Podium Made $150M Answering the Phone

Small businesses are paying for AI when it completes a recognisable job: answer the phone, prepare payroll, chase a lead, check an invoice or plan stock. The strongest products fit into work that already exists, use the business's own context and keep consequential decisions visible to the person responsible.

Podium's latest number is hard to ignore. The company says its AI-specific annual recurring revenue has passed $150 million in two and a half years, while its original software business took eight years to reach the same milestone.¹ That result sits alongside a week of launches that point in the same direction: small businesses appear far more interested in getting work finished than in getting another impressive conversation with a model.

The job is the product

Podium's agents answer calls, respond to leads, book appointments and follow up for local businesses in sectors such as automotive, home services and aesthetics.¹ Those are not experimental tasks tucked into an innovation budget. A missed call can mean a lost booking, and a lead that sits unanswered for three hours can be worth less than one answered in three minutes. The commercial value is attached to the job itself, which makes the buying decision much easier to understand.

Homebase made the same point from the world of hourly work. Its new assistant can handle the first steps after a 6 a.m. sick call by finding possible replacements, asking the owner for approval, then contacting staff until someone accepts; it can also check timecards, flag overtime and screen candidates.² The company says its earlier assistants have already handled more than 4.7 million administrative tasks. That is a more persuasive measure for a small employer than how many tokens a model can process or how well it performs on a benchmark.

Gusto is approaching the problem through payroll and people data. More than half a million small businesses use the platform, and its latest release lets owners connect that data to Claude, ChatGPT and other AI tools so an agent can prepare payroll, identify missing information and generate reports.³ The important design choice is that the owner still approves changes before they become final. The system takes on preparation without pretending that responsibility disappeared when the software got smarter.

This is a useful way to think about AI for small business more broadly. Owners rarely buy software because they want a new category in their technology stack; they buy it because invoices need chasing, shifts need covering, stock needs ordering or customers need replies. A tool that removes one of those recurring burdens can earn a place in the monthly bill. A tool that creates another place to log in, another set of prompts to learn and another output to check has to work much harder to justify itself.

Adoption is stuck in workflow

The adoption numbers from Australia show why access alone is no longer the most interesting problem. A Tech Council of Australia report estimated that broader AI adoption could create A$93 billion in annual value for Australian SMEs, yet only 25% currently use AI and only 5% use it broadly across the business; among firms already using it, 94% report measurable productivity gains.⁴ The gap between 25% trying something and 5% using it broadly is where the implementation problem lives.

Training aimed at small businesses is becoming more concrete for the same reason. Workshops are increasingly built around payroll, invoices, lead follow-up, SOPs, checklists and recurring administration rather than generic prompt-writing exercises. Anthropic's Claude Duet also follows that pattern: it can monitor connected apps, prepare briefings and handle tasks such as scheduling and financial forecasts, while requiring approval for sensitive actions.⁵ The useful question has shifted from "Can this model do something clever?" to "Can this fit inside Tuesday without making Tuesday harder?"

That distinction matters because small firms have very little spare implementation capacity. The person choosing a tool may also be handling payroll, supplier problems, customer complaints and marketing in the same afternoon. They cannot spend six weeks redesigning processes around a new interface, and they usually do not have someone whose job is to translate a general model into ten reliable workflows. A product wins when it arrives closer to the work than to the technology.

The same test applies to marketing. A small business looking at Instagram content planning does not need a blank generator that can produce fifty captions; it needs a repeatable way to turn its own products, offers, photos and brand cues into useful posts without losing the owner's judgement. That is also the difference between useful Instagram AI content and generic filler. For anyone asking how to automate Instagram content creation, the better answer starts with the existing content process, not with the model.

The interface is disappearing

Several launches this week suggest that the best small-business interface may increasingly be the one the owner already uses. Loyyal's Perxi puts ordering, customer service and loyalty inside WhatsApp under the business's own brand, so customers do not need another app or account.⁶ Google is taking a similar route from the assistant side, letting small-business owners give Gemini a website, brand kit, fonts or colour palette, analyse business records and connect to services including Google Business Profile, Ads, Workspace, Canva, Shopify, Wix, Zoho and Xero.⁷

VistaPrint makes that shift unusually tangible. US users can search more than 7,800 print products inside Gemini, turn an image or PDF into a print-ready design, preview it and send the finished version to a VistaPrint cart without leaving the conversation.⁸ The value is not that a chatbot can talk about printing. It is that the distance between thinking "I need new menus" and having something ready to order gets shorter.

This creates a different competitive problem for software companies. A polished standalone interface still matters when people spend their working day inside it, but it matters less when the customer reaches the product through an assistant, messaging app or another tool they already use. The software's data, permissions and actions then become more important than the number of screens it owns. Integration becomes part of the product experience, not a technical feature hidden on a settings page.

For a small firm, that can be liberating. There is no prize for maintaining a perfect collection of SaaS tabs, and every hand-off between systems costs attention. The practical version of "agentic" software may therefore feel surprisingly ordinary: ask for a job in a familiar place, let the system gather the right context, review the part that matters, then move on. Good software starts to disappear into the work.

Approval is a feature

Human approval is showing up too consistently to dismiss as temporary caution. Gusto prepares payroll but keeps final changes with the owner.³ Claude Duet requires approval for sensitive actions.⁵ In professional services, Thomson Reuters' new agentic CoCounsel platform is designed to complete tax and audit workflows and give professionals work that is ready to review rather than forcing them to prompt a chatbot through each step.⁹

That pattern solves a basic problem with broad claims about autonomy. "Approve every AI action" is not workable because it turns the person into a full-time supervisor of software. "Review the prepared payroll before it runs" is clear. So is "check the drafted engagement letter before it goes to the client" or "review the forecast before it informs a purchase". Human oversight becomes useful when the review point is attached to a consequence people already understand.

The distinction also changes how products should measure success. If an agent prepares 90% of a task but leaves the last 10% easy to inspect, that may be more valuable than a fully autonomous system whose work is difficult to verify. In finance, legal work, security or anything touching customers, reviewability is part of the output. Speed without a clean way to see what happened simply moves the cost from doing the work to checking the work.

This matters for creative work too. A small business can use AI captions for Instagram business posts, generate variants and speed up routine production, but the owner still knows whether a sentence sounds like the business, whether a claim is true and whether a photo represents what a customer will actually receive. The useful role for AI is to compress the production work around that judgement. The decision itself remains valuable precisely because the tool cannot own the consequences.

Machines are becoming customers

There is another reason business context matters: some customers may soon arrive through software rather than a browser. BCG says 68% of retailers expect agentic commerce to have a significant impact within two years, while 42% are already making their data discoverable to any external shopping agent and another 32% allow selected agents.¹⁰ For retailers, the audience is starting to include machines that compare products, prices, availability and offers before a human ever sees the options.

That creates a strange double requirement for smaller businesses. Product data needs to be clean, structured and legible enough for an external agent to understand, while the brand still needs to give the human customer a reason to care which result gets chosen. If every business becomes equally machine-readable but indistinguishable to people, the agent may reduce them to price, availability and fulfilment. BCG's warning that retailers risk becoming interchangeable fulfilment nodes is particularly uncomfortable for independents that compete through identity and trust.¹⁰

This is where small business Instagram strategy and product data unexpectedly meet. Instagram for small business is often where a customer learns what a restaurant feels like, what a salon's work looks like or why a boutique carries one product instead of another. An agent can parse price and stock; it cannot manufacture a credible history of taste, service and consistency after the fact. Businesses need operational information that machines can read and brand signals that people can recognise.

That is also why "more content" is such a weak goal. Instagram marketing for restaurants, Instagram for salons and ecommerce content all benefit from faster production, but speed only helps when the material remains anchored in the business itself. The more routine work software absorbs, the more visible the remaining human choices become. A recognisable point of view is not an inefficiency to automate away.

The software bill needs a job

The most convincing small-business products this week were easy to explain without talking about model architecture. One answers the phone, another covers a sick shift, another prepares payroll, and others move a print design into a cart or prepare professional work for review. The specificity is the advantage because the owner can see where the time, money or error reduction comes from.

That also gives small businesses a practical buying rule. Start with a recurring job that already has a cost, decide what a good result looks like, identify the point where a person must approve or intervene, then judge the tool on whether it makes that loop materially easier. A general capability can still be useful, but capability only becomes value when it attaches to work. The monthly fee should have a job description.

For vendors, the implication is equally concrete. Products built for small firms need enough business context to act without requiring the owner to explain the company from scratch every time. They need to live close to existing workflows, expose the moments that carry consequence and keep the business's own voice and data intact. That is a much tougher product standard than adding a chat box, which is exactly why the better examples now look less like demos and more like work getting done.

Sources