AI for Consultants: Where It Pays in a High-Ticket Consultancy

For most firms, the answer is simple. Start in front of the calendar, not inside the work you deliver. AI does the rule-bound, repeat work of a sales front end well. It replies to a comment or DM in minutes. It asks what a partner would ask to qualify a lead. It sends some people a resource and others a booking link. It confirms the call. It follows up when someone does not show.

Inside client work, it is a mixed bag. On some tasks, AI is a fast, good drafter. On others, it is a confident risk. And the line between them is not where people expect.

The firms that win draw that line on purpose, task by task. They keep the diagnosis, the numbers and the advice with the people who sign the work.

Where does AI pay in a consultancy?

Here is the list for a firm that sells through strategy calls. It is ranked by what each use tends to be worth.

Rows 1 and 2 are the sales front end. Most small firms lose money there and never see it. A DM gets answered the next afternoon. A good prospect never gets a reminder. Rows 3 to 6 are inside the work, where the frontier matters.

The jagged frontier: what AI should and shouldn't touch in client work

The best proof here comes from a field test. Harvard Business School researchers ran it with Boston Consulting Group. It was preregistered. That means they set out the plan before they ran it. It used 758 consultants and real-world tasks. Some used GPT-4, an AI model. Some did not.

The authors call the line the jagged technological frontier. On tasks inside it, those with AI did 12.2% more tasks. They were 25.1% faster. And the quality was much better.

Then came a business problem picked to sit just outside the frontier. Here it went the other way. In the paper's words, those using AI were 19% less likely to be right. The HBS AI Institute gives the rates. Without AI, 84.5% got it right. With AI, 70.6% did. With AI plus a short guide to prompts, 60% did. The group with extra training on the tool did worst.

"Jagged" is the key word. The frontier does not follow how hard a task is. Some tasks look hard but sit well inside it. A polished first draft of a long report is one. Some look easy but sit outside. Think of matching the client's numbers into one table. There, the model will give a clean answer. It looks right, and it is wrong.

The figure at the top maps a typical firm's tasks against that line. We drew it from our build work, not from the study.

Here is a rule of thumb for each task. Would a wrong answer look as polished as a right one? Could the client act on it? Then treat the task as outside. It stays outside until proven safe. A consultant checks it line by line. Do you search across your own past work? Then insist on citations to the source file. We explain how in how to make a RAG system cite its sources.

Qualify before the calendar

The scarcest thing in a consultancy is the partner's calendar. A sales call with a poor fit costs an hour. It is the firm's most costly hour. And each slow reply to a good prospect can cost the deal.

Speed still matters. The Harvard Business Review audited 2,241 firms. Some made contact with a lead within an hour. They were nearly seven times as likely to qualify it as firms that waited an hour longer.

The pattern that works gives each new chat two exits. Someone just looking gets a useful resource. They get a follow-up later. Someone with a real need gets the qualifying questions. Then they get a booking link.

We built one for a high-ticket advisory firm. A keyword comment or DM on Instagram starts an AI chat. The AI gets the person's email. Then it offers those two paths. It books strategy calls. And it reaches back out to anyone who cancels or does not show. It all runs across 9 linked workflows.

We estimate a funnel like this is worth $75K to $500K+ a year. That is for a firm that sells through booked calls. The range turns on your traffic and your price. Much of it comes from calls that happen instead of being missed. The model assumes the confirm and re-engage steps cut no-shows. It puts the cut at an estimated 30% to 50%. Will a build like this clear its cost? Use the math in how to calculate AI ROI to test it.

How do you reduce no-shows on discovery calls?

Confirm, remind and make it easy to move the time. That is all it is. Most firms skip it.

The best proof on reminders comes from health care. A Cochrane review looked at 8 trials. With a text reminder, 78.6% came. Without one, 67.8% came. Sales calls are not clinic visits. But people forget in the same way.

The build above sends 3 emails to confirm the call. It assigns an advisor at booking. Say someone cancels or misses the call. Then a set of emails starts to win them back.

Two details matter more than the wording. Send the reminder from the person they will meet. Don't send it from "the team." And give a one-click way to move the call. A moved call is worth far more than a no-show.

The bundling problem, and the platforms' rules

People do not DM in paragraphs. They send "hey." Then "saw your post about exit planning." Then "is this something you do for family businesses?" All of it comes in half a minute. A bot that answers each piece sends three replies to half a thought.

The fix is a short wait. The build above collects quick messages over a 1-minute window. Then it answers the bundle once. The whole question is in view.

The platforms set the outer limits. Ignore them, and a vendor puts your account at risk.

• Instagram. A business can message a person only after they have messaged it. For most actions, the reply window is 24 hours. After a click-to-Direct ad, it can be up to 7 days. A person replying can use the human agent tag for seven days. A public comment can get one private reply. It must go within 7 days of the comment. Follow-ups are allowed only if the person responds. Where the law requires it, a bot must say it is a bot. Meta advises saying so anyway.
• LinkedIn. Its user agreement bans bots that send or redirect messages. It also bans other automated methods it has not allowed. AI can draft a reply for a person to send. It cannot run your LinkedIn inbox.

What does it cost to run?

Our build library is the record we keep of each build. It puts the DM funnel's tools at $160 to $960 a month. That covers the CRM and workflow platform and the model calls. It also covers a small data store, scheduling and alerts.

Model costs are the smaller part. Each DM is short. The bigger costs are the CRM tier and someone's time. Someone reads a sample of chats each week. They update the questions when the offer changes. And they own the set of resources the AI sends.

Inside client work, the cost is mostly review time. Say a drafting tool saves a consultant three hours. It costs a partner one hour of checking. It still pays. But if it saves three and needs three to check, it does not.

Confidentiality and the rules by field

Consulting has no single AI rule book. But the field nearest yours may well have one.

• Investment advice. Are you a registered investment adviser? Then the SEC's Marketing Rule applies. It covers testimonials and claims about returns. And it covers all the AI writes for marketing. The SEC has fined two advisers over claims about their use of AI. The fines came to $400,000 in all.
• Tax. Circular 230 is the IRS rule book for tax pros. It says written advice must rest on reasonable assumptions about the facts and the law. Say you rely on others' work. Then you must take reasonable care to supervise it. An AI draft counts as someone else's work here.
• Work near the law. The ABA is the American Bar Association. Its Formal Opinion 512 came out in July 2024. Some AI tools can learn from what you put in. Before facts about a client's matter go into one, the client must agree. That consent must be informed. Stock wording in an engagement letter is not enough. You may not be a lawyer. But it is the clearest guide to the privacy your clients will expect.
• Marketing claims. The FTC has said AI gets no pass from the laws on the books. And its 2024 rule bans fake reviews and testimonials. That includes ones written by AI.
• Clients in the EU. GDPR is the EU's privacy law. It can apply to a firm outside the EU. All it takes is serving people in the EU. Its Article 22 limits decisions made only by machine. That is for decisions with a big effect on someone.

In practice, client files go only into tools under contract. The contract rules out training on them. And a named consultant reads each draft before it leaves the firm.

What goes wrong

• A polished wrong answer. This is the frontier problem. Check numbers and advice line by line, whoever drafted them.
• The bot answers fragments. There is no bundling window. So one question gets three replies.
• Qualifying that never says no. A funnel that books everyone fills the partner's week. The calls never close.
• Claims no one approved. Say an AI makes up results: "clients typically save six figures." Now your firm is liable for a claim it cannot back up.
• Automation outside the rules. Say a bot chases a silent prospect in DMs after the window closes. That puts the account at risk. So does automating LinkedIn messages.
• Client data in the wrong tool. A consultant pastes a client's financials into a consumer chatbot. The goal was to "tidy the table."

Questions to ask an AI vendor

1. Which tasks in our work do you think sit inside the frontier? How did you test that on our material?
2. What does the qualifier ask? What does it say to someone who is not a fit?
3. How does the system handle quick bursts of messages? How does it keep to the Instagram reply window?
4. What happens when someone cancels or misses the call? Who sends the reminder?
5. Which claims is the AI allowed to make, and who approved that list?
6. Where does client data go? Is it ever used to train a model?
7. Whose accounts do the CRM, the data and the model bills run on?

When not to automate

Low volume. For a DM funnel, our build library's line is about 10 DMs or comments a day. Below that, use a fast human with a few saved replies. It costs less.

No booked-call step. Do you sell without a call first? Then the booking setup is the wrong tool.

No audience yet. A funnel turns attention into calls. It does not create the attention. A brand-new account with a few hundred followers needs content first.

Judgment work. Clients pay a consultant for the diagnosis, the advice and the price. Use AI around them, and keep your name on them.