The $2.7K-a-Day AI Business: The Support-Agent Model

A $2,700-a-day revenue sounds like a dream, yet the math behind it is surprisingly plain: set up an AI support agent for a small business for a $2,000-3,000 setup fee, then collect $500-1,000 a month in care fees. The person doing the setup needs no technical background; the system reads documents, answers questions, and writes its own report at month's end.

At the center of the story stands Aaron Edwards. While still employed, he tried eight different AI ideas in six months; the first involved visuals and none of them stuck. The ninth idea, building support agents for businesses, took hold three and a half years ago and has grown since. Today he has no office, three people beside him, and yearly earnings past the $1 million mark. Even the first client signed by his 19-year-old son calculated the service would save $100,000 a year.

The product is called Docspot and lives at docsb.ai. The logic runs like this: a firm's scattered files, help pages, and internal notes gather in one knowledge pool, and the AI answers customer questions by feeding on that pool. The technique is giving generative models context alongside data, so instead of inventing facts the model finds the matching passage in the documents. The inside of that method was shared openly in the conversation recorded in the LifterLMS podcast.

The scale side of the story runs deeper. According to the Weaviate case study, Edwards started alone; a single post by a Japanese influencer drew 2.3 million views and hundreds of sign-ups poured in overnight. When the first version buckled under the load, the infrastructure was rebuilt; today tens of thousands of isolated client pools run on the same system and the platform answers 6.1 million questions a year. One founder carrying that load became possible by choosing ready infrastructure.

Now for the unit numbers. The price tiers run like this: $50 for starters, $150 for the standard business, $500 for larger self-serve setups, plus enterprise plans for custom jobs. Average revenue per user sits near $130, meaning most clients cluster in the middle. Lifetime value clears $2,000 while monthly revenue loss hovers near 3%. A free plan exists too, but the rule from day one stays firm: seriousness is not measured without a card on file, and entry is paid with a refund guarantee.

What does that 3% say? The Optif study treats 3-5% monthly loss as normal at small-business scale, 1.5-3% for the middle tier, and 1-2% for large accounts. So the Docspot line sits on the good side of the band. The study also warns that multiplying a monthly figure by twelve misstates yearly loss; the honest math compounds. At 3%, in a world where small clients walk away fast, the signal reads solid product plus steady attention.

Agency window prices complete the picture. The Taskip guide puts small-business care fees at $500-1,500 a month, with one-time setups in the $1,500-5,000 band. The guide lists six pricing paths: care agreements, project fees, performance-based, hybrid, productized packs, and value-based. For sellers targeting small firms the advice stays crisp: one workflow, a fixed price, a predictable outcome.

The whole market looks even hungrier. The ArticSledge analysis sizes the AI agency market at $7.63 billion in 2025, heading for $50.31 billion by 2030 at 45.8% yearly compounding. Typical setups run $2,500-15,000 with monthly care at $500-5,000, and gross margins drift in the 70-90% band. Consulting and model-built custom jobs rank as the most lucrative lines.

But is demand real on the buyer side? The CapsuleCRM roundup lines up several surveys: generative AI use at 58% among Chamber members, two-thirds of Thryv users reporting $500-2,000 in monthly savings, and 76% usage among Goldman Sachs participants. Client communication ranks among the top three uses. So the business being pitched has usually heard the story before; most have tried, grown tired, and now wait for results.

Prices on the rivalry front stay transparent. On the Intercom side, the Fin agent charges $0.99 for every ticket it resolves. The Tech Insider review notes independent tests landing at 38% on average, with a realistic 30-55% band for small-business setups. Sierra opens the enterprise door at $150,000 a year, while Zendesk asks $1.50-2 per verified resolution. The seller's pitch writes itself: rates move with documents, costs stay per ticket.

At the other end of the scale sits messaging. The TailorTalk comparison finds well-built agents in 2026 handling 65-83% of questions automatically, answers dropping under a minute, and support spending falling by up to 80%. For a small business waiting on WhatsApp and Instagram lines, that picture means a bigger shop window without hiring. The comparison adds that each platform cures a different pain: some play for conversation, others for the knowledge pool.

Finding clients is the part everyone asks about. The ranking shared runs like this: face-to-face meetings at chambers and fairs bring the strongest leads, phone calls come second, and cold outreach on LinkedIn third. The common thread is niche choice: winners know one sector well rather than chasing everyone. First jobs usually arrive through friends and family, then the referral chain starts.

Behind the sale runs the partner arrangement. The agency delivers a white-glove setup, then walks the client through 90 days of close onboarding. Partners then demo the agents to small businesses, sell them, and share the revenue. The program opens through docsb.ai, with a $1,000 referral reward per client brought in. So the seller never stands alone; the product owner shares the selling.

The real charm lands in the inbox at month's end. The AI scans the client's logs, lists answered and unanswered questions, and mails the business a neat logo-bearing PDF. The seller may have touched nothing all month, yet the $500 still reads earned. The report also flags missing documents: write a help note on this topic and your agent will answer better. Value gets documented monthly, leaving no excuse to cancel.

The other side exists too: after a while, clients want to build it themselves. Early on 1% of cancellations cited that reason; the share has since climbed to 13%. Small accounts leave while large ones stay and keep paying, and whoever carries maintenance also shoulders updates and document order. Hence the seriousness filter from the start: no tire-kickers on free trials, only clients who pay their way in.

Model choice is the technical heart of the tradeoff. The advice weighs cost against benefit: not the largest model for every job, but the model each job deserves. Current models get tested on support bots, and whichever proves accurate and cheap gets picked. The names GPT and Claude surface in that context. Since models keep shifting, the rule stays fixed: watch the newest, mind the bill, keep documents fresh.

Voice raises the bar further. The Open team's Realtime guide prints the numbers openly: $0.06 per minute of audio in, $0.24 per minute out, so a typical 2-3 minute support call costs the model $0.30-0.60 before telephony. The target is latency that never feels above 2 seconds. The maker's gpt-realtime release cut prices 20%. Voice agents simply cost more than text; minutes must be priced into every sale.

The close brings a fairground tale that reads like a lesson. Instead of paying $25,000 for a booth, the team had AI draft 3D-print models for tiny shoulder-sitting robots at $50 each. Wired to the voice agent, the robots crack jokes and pitch the business while the team walks the floor, turning every head in the hall. The message lands clean: memorable demos beat pricey showcases. This trade rewards attention and trust, not deep pockets.