How to Build AI Agents for Consulting

Consulting solo often means juggling prospect research, proposal writing, and follow‑ups while trying to bill enough hours. After reading this guide you’ll be able to spin up a lightweight AI agent that pulls lead data, drafts a tailored proposal, and logs time entries — all without writing a line of code.

What most guides get wrong about AI consulting agents

Most tutorials start with a vague promise to “automate everything” and then dump a list of generic prompts. They ignore the fact that consultants need traceable outputs, not just clever text. I’ve seen guides suggest feeding raw LinkedIn URLs into a model and expecting a perfect proposal — that rarely works because the model hallucinates missing details. The real mistake is treating the AI as a black box instead of a configurable pipeline.

Core pieces you actually need

You need three things: a data source, a language model, and a workflow orchestrator. For data, a simple Google Sheet works fine for storing leads and call notes. For the model, Claude 3.5 offers a large context window and decent pricing. For orchestration, Make (formerly Integromat) lets you connect sheets, HTTP calls, and the model API without code. Those three pieces are enough to build a functional consultant agent.

How to debug when the agent gives wrong outputs

When the proposal sounds off, first check the raw data that entered the model. In Make, add a module that logs the incoming JSON to a text file or a second sheet. If the data looks fine, look at the prompt you sent. I often find that a stray line break or an extra space changes the model’s interpretation. Finally, verify the model’s temperature setting; a high temperature can cause wild guesses. Keeping a debug log cuts troubleshooting time from hours to minutes.

Why does the agent hallucinate numbers in proposals?

Hallucinated numbers usually come from two places: missing source data and over‑creative model settings. If the lead’s budget field is empty, the model will invent a number to fill the gap. The fix is to add a validation step that blocks the proposal generation when any required field is blank. In Make, use a router that checks each field and stops the scenario if something is missing. Lowering the temperature to 0.2 also reduces creative filling.

Putting it together: a real example with Make, Claude, and Google Sheets

Here’s a concrete scenario: you receive a new lead via a webhook from your website form. The webhook drops the lead’s name, email, and company into a Google Sheet. A Make scenario watches that sheet for new rows. When a row appears, it pulls the company’s website, scrapes the About page via a simple HTTP request, and sends the text plus your service description to Claude with a prompt that asks for a one‑page proposal. Claude returns the proposal, which Make writes back to the sheet and emails to you as a PDF attachment.

First‑time tool mentions are bolded below.

1. Create a Google Sheet with columns: Timestamp, Name, Email, Company, Website, RawAbout, Proposal, Status.
2. In Make, add a Webhooks module to capture the form POST and write the basic fields to the sheet.
3. Add a Google Sheets > Watch Changes module that triggers on new rows.
4. Add an HTTP > Get a file module to fetch the company’s homepage (use the Company column to build a URL).
5. Add a Text > Extract module to pull the first 1500 characters of the HTML body (you can strip tags with a simple regex).
6. Add a Claude > Create a completion module. Prompt example:
   
   You are a senior consultant. Using the following company description:

   {{RawAbout}}

   and our service offering: "We help B2B SaaS firms cut churn by 20% through targeted onboarding."

   Write a one‑page proposal that includes: problem statement, suggested solution, expected outcome, and rough timeline. Keep the tone professional and concise.

   Temperature: 0.2
Max tokens: 800


7. Add a Google Sheets > Update a row module to store the returned proposal in the Proposal column.
8. Add an Email > Send email module to send yourself the proposal as a PDF (use Make’s PDF converter).
9. Set the scenario to run continuously and enable error handling that emails you if any module fails.

This flow takes about three minutes to set up and runs automatically for each new lead.

Price check: what you’ll actually spend per month

Here’s what I pay for a solo consulting setup: Google Workspace (for Sheets and Gmail) $6/mo, Make core plan $29/mo (allows 10k operations and webhook access), Claude 3.5 API $0.008 per 1k tokens — roughly $4/mo for ~500k tokens used in proposals and scraping. Total comes to about $39/mo. I think $29/mo for Make’s core plan is fair; $199/mo for the enterprise tier is ridiculous for what you get. The free tier of Make is a joke for anything beyond testing because it caps webhook payloads at 4 KB, which I ran into when trying to pull full HTML pages.

Who should skip this and hire a freelancer instead

If you need multi‑step legal review, custom branding on every proposal, or integration with a legacy CRM that lacks an API, building this yourself will eat more time than it saves. In those cases, a freelancer who knows Make and Claude can spin up a tailored system in a day for under $500. For most solo consultants selling standardized services, however, the DIY approach pays off in the first month.

Adjacent reading: deeper coverage of AI agent platforms.

If you’d rather skip the build and deploy a working version in an afternoon, we’ve packaged this workflow as a blueprint at deepusecase.com/vault/ai-agent-builder-kit.