AI for Professional Services: Start With Prospects

AI for professional services usually means tools for the inside of a firm: research, drafting, reconciliations, document review. The front of the firm gets less attention, yet that is where a prospect decides whether to call. They read your services page, look for fees, check who you are, and leave when a simple question goes unanswered while you are in a meeting. Here is what an AI agent can answer on an accounting or advisory firm's website, what it hands to a person, and a live intake check you can ask for in one message.

What does AI for professional services look like on your website?

An AI agent for professional services answers prospective clients on the firm's website from the firm's own pages, says plainly that it is an AI and gives no advice, and moves anyone who wants help with their own situation to a first meeting with a person.

Ultimo Bots is an AI agent you train on your website, files and Q&A pairs: it answers visitors in the chat, sends them to a first meeting, forwards requests to your team with the whole conversation and lets a colleague take over when someone is online. It installs on Wix, WordPress, Webflow or any site that takes a short snippet, and the same agent can answer Messenger, Instagram and WhatsApp.

The first two make your team faster. The third decides whether a stranger becomes a client, and it is the one this guide covers, with AI for accounting firms as the worked example.

Tuesday, 11:20 am: both partners are busy

Take Oriel & Finch, a fictional accounting and advisory firm in Madison, Wisconsin: two partners and three staff. Dana Oriel runs the website. On a Tuesday in late January she is two hours into a year-end review with a client, Marcus Finch is at a client's warehouse, and nobody has switched online in the portal.

Luca, who runs a landscaping company with six employees, opens the chat on the firm's services page: "Do you do bookkeeping and payroll for a company my size? And can I write off the new truck this year?"

The agent replies: "I'm Oriel & Finch's AI assistant. Yes, we do monthly bookkeeping and payroll for small businesses, and bookkeeping starts at $400 a month. Whether the truck is deductible depends on your own numbers, so that is a question for one of our advisors, not for me. Would a short intro call help?"

Luca says yes. The agent asks what he needs, recommends the 30-minute small business intro over the 60-minute advisory meeting and, once he agrees, gives him a personal Calendly link; he picks the time on Calendly's page. Each part of that reply has a source: the services and the $400 come from the firm's pages, the AI line and the refusal to judge the truck from two behaviors, the link from a ready-made skill.

Which checks do prospects run before they hire a firm?

Prospects arrive with a checklist, and the IRS publishes one. Its Topic No. 254 on choosing a tax return preparer (updated October 7, 2026, read October 29, 2026) tells taxpayers to make sure a preparer has a PTIN, to avoid preparers whose fee is a share of the refund and to consider whether the firm will still be around later to answer questions. It notes that only attorneys, CPAs and enrolled agents can represent a client before the IRS in all matters.

So write the answers down: each person's credential on the team page, how you bill and the starting prices you are willing to publish on the fees page, who you work with on the services page, and the next steps, including what happens in a first meeting, on the process page. The same goes for AI for consulting firms, architects or engineering practices, with sectors and project starts in place of credentials.

The agent reads the text on your pages, images only by their alt text, embedded frames (iframes) not at all. A fee table posted as a picture is invisible to it, so put prices in text or upload a text PDF to the knowledge base. It answers from your content, not from guesses about your business, and says so when a question goes past what you published.

Advice, documents and the yes stay with people

An agent that sounds like an advisor is a liability for a firm. Three things never belong in the chat, and each gets a behavior, a short rule the agent follows in every conversation. In the AI Studio, open the Behaviors tab of the "Your agent" panel, click Add Behaviors, paste the text and click Add. Each holds up to 250 characters and is stored word for word.

1. No advice. "Never give tax, accounting, legal or investment advice about the visitor's own situation. Say you are an AI assistant and cannot advise, describe the service that fits and offer a first meeting with an advisor."
2. No documents or numbers. "Never ask for tax documents, Social Security numbers, account numbers or income figures. If someone starts to share them, ask them to stop and say the team requests documents after an engagement starts."
3. No engagement promises. "Never confirm that we will take someone on as a client or quote a final fee. Give only the fee ranges on our website and say a partner decides after a first conversation."

The website chat has no file upload, so nobody can drop a W-2 into it; the second rule covers what people type. Make the disclosure rule the fourth behavior. The website chat already shows "You are chatting with an AI assistant" above the first message, but Messenger, Instagram and WhatsApp show no such notice:

Existing clients ask too: "Is my return filed yet?" The agent cannot see client files and says so, then offers once to pass the question on. Only after a yes does it ask for the contact details it still needs, sum up the request in one sentence and forward it to Leads with the conversation, which emails you by default. It does not read the details back, so an email address reaches you as typed.

From the chat to a person: meetings and live chat

A firm sells meetings before it sells engagements. Connect Calendly on the Apps page by signing in (Mo shows the button if it is missing) and add the ready-made skill Recommend the right meeting. It asks one or two short questions, recommends the meeting type that fits, with name and duration, and gives a personal booking link. The visitor picks the time on Calendly's page; the agent never claims it booked anything. With Cal.com, the agent can book inside the chat after the visitor confirms. Both are app actions included on every plan.

A chat that ends with a Calendly link records no lead in Ultimo Bots; the meeting is your record. During the day, a colleague can also take over: when a staff accountant switches online in the portal, a "Talk to a human" button appears in the website chat, and the agent hands over when a visitor asks for a person. The colleague sees the whole conversation, and the AI stops replying until they leave (live chat).

Build it with Mo: open intake from your practice system

"Are you taking new clients?" is the question your pages answer worst. In March the honest answer for individual returns may be a waitlist, while bookkeeping is open all year. If your practice software or your own system holds that status and has a public HTTPS API that takes an API key or bearer token, Mo can build the lookup. Dana tells Mo, the builder in the AI Studio:

When someone asks whether we take new clients for a service, or what it costs, look that service up in our practice system's API and answer with its intake status and published fee range. Never look up a client or their files.

Mo reads the system's public API docs, or the part Dana pastes in, and asks for the key in a secure field; the key is stored encrypted and never sent to the AI. Mo builds the skill, tests the lookup on the firm's service list and reports what the agent can now answer. From then on, "Are you taking new tax return clients?" gets the status your practice system shows at that moment.

The limits are part of the design. Calls to your own system cannot check who is asking, so the skill reads only firm-wide facts such as intake and published fees, never a client's return or balance. Each call has 20 seconds to answer, and systems behind a firewall, or ones that offer only a sign-in instead of an API key or bearer token, are out of reach. Within that, you build what your firm needs by describing it.

Will AI replace professional services firms?

No. AI takes over parts of the work, and on your website it takes over first questions, but prospects hire people. The IRS asks taxpayers to consider whether a firm will still be around to answer questions later; no agent answers that for you. Firms that do well with AI tools for professional services firms keep judgment where it is: advice, pricing an engagement and saying yes to a client stay with a partner. The agent makes sure the partner gets that conversation, instead of a competitor whose website answered first. Law firms draw the same line under their own rules; our guide for law firms covers it.

What it costs a small firm

Ultimo Bots is priced per agent in AI credits, and teammates are free, so two partners and three staff pay what a sole practitioner pays. As of October 29, 2026:

Smart includes 3 skills and 5 behaviors. Boost is $59 a month, or $39 billed yearly, with 2,000 AI credits, 10 skills and 15 behaviors. Ultimo is $249 a month, or $199 billed yearly, with 10,000 AI credits, 30 skills and 50 behaviors.

A standard answer uses 1 AI credit; the higher answer levels use 4 or 8. Mo works from the same balance: a quick question uses a few credits, building a skill more. Oriel & Finch fits on Smart: the intake lookup and the Calendly skill are two of three skills (the built-in lead skill does not count), and four behaviors leave one free. On Wix, billing runs through the Wix App Market, where prices can differ. Details on the pricing page.

Test it the way a prospect would

Before the agent goes live, read your fees or services page as Luca would. If it does not say who you work with, how you bill and how a first meeting works, write those three paragraphs in text. Then add the website chat, point the agent at your site and ask it three things in the AI Studio's Test tab: a service question, a fee question and "Should I set up an S corp?" The first two should come back with your facts, the third with a polite no and a meeting. Test-tab chats are real, so a test that forwards a request sends you a real lead email.

Questions and answers