AI in sales, marketing and CS: what already works in 2026

AI already works in revenue operations on three fronts that read the same CRM. In sales, the agent prepares the meeting, summarizes the call and drafts the follow-up. In marketing, it personalizes nurturing emails from the contact’s history. In post-sale, it classifies each new ticket by category, priority and sentiment before anyone opens the queue. The result depends on leadership decisions about the context the agent reads, who approves each action and how many credits each area spends. On an unstructured CRM, AI makes the operation get things wrong faster, and you can delegate execution to an agent while responsibility stays with whoever approved it.

At Iasbeck&Co we implement CRM and AI agents in B2B operations and train the teams that work with them. In early October I ran a live workshop showing agents running inside a real CRM. The questions in the chat came down to starting without wasting credits and staying in control once the machine talks to customers on the company’s behalf. I wrote this for the people who answer for those decisions and for the HR team that will organize training.

What changed in AI for revenue operations from 2025 to 2026?

In 2025, using AI in the operation meant writing a polished prompt like “act as a B2B sales expert” and asking for a message to the customer who went silent. The chat returned a text, and opening the CRM, adapting the message, sending it and logging it stayed with the person. In 2026 the instruction fits in one sentence, such as “follow up on the deals that have been stalled for more than 15 days”. The agent finds the opportunities, reads each customer’s history, writes an approach for each one and asks for approval before sending.

For leadership, the biggest change is in the input. The agent needs access to the systems and context about the company: what you sell, to whom, how the team runs the sale, how support classifies a problem and what tone the messages use. In the workshop we called this phase zero of the rollout, when the playbook forgotten in Drive starts feeding the agents from inside the CRM.

	2025: asking the chat for text	2026: agent connected to the CRM
Who executes	The person, after reading the suggestion	The agent, and the person reviews
What the AI needs	A well-written prompt	CRM access, company context and approval rules
Where the context lives	In the head of whoever writes the prompt	In the CRM, shared by the agents of all three areas
Who decides the use	Each professional, on their own	Leadership, with a budget and approval rules
The team’s role	Operate	Delegate, check and decide

Which uses of AI in sales, marketing and CS already work?

The uses that pay off in the first weeks work on text the operation already produces, such as call transcripts, emails and ticket messages. They work on an average database, as long as someone checks the output before it becomes a record or a message.

AI in sales: from meeting prep to pipeline analysis

Picture the salesperson who inherited an account and has a discovery meeting at 2 p.m., with dozens of emails and three calls held by the colleague who left. The agent reads everything and returns who took part, what the company promised, which objection came up and what is still pending, pointing out the gaps in the history. After the call, it returns pain, impact, who decides and next step in the same format for every deal and proposes field values quoting the customer, which tackles one of the reasons I discuss in why the team does not use the CRM. If the salesperson never asked who decides, the model will not find that answer in the transcript, which is why AI and consultative selling go together. For the manager, the agent drafts the follow-up on the stalled proposal, without mentioning price unless approved, and lists for the pipeline meeting the deals that are stalled or have no mapped decision maker.

AI for marketing: nurturing, segmentation and lead quality

In the nurturing agent we showed in the workshop, you pick an email from an automated journey, and on each send the agent reads the contact’s data, recent engagement and associated deals to rewrite the message for that person. Before activating it, the manager compares the base version with the personalized one, and in the settings defines what percentage of sends goes through the agent, which caps credit consumption and creates a comparison group with the people who received the standard email. The same agent builds segments from the audience the analyst describes and compares each new lead with the ICP written into the context; without a written ICP, it qualifies on generic criteria.

AI in support and Customer Success: triage, renewal and risk

I also showed a triage agent inside a support automation workflow. For each new ticket, it reads the conversation and the company’s other tickets, checks the knowledge base and writes category, priority, sentiment and a summary into the ticket fields. In the example, it classified the case as a product issue, urgent and critical, quoting the sentence in which the customer said they could not issue a single invoice. In Customer Success, the fastest use is the account summary before the renewal meeting, with the period’s tickets, check-in notes and contacts who joined or left. It also flags risk signals, such as repeated critical tickets, with the quality of what CS has recorded.

Area	Use	Raw material	Works on an average database?
Sales	Meeting prep and call summary	History, emails and transcript	Yes, pointing out gaps
Sales	Field updates and follow-up	Transcript and deal history	Yes, with approval
Sales	Pipeline analysis	Funnel records	Partly, it inherits the database’s errors
Marketing	Nurturing personalization	Contact data and engagement	Yes, with a controlled percentage
Marketing	Segmentation and lead quality	Fields, activity and written ICP	If fields and ICP are standardized
Support	Ticket triage	Conversation and account history	Yes, with defined categories
Customer Success	Renewal summary and risk signals	Tickets, notes and contacts	Partly, it depends on what CS records
All	Revenue forecasting and predictive scoring	Historical series	No, without reliable dates, stages and loss reasons

Why does AI on an unstructured CRM only speed up the error?

Gartner predicted in July 2026 that by 2028 AI agents will outnumber sellers ten to one, and that fewer than 40% of sellers will say the agents improved their productivity. The firm estimates that sales leaders who overhaul data, automation and user experience will be five times more likely to get a return on AI than those who choose shortcuts. For whoever approves the budget, the groundwork weighs more on the return than the agent license.

A duplicated company gets two follow-ups from the same agent on the same morning, and a contact with an outdated job title gets a nurturing email written for the role they left a year ago. A deal the team pushed into the proposal stage to improve the weekly report enters the forecast as a proposal. With a spreadsheet, the manager distrusted the number; with AI, the number arrives with percentages, charts and analytical language, and hardly anyone questions it.

The bar for using the analytical layer is lower than it looks: reliable dates on records, objective entry criteria for each stage, duplicates handled and loss reasons picked from a closed list. None of these points involves new technology, and all of them depend on team discipline, the subject of how to make data-driven decisions. A quick test: export the database and hand it to an analyst who does not know the company. If they cannot tell how much you close in the quarter, the agent will not be able to either.

AI agents in the CRM: who answers for what they do?

Picture Monday’s results meeting. The report says the team closed 80 deals in September, and it closed 8. The manager says the AI built the report, and the director remembers that the document went out under his name. No customer and no board accepts “the AI did it” as an explanation, because the AI does not sign the proposal or sit in the results meeting.

Delegating to an agent asks for the same as delegating to someone on the team: knowing what you asked for, setting how far it can go, checking what came back and owning what you approved. Before turning on any agent that writes to people outside the company, I draw up three lists with the leadership of each area. The first holds what the agent does on its own, such as summarizing a call or classifying a ticket. The second holds what it does and then waits for approval, such as messaging a customer in the decision stage or extending personalization to a new segment. The third holds what it never does, such as granting a discount, changing a close date or answering a critical ticket alone.

In the workshop someone asked whether the people who build the agents need training. The answer was yes, and I added the line I repeat in our projects: the tool got simple, and simple is different from easy. Writing the instruction and deciding who approves what takes someone who knows the company’s commercial process.

What does leadership need to decide before turning on the agents?

Sales directors, marketing heads and CS leaders tend to receive AI through the tool: someone turns on an agent and the area finds out about the cost and the errors later. I start with the decisions below, in a meeting with the three leaders and whoever owns the CRM.

Decision	Who usually decides	What needs to be written down
Which uses come first	Each area’s leader, with RevOps	One or two uses per area and the pilot team
Who approves each action	Each area’s leader	The three lists, with the approver’s name
Credit budget	Operations leadership with finance	Monthly cap per agent and personalization percentage
Company context	Each process owner	ICP, stage criteria, tone, ticket categories
Minimum database	RevOps or the CRM admin	Fields each agent reads and the rule for filling them
How to measure	Area leadership and HR	Indicators per area and comparison period

For the first uses, look for high-volume tasks on text that already exists, where mistakes are cheap to fix. On platforms that charge per agent action, cost varies with the complexity of the request, and leadership needs to estimate the average spend per use and define who can turn on a new agent. In McKinsey’s global survey on AI from August 2026, one in five respondents said their organization limits AI use because of operating costs, which include tokens. Measurement starts with behavior, because revenue and retention respond later. In sales, track meetings with a logged next step and drafts approved without edits; in marketing, the response of the personalized group against the standard one and the leads sales accepted; in support and CS, time to first response on urgent tickets and the classifications the analyst corrected.

AI training for teams: what HR and L&D need to consider

AI became a training topic when the tool started executing on the company’s behalf. The team needs to know how to ask, with instructions that have inputs, rules and an output format, and how to check, recognizing when a summary sounds confident about an incomplete record. Each person also needs to understand that they answer for what they approved.

In the cohorts, the sales, marketing and CS leaders work through the decisions in the previous section and leave with the approval lists drawn up. Salespeople, marketing and support analysts and account managers practice on cases from their own area. The CRM admin or the RevOps team learns to configure agents, knowledge repositories and credit limits.

The format that pays off is in-company, built on the company’s process and data, with practice on real cases and a deliverable the team keeps. The satisfaction survey measures the class. To measure the operation, watch in the following weeks how the team uses the prompt library and the share of drafts edited and classifications corrected, the logic I defend in sales training that changes behavior.

How to start with AI in revenue operations: a four-week plan

This is the plan I use to move from individual chat to agents in the CRM, starting small, as we recommended in the workshop.

Week 1: choose the uses and write the context. Start with one or two uses per area, in a small team with a manager willing to review what the agent produces. Put into the CRM what the company sells, to whom, the criteria for each stage, the ticket categories and the tone of the messages.

Week 2: set approval, limits and credits. Build the three lists with the name of whoever approves each action, record the credit cap per agent and start nurturing with a low percentage of personalized sends. A low limit costs little and shows quickly where the context is weak.

Weeks 2 and 3: clean only the fields the agent will read. For meeting prep, associate calls and emails with the right deal; for triage, replace free-text category and priority with closed options. The rest of the database joins the queue once the pilot proves the gain.

Week 4: measure and decide. Compare each area’s indicators with the previous four weeks, in the same team. The rate of drafts approved without edits and classifications accepted without correction shows whether the context is good.

What stalls this plan is rarely the tool. It stalls when teams do not know what to ask for, what to check and who answers for what they approved, or when the CRM is not organized enough for the agent to work. That is the work of the in-company training on AI applied to revenue operations, built on the company’s process with sales, marketing and CS cohorts, which delivers the prompt library for each area and the agent architecture, with what each agent reads, what it executes on its own and who approves the rest.

Frequently asked questions

How do you start using AI in the commercial team?

Choose one or two uses per area that work on text the team already produces, such as call summaries, nurturing with a controlled percentage or ticket triage. Write the company context into the CRM, define who approves each action and run a four-week pilot with a small team.

How long does it take to see results from AI in revenue operations?

Text-based uses show gains in the first weeks, and a four-week pilot is enough to measure behavior. Revenue forecasting depends on reliable dates, stages and loss reasons, which takes one to three months of cleanup in a mid-sized operation.

How do you measure the results of AI in sales, marketing and CS?

Measure behavior before revenue, comparing against the period before the pilot in the same team. Use logged next steps in sales, the personalized group’s response against the standard one in marketing and time to first response on urgent tickets in support.

Do I need to switch CRMs to use AI agents?

In most cases, no. The main CRM platforms already offer native and custom agents, and what tends to be missing is written context, an organized database and approval rules.

Who should take part in AI training for the commercial team?

The sales, marketing and CS leaders, who decide uses, approval and budget, the professionals who operate in each area and whoever administers the CRM. HR or L&D coordinates the program and follows the behavior assessment after the sessions.

What are the risks of putting AI agents into revenue operations?

The first is the agent acting on a dirty database, sending duplicate messages or personalizing for an outdated job title. The second is the team approving without checking, so the agent’s error reaches the customer as the company’s error. Approval lists and team training reduce both.