AI in the CRM

A week with the Assistant at a consultancy

The Assistant looks things up, cross-checks figures and prepares the work; the partner reviews each change and decides what gets approved.

At a marble table in a café in Seville, a grey-haired partner in a green jacket checks her laptop while a consultant, coffee in hand, points to a line in a notebook.

An agentic CRM for consultancies lets the Assistant look up data, cross-check figures and prepare the week's work (an opportunity, a proposal, a project with its phases) while the partner decides what gets approved. In AgentticCRM the Assistant works with the name and permissions of the person writing to it, shows each change before making it and runs nothing until that person approves. An agency works the same way: the engagements change, and the split between what the Assistant prepares and what a person decides stays put.

Picture Carmen, a partner at a six-person consultancy that runs process diagnostics for mid-sized companies. This week she has two engagements under way, a proposal to write and a client who pays late. She has no personal assistant, and a good part of her Monday used to go into opening screens to see where each thing stood.

What an agentic CRM does in a consultancy's week

The Assistant has its own screen inside AgentticCRM, and you write to it the way you'd write to a colleague. When you ask for a piece of data, it looks it up and answers on its own, because a lookup changes nothing. When you ask for something that modifies a record, it shows you a card with the steps it would take, the data for each one and what it won't do. You can drop steps or correct fields right there, and nothing happens until you approve. If you leave the proposal pending, it expires on its own.

Before every action it reads your role and permissions again, so it can't reach anything you wouldn't see by opening the screen with your own user. And every action it carries out records who asked for it, who approved it and what changed.

Monday: where each client stands without opening ten screens

Carmen starts by asking which invoices are still unpaid, which open tickets each client has and which meetings are on this week. The Assistant answers with the data and a link to the record it came from, so she only opens the two cases that worry her.

One of the tickets is a client's question about the interview schedule. Carmen asks the Assistant to draft the reply, and it sits ready in the ticket for her to read, adjust and send. Her conversation with the Assistant is hers alone: nobody else on the team sees it.

Tuesday: a new opportunity and its proposal

A transport company asks for a diagnostic of its invoicing process. Before quoting, Carmen asks how the consultancy prices diagnostics of that size, and the Assistant looks for the answer in the firm's knowledge base, among the documents the team uploaded.

Then she asks it to open the opportunity and prepare the proposal with services from the catalog. The card shows the steps: the opportunity with its stage and the draft proposal with its lines. Carmen corrects the price on one line and approves. Afterwards she reviews the PDF in Proposals and sends it herself, because scope and price are the partner's decisions.

Wednesday: the accepted engagement, turned into a project

The other client accepted its proposal last week from the portal. Carmen asks the Assistant to prepare the project with the phases of her method (Diagnostic, Interviews, Analysis, Final report) and a task with an owner and a date for each consultant. On the card she changes one owner, because one of her consultants is going on vacation, and approves.

From there the engagement lives in Projects: each consultant logs hours on the task where they worked, and Carmen compares what's been used against the budget when she reviews the project. We explain how to run that tracking in a consulting engagement, hour by hour.

Thursday: invoiced against collected

The client who pays late has two overdue invoices. Carmen asks the Assistant to compare what was invoiced with what was collected for that client and to tell her which projects carry an overdue invoice. The figures come from two different places in AgentticCRM, and the Assistant puts them together in one answer, with a link to each invoice.

Carmen decides to call the client's finance manager herself. She asks the Assistant for a task for Friday, reviews it on the card and approves it. She writes the follow-up email with the email writing assistance, which prepares a draft for her to correct and send.

Friday: reviewing what changed during the week

Before closing the week, Carmen opens the activity log and filters by action type to see the Assistant's actions: what was proposed, what was approved or rejected and what was reverted, each next to the person who asked for it. She sees that a task she approved on Thursday ended up with the wrong date and puts it back the way it was from Agent activity. When an action can't be walked back, such as an invoice already issued or a corrected payment, the screen says so and offers the correction that applies.

If someone on the team works from a connected ChatGPT or Claude, their actions show up in the same log under their name, with the agent's name in the detail.

What the partner doesn't delegate

The price and scope of each proposal, what goes to the client and when, who runs each engagement and the collection call remain Carmen's work; the Assistant saves her the searching and the assembling. The passwords the team keeps in AgentticVault are out of its reach.

The consultancy brings the artificial intelligence: it connects the account of the provider it prefers, sees what it spends and changes it whenever it wants. The AI CRM page covers the rest of what AI does inside AgentticCRM.

Frequently asked questions

Can the Assistant send a proposal or an email to the client on its own?

No. Inside AgentticCRM it proposes each change and a person approves it, and the writing assistance leaves the email as a draft for you to send. If you connect an external agent, such as ChatGPT or Claude, sending emails, proposals, invoices or reminders also requires approval.

Do my consultants see what I ask the Assistant?

No. Each conversation belongs to whoever opened it, and only that person reads it. What the team shares is the record of the actions that were actually carried out.

Do I have to pay for the artificial intelligence separately?

Yes. Your organization connects its provider's key: OpenAI, Anthropic, Google Gemini, OpenRouter, DeepSeek, GLM from Z.ai, Kimi from Moonshot or Huawei Cloud MaaS. The key is stored encrypted and never goes back to the browser.

Does it work the same way for an agency?

Yes. In an agency the week has campaigns and client approvals instead of interviews, and the Assistant prepares the same things: opportunities, proposals, projects and tasks that a person reviews and approves. We explain what makes a CRM agentic in what the agent does and what you decide.

Request a demo using a real case from your business.

We look together at where information gets lost today, which parts AgentticCRM can join up, and which ones are not its job.

Prefer to explore on your own? Work out your current costs, see how the Assistant works or use the CRM evaluation guide.