authorization and a legitimate purpose. Employees should only use approved applications and follow internal data protection rules. If in doubt, the responsible unit should be consulted before any data is entered.

What further training is advisable?

The most effective approach is training tied to the actual use case: start with basics and risks, then a guided example, clear validation criteria and regular updates. General prompt courses alone are not sufficient for responsible retail processes.

How can I recognize a good employer when it comes to AI?

Good companies specify allowed tools, prohibited data, responsible persons, and escalation paths. They provide training before use, record errors, and involve employees from sales and warehouse in designing processes.

Conclusion: Retail knowledge makes AI usable

AI in retail doesn't simply take over an entire occupation. It shifts work from design, search, and sorting toward review, exception handling, and authoritative communication. Employees benefit when they combine their product and customer knowledge with data literacy, clear work instructions, and responsible approval.

The next sensible step is small: choose a recurring process, record allowed data and verification criteria, and test it with clear human responsibility. This creates demonstrable AI competence that counts in everyday work and in the next job application.