Chili's Bought 23,000 iPads First

Chili's upgraded its Wi-Fi, replaced ageing devices and simplified ordering before narrowing dozens of proposed AI uses to six or seven. That sequence matters because small firms gain more from dependable workflows than from adding an intelligent layer to unreliable systems.

Small-business AI is arriving through ordinary work: a £12 phone plan, an Instagram message, an inventory spreadsheet, a customer call. The businesses getting value are choosing a repeated job, connecting it to information they trust and keeping a person responsible for the result.

The £12 entry point

Dialpad and RingCentral now include transcription, call summaries and meeting assistance in communications packages priced for small teams, with some plans starting around £12 per user each month.¹ For many owners, the first useful AI feature will not arrive through a technology programme. It will appear as a setting inside software already paid for, attached to a task staff already perform.

That route is more important than it looks. A café owner does not need to debate an agent architecture before deciding whether missed calls are costing bookings. A tradesperson does not need a data strategy to test whether call transcription produces a clearer record of what a customer asked for. The practical starting point is the existing bill, the repeated task and the next action someone must take.

The sequence can be simple. Capture the call, summarise the request, draft the follow-up and ask a person to approve it. Each step removes clerical effort, while prices, promises and unusual requests stay with someone who understands the business. The value sits in shortening the route to finished work, not in giving the tool the widest possible remit.

This also explains why broad AI advice can feel detached from small-company reality. Owners are often told to prepare every dataset, redesign responsibilities and build a long-term strategy before testing a single use case. That may be sensible for a regulated company deploying automation across hundreds of teams, but it can delay a five-person firm from learning whether one feature saves twenty minutes a day.

A better first audit starts with software already in use. Phone, email, accounting, scheduling and social platforms are adding assisted features quickly, often without much ceremony. The owner should inspect those tools before buying another subscription, because the cheapest new product can still be expensive when it creates another login, another stream of notifications and another place where business context has to be rebuilt.

The workflow before the model

Chili's offers the clearest example of order of operations. The chain upgraded Wi-Fi across 1,200 locations, bought 23,000 iPads and simplified ordering before selecting a small group of AI ideas for further investigation.² Inventory forecasting remained under consideration, while AI phone ordering did not make the cut. A monthly governance group expects to reject more proposals than it accepts.

A small restaurant will not copy that programme, but it can copy the logic. Reliable connectivity, current menus, accurate opening hours and a clear booking process matter before an automated assistant begins answering customers. If those foundations disagree, the assistant does not remove confusion. It repeats the confusion faster and in a more confident voice.

Yelp's restaurant phone service shows where automation can earn its place. It can take food orders, connect with systems including Toast and Square, check OpenTable availability and communicate in 17 languages.³ For a busy takeaway that misses calls during service, that is a direct route from enquiry to revenue. For a full-service restaurant with outdated menu data and unreliable integrations, the same capability can produce wrong prices, unavailable dishes and bookings staff cannot honour.

Inventory teams are describing a similar mismatch. An inFlow survey found that 81% wanted AI in warehouse or inventory work, yet only 11% were using it, while 85% still relied on spreadsheets as a primary tool.⁴ The desired outcome was not a general assistant. Operators wanted a dependable answer to an ordinary question: what should we reorder, and when?

Spreadsheets are not evidence that a business has failed to modernise. They often survive because employees can see the numbers, correct an error and continue working without waiting for an implementation project. A new system has to beat that combination of familiarity and control. A polished recommendation is worthless when the owner cannot trace the stock figure, supplier lead time or sales history behind it.

The same buying discipline becomes essential when mistakes carry legal or financial consequences. In a DataTrace review of 200 residential title files, an AI search relying on public records missed at least one meaningful matter in 40.8% of searchable files when compared with searches supported by the company's title-plant data.⁵ AI may still organise records, flag patterns and prepare material for review, but speed does not turn incomplete evidence into an insurable decision.

Owners can apply a consequence test without creating a committee. Drafting a social post from approved photographs is low consequence because a person can review it quickly. Preparing a customer quote, stock order or contract summary requires stronger data and a named reviewer. Work involving legal rights, safety or substantial money needs specialist validation and a record of how the answer was reached.

This turns product selection into a more useful conversation. Ask a vendor to demonstrate one completed workflow using realistic information, show what evidence the system used, identify the human review point and calculate the monthly cost at normal volume. A product that cannot answer those questions is selling access to capability rather than a dependable business result.

Work moved, people stayed

Large companies often discuss AI in the language of headcount reduction. Small firms usually have no spare department to remove, so the same tools produce a different effect. Reporting gathered by the Guardian found small businesses using AI to help existing employees handle quotes, customer questions, documentation and other tasks that lean teams struggle to cover.⁶

OpenAI's analysis of workplace usage gives that pattern a useful explanation. Task crossover is more common in smaller organisations, where a person uses AI for work that would normally belong to another occupation.⁷ A restaurant owner becomes a copywriter for an hour, a salesperson explores customer data, and a founder prepares a first pass at a financial or legal question before deciding whether specialist help is needed.

That is not the disappearance of work. It is a change in who can begin it, how far they can take it and when expertise enters the process. AI compresses the distance between a business question and a workable first pass, which matters most when there is no analyst, copywriter or operations manager waiting in the next room.

The rise of high-revenue solo companies makes the point more dramatically. Stripe data reported by The Wall Street Journal showed that the number of one-person businesses earning more than $1 million doubled between 2023 and 2025, while the number passing $10 million nearly tripled.⁸ One founder was on course for $10 million in annual revenue with AI handling parts of coding, support, email, subscriptions and refunds.

The founder's contribution did not shrink. It became more concentrated in market choice, product standards, unusual customer cases and the promises the business was willing to make. Software absorbed coordination and repetitive execution, but someone still had to decide what good looked like and notice when the ordinary pattern no longer applied.

There is a less glamorous risk inside that model. The owner can become the permanent reviewer of customer replies, invoices, campaign drafts, schedules and automated recommendations. Every saved task returns as a queue of approvals, and the business begins to depend on one person signing off work late at night.

The answer is not maximum autonomy. It is clearer responsibility and a tighter definition of the job. A small team should know which outputs can be accepted routinely, which require review and which must move straight to a qualified person. It should also track repairs, because an assistant that saves an hour and creates forty minutes of correction is not producing meaningful capacity.

This is why the most useful target is rarely the fewest employees possible. It is the smallest team that can serve customers well, protect quality and keep learning. Use AI to remove work that drains attention, then preserve human time for judgment, relationships and original expertise, the reasons a customer chooses one small business over another.

How can a small business save time using AI for Instagram?

Instagram work contains the same pattern as calls, inventory and customer service. The repeated burden is easy to recognise: choosing photographs, shaping an idea, drafting a caption, checking the tone, planning the date and returning later to publish. A practical small business Instagram strategy removes parts of that sequence without inventing a generic identity for the business.

The source material should come from the company itself. Product photographs, menus, services, offers, events and previously approved language give the system something real to work with. AI captions for Instagram business posts become more useful when they are assembled around those materials, rather than generated from a vague instruction to make something engaging.

For restaurants, this can mean turning a dish photograph, opening-hour change and weekend offer into a reviewed post without asking a manager to begin from a blank page. For salons, it can mean using before-and-after images, treatment details and appointment availability while keeping claims accurate and the wording recognisable. Good Instagram AI content reduces repetitive production while leaving approval with the person whose reputation appears above the post.

A salon deployment in Singapore shows what happens when the workflow extends beyond publishing. Fushi Tech said its agent could answer messages across WhatsApp, Facebook Messenger, Instagram and TikTok, check branch-specific prices and hours, find the nearest location, book an appointment, send reminders and request a review afterwards.⁹ The useful result is not an impressive conversation. It is the removal of the hand-off between a late-night question and a confirmed booking, with complex requests transferred to staff.

Adoption figures suggest many owners are ready for that kind of narrow result. An OnDeck and Ocrolus survey of 805 US small businesses with working-capital loans found that 61% were using AI, and 91% of those users reported a positive effect.¹⁰ Those numbers do not prove that every deployment is sound, but they do challenge the idea that small-business adoption is waiting for a future wave of education or strategy work.

The strongest Instagram tools should complete recognised production work and leave the owner with a clear approval decision. They should choose from approved media, use current business details, prepare the caption and place the draft into a workable content calendar. That is how to automate Instagram content creation without asking the software to invent what the business stands for.

The distinction matters because small businesses lose trust at close range. A poor corporate message may be absorbed into a vast customer-service system. A poor post from a local café, stylist or shop can reach someone who knows the owner, recognises the premises and notices immediately that the language sounds borrowed.

Instagram content planning should therefore reduce blank-page work, missed posting dates and repeated formatting. It should not produce a larger volume of interchangeable posts. The gain is consistency with evidence of the real business still visible, not activity for its own sake.

The hand-off earns the budget

This week's stories make a stronger buying rule than any general call to adopt AI. Spend money where a repeated hand-off loses time, revenue or information, and where the result can be checked without creating a second job. A missed restaurant call, an unrecorded customer request, a stock reorder and an unfinished social post all meet that test more clearly than a broad promise of an artificial employee.

The useful systems also expose their boundaries. They show the information used, make correction easy and transfer consequential decisions to a person. That design respects the reason small firms can benefit so much from AI and the reason they can be damaged by it quickly: every hour matters, and every customer interaction carries a larger share of the company's reputation.

Chili's bought the devices and repaired the working layer before choosing six or seven ideas. A five-person business can follow the same order at a smaller scale by fixing one workflow, defining the approval point and measuring the result for a month. The next purchase should leave fewer hand-offs, fewer corrections and less Sunday-evening administration than the process it replaces.

Sources