Italiano 
The 6 pm enquiry: where sales teams lose time
On an ordinary day, a sales office receives requests from the website form, emails about products, quote requests with attachments and replies to a campaign. Someone has to read them, work out who sent them and how urgent they are, find the right information and write a reply. If the request lands at 6 pm, all of that often slips to the next day.
Time is rarely lost in the actual negotiation, which remains a job for people. It is lost in the steps around it: copying details from an email into the CRM, looking for the last offer sent to that customer, piecing together the history before a call, remembering to get back to someone who asked to be contacted again in two weeks. These tasks are repetitive, but they involve reading and understanding text, which is exactly where artificial intelligence has become useful.
If you have not yet decided where to start, the article on how to choose the first process to improve with AI offers a simple method. Here we take the next step and look at what happens when that process belongs to marketing and sales.
Model, agent, integration: what changes in practice
Three words come up again and again and lead to very different projects, so it is worth telling them apart.
The model is the engine that understands and writes text. On its own it does not know your customers, your price lists or your rules: it answers with what it has learned in general.
The agent is a system built on top of the model to carry out a specific task. It has instructions, a tone of voice, sources to consult and clear limits: it knows, for example, that it must answer only from approved product sheets and that any request about a discount must be passed to a salesperson. It can simply make suggestions or carry out controlled actions, with confirmations and a log of what it has done. Custom AI agents for businesses are designed in exactly this way, starting from the processes, documents and roles of the people who will use them.
The integration is the connection between the agent and the tools you already use: CRM, email, website form, management software, document archive. It is what allows the agent to read a customer's history or create a reminder without anyone copying and pasting data.
And the chatbot? It is the conversational interface, the window where you type and read the answer. It can front a simple auto-responder or an agent connected to company systems: the difference is not the window but what sits behind it. To see how these pieces come together in a project, the overview of AI services for businesses describes the path from use case to adoption.
In sales: requests, CRM, quotes and follow-ups
Back to the 6 pm enquiry. With an agent connected to the website form and the sales inbox, the request is read as soon as it arrives. The agent classifies it (information, quote, support, job application), recognises whether the sender is already a customer, assigns a priority and prepares a draft reply based on approved information. The salesperson finds the request already sorted, with a reply to review rather than write from scratch.
Before a call, the agent can summarise the customer's history: recent emails, offers sent, meeting notes, open requests. Today this means opening several tools, and it is often skipped for lack of time.
With quotes, the most realistic help is in the preparation. The agent can read a request with attachments, pick out quantities, deadlines and requirements, and set up a draft from approved templates and price lists. Checking prices, terms and discounts, however, stays with the salesperson, a point we will come back to shortly.
Finally, the follow-up, meaning getting back in touch after a first exchange. Connected to the CRM, the agent can notice that an offer has been sitting idle for a while, create a reminder and suggest the text of the message. How the connection with CRM, email and management software works is explained on the page about integrating AI into business processes.
In marketing: content, communication and reports
For marketing, the main value is speed on the first draft, always with a person reviewing it. An agent briefed on the company's tone, on the terms to use and those to avoid, can prepare drafts of posts, newsletters, product descriptions or campaign copy, adapting the same content to different channels.
A word of caution on website content. AI helps produce drafts and organise topics, but the quality of a business blog that brings visits and enquiries still depends on expertise, real examples and consistency. The way search engines and AI assistants read web pages is also changing, as explained in the article on AI and site indexing.
Then there are reports. Campaign, contact and sales data are often spread across several tools. An agent with authorised access to these sources can prepare periodic summaries, compare different periods and flag anomalies, leaving managers time to interpret the figures rather than assemble them. Event organisers know this benefit well: we covered it when discussing event secretariats and artificial intelligence and, more broadly, how AI is changing event organisation.
What not to hand over to an agent without control
A useful agent is not one that does everything on its own. In marketing and sales some decisions must stay with people: prices and discounts, contract terms, delivery commitments, sensitive messages to key customers. In these cases the agent prepares and the person decides.
That is why a well-run project defines from the start what the agent can read, what it can change and when it must ask for confirmation. Roles, permissions and an action log are not technical details: they are what allows the team to trust the tool over time.
A practical tip. Before connecting an agent to your CRM and email, decide which data it can use and with which tools. The article on which company data to share with an AI assistant offers a simple classification to start from. For privacy and data protection matters, always check with your advisor.
Source quality matters too: if the price list in the archive is out of date, so will the draft quote be. In the first few months, someone needs to check the answers and flag what needs fixing.
How to launch a pilot in 5 steps
The most cautious way to start is a pilot, a trial on a narrow workflow with clear goals and a set time frame.
- Choose a small, frequent workflowFor example, sorting requests from the website form or summarising customer history before calls. Pick a task that happens every day and currently involves obvious manual steps.
- Gather the sources the agent will useProduct sheets, price lists, approved answers, offer templates, tone of voice guidelines. You do not need a perfect archive: start from what exists and assess what is already usable.
- Decide what it can do and what a person must approveWrite down clearly which actions are allowed, which require confirmation and when the agent must stop and pass the request to a colleague.
- Measure the pilot with simple criteriaQuality of drafts, usefulness as perceived by the team, time spent on the task before and after, how many people actually use it. Note the cases where the agent gets it wrong: they are the basis for improving it.
- Bring the agent into daily work and keep improving itIf the pilot works, the agent becomes part of everyday tools. Sources and instructions are updated based on real questions and feedback from the people using it.
Checklist before you start
Before launching the pilot, go through these points with your team.
Before launching an AI agent in marketing and sales
Defined task. It is clear which workflow you want to improve and who handles it today.
Sources ready. Product sheets, price lists and approved answers are up to date and kept in one place.
Written limits. Allowed actions, actions needing approval and cases to hand over to a person are defined.
Data under control. You know which information the agent may use, and your privacy advisor has been involved.
Internal owner. One person follows the project, updates the sources and gathers colleagues' feedback.
Frequently asked questions
Does an AI agent replace salespeople?
No. It takes on the repetitive steps around selling, such as sorting requests, preparing drafts and keeping track of follow-ups. The customer relationship, the negotiation and decisions on prices and terms stay with people.
Do we need a tidy CRM to get started?
Not necessarily. You can start from documents, archives, FAQs or partial data, assessing with the project team what is already usable. The pilot is also a good moment to find out which data needs cleaning up.
Can the agent email customers on its own?
It can, if the process allows it, with permissions, confirmations and an action log. In marketing and sales, however, it is wiser to start with drafts approved by a person and extend autonomy only where results are stable.
Can it write in our company's tone of voice?
Yes. The instructions define style, terminology, allowed answers and priority sources, as well as the cases in which the agent must stop and ask for a check.
Do you have a repetitive sales or marketing task?
Tell us about it: we will look at it together and assess whether and how to launch a pilot with a custom AI agent connected to the tools you already use.
Discover AI agents
