Shopping assistants can now contact businesses and, in limited settings, help customers complete purchases. But they rely on merchants' prices, product information and availability records. If those records are wrong, better discovery can lead to more disappointed customers rather than more successful sales.
Last week's announcements put two very different versions of progress next to each other. Meta and OpenAI are making it easier for software to act on a shopper's behalf, while research from Scandit suggests retailers still struggle to keep basic stock information accurate. For an independent shop, the interesting question is whether the business can deliver what a persuasive digital assistant promises.
Meta's Muse can now place calls to US businesses for paying subscribers. A customer might ask it to enquire about a restaurant booking, rather than looking up a number and waiting on hold. It is an early, geographically limited service, but the change in behaviour is easy to picture: a business receives an enquiry from software representing someone who may never speak to the staff directly.1
A restaurant that answers the call can gain a booking without changing its menu or running another advert. Yet the conversation has to achieve something useful. An automated caller asking about a fully booked Saturday evening needs an accurate answer, a possible alternative and a way of confirming arrangements. Treating every AI caller as spam could lose genuine demand; letting every agent monopolise the line could waste a small team's time. Both problems are foreseeable even before agent calls become common.
OpenAI's ChatGPT Wallet brings a different task closer to the assistant. Its documentation says eligible US Pro users can use Wallet with their dot to pay for purchases, subject to rollout, card eligibility and user approval. Wallet is not available in ordinary ChatGPT conversations or ChatGPT Work, and it should not be mistaken for universal automated checkout.2
The restrictions matter because they separate a working early feature from predictions about how everyone will shop. The direction is nevertheless significant. A customer can increasingly delegate parts of searching, asking and paying, while the merchant remains responsible for supplying the product, honouring its price and resolving problems. The assistant can reduce a buyer's effort without reducing a seller's obligations.
There is a less obvious consequence for customer service. A good shop assistant used to recognise what a customer meant when they asked whether a particular colour was available, or whether an item could arrive before a birthday. An AI assistant making the same enquiry may have to infer the answer from a catalogue entry, website or phone conversation. Ambiguous product descriptions, inconsistent opening times and missing delivery information therefore have a practical cost. They make a business harder to buy from, even when its products are exactly what the customer wants.
This does not mean every café needs an agent answering its telephone tomorrow. It means checking the information a customer can already obtain without talking to an owner: opening hours, menus, booking links, product sizes, delivery cut-offs and refund terms. If those basics disagree across the website, social profiles and till system, a cleverer interface will not reconcile them for you. More automation can actually make the inconsistency travel faster.
Scandit supplied the week's most sobering statistic. In its research, surveyed retail associates said stock information matched the shelves only 46% of the time, and only 8% said they completely trusted their store data. Employees described working around errors with handwritten notes, team chats, personal photographs and memory. These are findings from Scandit's retail research, not a measured stock accuracy rate for every independent shop, but the underlying problem will be familiar to anyone who has searched for something that a system insists is available.3
A missed stock update used to irritate one customer who walked into a shop. Now consider the same error appearing on a generated website, in search results, in a shopping assistant's recommendation and at checkout. One inaccurate number can be repeated through several apparently authoritative interfaces. More connected channels can multiply a mistake as easily as they can multiply sales.
Fiserv's Clover announcement is relevant precisely because it attempts to connect those channels. Its new ecommerce capabilities let merchants build an online shop using AI, while linking products, inventory, orders and payments with the Clover system already used in-store. It also includes tools intended to improve how products appear in traditional and AI-powered search. These are announced product capabilities, not proof that every merchant's records will suddenly become accurate.4
The distinction is worth making. Synchronisation helps prevent a shop from maintaining three competing versions of a stock count. It does not know that the last item was damaged, that someone put it in the wrong place, or that the staff member who packed an order forgot to record the return. Software can connect records, but the business still needs a reliable habit for updating them. Better integration gives accurate information somewhere useful to go; it does not create accuracy from nothing.
Imagine a small fashion boutique preparing for a weekend promotion. It has five sizes listed for a popular jacket, photographs taken in its own shop, and an Instagram post scheduled for Friday. If the last medium was sold on Thursday but the count was never corrected, more traffic produces a specific failure: the customer chooses a jacket the retailer cannot supply. An assistant that helps complete the purchase has shortened the path to disappointment. The example is hypothetical, but the operational weakness is straightforward.
This is also why small businesses should be measured about the rush to generative engine optimisation. A UK government Business Academy event last Friday explained how companies can make their information more visible in AI search.5 That is a worthwhile subject, especially for independent businesses with distinctive products. Yet visibility is valuable only when the facts being made visible are current. A shop that can explain every item perfectly but cannot confirm availability has solved the easier half of the job.
The first investment may be surprisingly mundane. Check whether online and in-store stock agree for the twenty products that sell most often. Check whether prices and delivery estimates match what staff actually quote. Establish who corrects errors and how quickly those corrections reach the public catalogue. Those checks are less exciting than a new shopping bot, but they make the new channel worth having.
Square's launch of Squaremates offers a useful counterpoint to AI tools that only generate text. The agent can work with a seller's information inside Square, examine sales, identify lapsed customers, prepare a campaign and suggest changes to catalogues or staffing. It can bring a proposed action back to the seller as a usable result rather than leaving them to copy an answer into another application.6
That matters for a small shop because the owner seldom has a clean separation between marketing, inventory, staffing and customer service. A promotion can bring demand on a day when an employee is absent. A price change can affect the margin on an already thin-selling product. A reminder to return customers can work only if the shop has something worthwhile to offer them. The business record is part of the decision, not background material for a chatbot.
Gusto's State of Small Business 2026 report gives a sense of how far adoption has gone and how far it still has to travel. Its survey found that 67% of US small-business owners used AI for at least some tasks, but only 20% used it regularly across several functions or integrated it wherever the technology allowed. Nineteen per cent were experimenting and 28% used it for some tasks. Gusto surveyed its own customer base, so these results should not be presented as a universal census.7
The figures are revealing because an occasional draft is easy to count as adoption. A business process that works every week, uses current information and has a clear owner is much harder to establish. Someone still needs to know whether the system saved time, led to better service or created a new correction job. The sensible metric is whether staff can complete familiar work reliably, with fewer unnecessary steps.
OptiMantra's AI Scribe is a helpful example from a very different industry. It produces draft clinical notes inside an independent practitioner's existing record system, after recording a consultation with patient consent. The practitioner reviews and signs the note; the product does not finalise it or make a diagnosis. Its advertised starting price is $19 per provider per month.8
The interesting design choice is where the draft appears. Clinicians do not have to transfer text between a chatbot and their record system, or pretend that a plausible note is automatically correct. The tool reduces transcription and administration while keeping clinical responsibility with the practitioner. A similar principle works for a retailer checking a reordered item, or a salon reviewing an appointment reminder: put the useful draft next to the information and person needed to finish the task.
Manufacturing offers another illustration. Infor's partnership with Anthropic allows users to ask questions about orders, suppliers and production within workflows connected to Infor systems. Some actions, such as changing purchase orders or reallocating stock, can be carried out after a user reviews and approves them, using permissions already configured in the business software.9 A smaller manufacturer does not gain much from a beautifully written quotation if its material costs or capacity numbers are stale. Putting an assistant in contact with the business's real records makes the proposal more useful, although the estimator must still check the commercial decision.
There is an important cost implication. Small businesses may spend less on getting a first draft, a basic website or an automated reply. They can still spend a great deal correcting bad information, investigating missed orders or explaining mistakes to customers. A subscription price tells you almost nothing about that second cost. Ask vendors what information their product reads, how often it is updated, what happens when a field is missing and which actions require a person. Those answers will tell you more than a demo made with perfect sample data.
American Express's 2026 Shop Small Impact Study provides the clearest reminder that customer behaviour cannot be reduced to an interface. In its US consumer survey, 65% cited recommendations from friends and family as a way they discover small businesses. Reviews were cited by 45% and social posts by 39%. In the accompanying owner research, 90% expected to use AI tools during the holiday season.10
Those percentages measure different things and should not be treated as competing sales conversion rates. They do, however, show an uncomfortable mismatch in emphasis. Owners are preparing to automate, while customers still rely heavily on other people's experiences when deciding which independent businesses deserve a visit. More content might improve discovery, but the recommendation that follows a good visit may be the more durable asset.
For a restaurant, that might mean a customer sharing a photograph of the dish they actually received. For a hair salon, it might be a client showing a friend the result of an appointment. For a boutique, it might be someone recommending a jacket because the fit and quality matched the shop's description. None of these experiences becomes more convincing because a tool has generated fifty versions of the same promotional message. The business has to be recognisable in what it publishes, and dependable when someone walks through the door.
This makes Instagram for small business a useful test of whether AI is amplifying something genuine. A small business Instagram strategy can use scheduling, captions, cropping and content planning to make existing photographs and stories easier to share. Instagram AI content built from a shop's real products, a salon's actual work or a café's own dishes has a claim to the customer's attention that synthetic filler lacks. A beautiful invented meal is a poor advert if the restaurant cannot serve it.
The practical marketing goal is to help people make an informed choice. Instagram marketing for restaurants should show what is available, what it looks like and why it might be worth visiting. Instagram content planning for a local retailer should account for new stock, sold-out items and seasonal demand rather than blindly filling a calendar. AI captions for Instagram business posts are useful when they preserve the details a customer can recognise, not when they replace those details with claims any competitor could make.
It also changes how to judge success. Posting frequency and reach are signals, not the outcome. Look at the questions customers ask after seeing a post, whether they can find the advertised item, whether bookings were handled properly and whether people return. These observations are available to small teams without elaborate analytics. They tell the owner whether the public picture of the business matches the experience being delivered.
The new shopping tools are worth taking seriously. A shopper who dislikes phone calls may appreciate Muse. An eligible ChatGPT Wallet user may find checkout more convenient. Clover and Square may help a merchant make better use of existing records. These developments can bring independent businesses closer to customers who would otherwise never find them, and they deserve more than a reflexive dismissal as hype.
But the merchant has to decide what happens when the records are wrong, when the advertised product sells out or when a request needs a person to understand it. Start with the information customers depend on and the moments where an error costs trust. Then use AI for small business to reduce the repetitive work around those moments. The order matters because automation sends whatever is in the system further and faster.
A practical standard for the next purchase of software is whether it makes a promise easier to keep. That might be an accurate stock count, a confirmed appointment, a clearly described product or a timely answer to a customer's question. A small shop cannot compete with every large retailer's technology budget, but it can compete on knowing what it sells and delivering what it says. A shopping bot is another route to the counter. The experience after that still belongs to the business.
Meta Muse business calls, TechCrunch↩
Eligibility and operation of ChatGPT Wallet, OpenAI Help Centre↩
AI search discovery guidance event, UK Business Academy↩
Clinical scribe integration and pricing, PR Newswire↩
Holiday shopping and discovery survey, American Express via StockTitan↩