A sales team given an AI assistant and no configuration uses it for emails and stops there. The same team given six assistants set up for its own recurring work, with the company’s templates, tone, pricing rules and product knowledge built in, changes how it works within a month. The configuration is not technical; it is knowing what the team does all day and encoding it once.

What a configured assistant is

In Microsoft 365 Copilot it is an agent; in ChatGPT it is a custom GPT; in Claude it is a project with instructions and skills. In all three it is the same thing: a standing set of instructions, reference material and, where the platform allows, connected data, so that the assistant starts every task already knowing the house style, the products, the rules and the output format. It is what turns “help me write this” into “produce the proposal in our format for this customer”.

Six for a sales team

1. Proposal drafter. The proposal template, the standard sections, the tone, the compliance wording, and the rules about what may and may not be promised. Input: the discovery notes and the customer. Output: a first draft in the format, for the salesperson to finish.

2. Visit notes to CRM. Takes voice or typed notes from a customer visit and turns them into the structured record the CRM expects, the next actions, and a follow-up email, so the notes keyed into the CRM and never turned into anything become a report.

3. Account brief. Compiles what the business knows about a customer from the systems the assistant is permitted to read, order history, open tickets, recent correspondence, into a one-page brief before a call.

4. Quote and configuration helper. For products with options: recommends configurations, flags compliance or compatibility issues, drafts the quote email. The copilot level of the quote scenario; the person decides.

5. Objection and competitor responses. Grounded in the battle cards and the approved positioning, drafts responses to common objections in the company’s voice, with the sources it used, so nobody improvises a claim.

6. Pipeline summary. Turns the week’s CRM data into the Monday summary in the format the sales director reads, with the changes highlighted.

The pattern for any function

Sample ten to fifteen recurring tasks. For each, collect the template, the examples of good output, the rules, and the data the assistant may read. Configure the assistant, test it on real cases with the people who do the work, and refine the instructions until the output needs finishing rather than rewriting. Record what may be shared with it under the usage policy. Then hand it to the champion to own.

What to keep in mind

Configured assistants are copilots and assistants on the four-levels scale: they draft and execute defined tasks; people decide. Where a task needs the system to act in the CRM or the ERP, approve a discount, or decide on its own, that is an agent, which is a build with evaluation and escalation, not a configuration. The line matters for governance, and a good rollout keeps it clear.

Configuration is part of the tool rollout and the function enablement under team enablement; the agent side is under AI agent development.