The invoice you meant to follow up on three weeks ago. The month-end close that swallows a Saturday. The client P&L you re-key out of QuickBooks every quarter because nobody ever built the report. That's the tax on running a small firm with no developer on staff — and it comes straight out of your realization rate.
Describe the work in plain English — "chase every invoice past 30 days, pull each client's P&L on the first of the month, flag anything that doesn't reconcile" — and InTouch AI builds the workflow, runs it on your schedule, and tells you exactly what InTouch AI found. No scripts. No per-task bill. And the client financials never leave your server.
InTouch AI lives on your own computer, not someone else's cloud. Your clients' numbers, files, and logins never leave your office — they never go anywhere, so nobody else can see them. That's the line a cloud service like Zapier or Make can't cross, and it's why a firm handling sensitive financials can actually use this.
Everything below is a real, ready-to-run automation you install with a click and point at your own books — your client list, your ledger, your dollar thresholds. You adjust it to how your practice works and check it before you rely on it. And nothing client-facing goes out on its own: every reminder and statement is yours to review and send.
Nobody enjoys writing the "just following up on invoice #1047" email, so it slides. Receivables age, cash flow tightens, and the follow-up that should have gone out at day 30 goes out at day 75 — if at all. A firm that chases on a schedule collects faster.
The first week of every month is the same ritual: export the reports, build the client packs, reconcile, email each one. Hours of repetitive assembly that looks identical to last month — because it is. It's the definition of work a machine should do.
Twelve QuickBooks companies, three on Xero, a couple still on spreadsheets. Pulling a consistent P&L or AR-aging across the book means opening each one by hand. There's no single view because nobody's had time to build it.
1099s in January, 941s quarterly, sales-tax filings by state, estimated payments, extensions. Miss one and it's a penalty and an awkward client call. The calendar is only as good as the person who remembered to put it there.
Describe the question, get a structured answer. Paste the input, read the result. Nothing to configure beyond your AI provider credential — or point it at a local model so nothing leaves the office at all.
Extracts structured data from invoice emails and PDF attachments — vendor, amount, line items, due date, payment terms — and flags unusual amounts or vendors it hasn't seen before. The keying step on your bookkeeping clients, done first so a person is checking rather than typing.
Inspects a CSV's headers and sample rows for schema problems, missing required fields, encoding issues and obvious data-quality red flags before it loads. Run it on a client's trial balance or transaction export and find the broken column now, not after it is in the ledger.
Scheduled or trigger-driven. Each one reads a source you already have — a Google Sheet you maintain, a folder your reports land in, or an export from QuickBooks or Xero — does the work, and drops a digest in the channel you live in. It runs whether the first week of the month is busy or not.
Reads a FilingDeadlines Sheet (1099s, 941s, sales tax by state, estimates, extensions) and sorts what is open into 30/14/7/3/1-day windows, flagging anything already past due. It does not calculate due dates — you enter the date your firm determined, and this is the second set of eyes on it.
Ages every document you are still waiting on, groups it by client worst-first, and tiers it into nudge, follow-up or partner call. For anything past the nudge line it drafts the request naming the specific missing items. The draft comes to you — it never emails a client on its own.
Checks each client's pack folder against the list of reports you expect, and tells you which packs are incomplete, what is missing from each, who owns it, and which are already late. A presence-and-naming check — it does not open the files, so it catches the missing schedule at 7am rather than in partner review.
Compares a GL export against a subledger or bank export, flags every account differing by more than your threshold, and separates accounts present in only one source. Then the AI writes the covering note. Every figure is computed in code before the model is called, so it describes numbers rather than producing them.
Chasing receivables too? Invoice Reminder on the Hub reads an invoice sheet and sends escalating reminders — friendly, follow-up, final — stamping each row so nobody gets chased twice. Note that one does email the client directly, unlike the four above; configure it deliberately.
Grab the free Personal edition of InTouch AI and run it on the office server or a partner's laptop. Browse the InTouch Hub, pick the skills and workflows you want, click Install. No scripts to write, no consultant to hire.
Each workflow ships with placeholders — client list, folder paths, ledger connection, dollar thresholds. Read the README, fill them in, point it at your exports or database. Verify a run does exactly what you expect before you schedule it.
Attach a daily, weekly, or first-of-month schedule and walk away. InTouch AI runs it, pushes the pack or worklist to your channel — email, Slack, a folder — and logs every run. When an export format shifts or a token expires, InTouch AI reads the failure, fixes what InTouch AI can, and tells you why. As a workflow earns your trust, dial it from AI-driven toward fully deterministic: same result every time, zero per-run AI cost, fully audited.
Straight talk. No surprises after you install:
Personal edition is free. Install it on the office computer, point it at your exports or ledger, and let it chase the receivables, build the month-end packs, and watch the filing calendar for you. And because it all runs on your own computer, your client financials never leave your office — nobody else can see them. That's the one thing a cloud tool can't promise a firm.
Every one installs from the hub and runs on your own server. Browse all over a thousand connectors, or describe what you need and the assistant builds it.