Why start with the back office, not the front?
Front-of-house AI gets the attention — chatbots, receptionists, sales assistants. But when we audit a service business, the hours actually bleeding out are in the back office: chasing invoices, sorting the inbox, compiling the same weekly numbers, remembering to close an ex-employee's accounts. None of it grows the business. All of it has to happen. That makes it the safest, highest-certainty place to put an agent, because the work is repetitive, rule-bound, and already written down somewhere — even if only in the owner's head.
One thing before the list: every agent below has a hard boundary, a thing it deliberately does not do. That's not a limitation we apologize for — it's the design. An agent that drafts but never sends, reports but never remediates, reminds but never threatens, is an agent you can actually leave running while you do your real job. Vendors who promise full autonomy on day one are selling you their risk tolerance, not yours.
Who chases the money? The AR Agent
Invoice follow-up is the task owners postpone the longest, because it's awkward. The AR Agent runs it on your tone rules: polite reminders before due dates, firmer sequences after, payment links attached, every touch and promise-to-pay logged in your books. It plugs into QuickBooks, Xero, and Stripe, so the follow-up starts from your actual receivables, not a spreadsheet you have to maintain.
The boundary: it never negotiates, never threatens, never makes a collections decision. Disputes, hardship cases, and accounts past your escalation threshold go to a human with the full history attached. Who benefits most: any business that invoices on terms — trades, agencies, clinics, consultants — where the gap between work done and cash in is measured in weeks and the follow-up depends on whoever remembers.
What handles the inbox and the calendar?
The Inbox Triage Agent sorts what arrives — urgent, routine, junk, opportunity — and prepares draft replies in your voice for the messages that need one. The boundary is absolute: drafts only, never auto-sends. Every outgoing word passes through a human eye first. That single rule is why you can trust it with your actual inbox instead of a sandbox: the worst case is a bad draft you delete, not a bad email your client received.
The Scheduling Agent is its natural pair: booking, rescheduling, reminders, and the back-and-forth of finding a time — the email ping-pong that turns a two-minute decision into a two-day thread. Together they suit anyone whose day is fragmented by their own inbox: owner-operators answering email at 10 p.m., office managers who are the de facto dispatcher, and small teams where 'who's replying to this?' has no answer.
Who does the security chores nobody does?
Most small-business breaches don't come from sophisticated attacks. They come from the boring controls nobody ran: the ex-employee whose accounts stayed open, the phishing email someone clicked because there was nowhere to forward it, the vendor breach nobody noticed. The Security Agent runs that routine on schedule — quarterly access reviews that flag stale accounts, triage of suspicious emails your team forwards (with a plain-language verdict back), password and MFA hygiene nudges, and a watch on breach disclosures for the tools you actually use.
The boundary here matters more than anywhere else in the catalog: it never remediates on its own. Read access, not admin control. It reports, recommends, and waits — revoking access, resetting credentials, and incident response are human decisions, made with the agent's evidence in hand. Anything that looks like a live incident escalates immediately. Who benefits most: any team on Google Workspace or Microsoft 365 with no one whose actual job is security — which is nearly every business under fifty people.
Who tells you the truth about the business — and about what clients say?
The Reporting Agent compiles the weekly digest you keep meaning to build: revenue, pipeline, jobs completed, receivables aging — pulled from your actual systems into one readable email. Its boundary is the one that makes the digest worth reading: it never invents figures. If a number can't be pulled from a source system, the digest says so and shows a gap, rather than papering over it with an estimate. A report you can't trust is worse than no report — you make decisions on it.
The Review & Reputation Agent watches what clients say publicly, drafts responses for your approval, and asks happy customers for reviews at the right moment. Its boundary: it never fakes a review and never gates them — no filtering happy customers toward Google while quietly diverting unhappy ones, a practice that violates platform rules and, more to the point, poisons the feedback you most need to hear. Who benefits most from this pair: owners flying blind between bank-balance checks, and local businesses — clinics, restaurants, trades — where the Google profile is the storefront.
In what order should you deploy them?
Not all six at once. The pattern that works: start where the pain is measurable in dollars — usually AR if you invoice on terms, Scheduling or Inbox Triage if your bottleneck is time. Run one agent for a month, learn what supervising an agent feels like, then add the next. Reporting and Security are the quiet compounders: neither shows a dramatic first-week win, but six months in, they're the ones owners say they'd keep if forced to choose. Each agent is productized — setup from CAD $2,500, plus Care from $495/mo for the monitoring and tuning that keeps it working as your business changes.
If you're not sure which pain ranks first, that's exactly what our free audit is for: we map where your hours and dollars actually leak, then rank what to automate first. Sometimes the answer is one agent. Sometimes the honest answer is none yet — fix the process before you automate it. You keep the analysis either way.