Which sales tasks can AI agents do reliably in 2026?

AI agents are reliable at account research, inbound lead qualification and routing, meeting booking, first drafts of outreach and content, turning call notes into CRM fields and flagging stalled or poorly evidenced deals. They work best with clean data, narrow rules and human review of anything a customer will see.

Should AI SDR agents send outbound email without human approval?

Most scale-ups should start with a person approving every outbound message and grant autonomy only for low-risk, well-defined workflows once edit rates, reply quality and spam-complaint rates are understood. Volume caps and a separate sending domain for outbound help protect the main company domain.

How can AI outbound damage email deliverability?

High-volume, poorly targeted outbound raises spam complaints and unsubscribes, which lowers the sending domain’s reputation. Google’s sender guidelines ask senders of more than 5,000 messages a day to Gmail accounts to authenticate their domain, offer one-click unsubscribe and keep the reported spam rate below 0.30%, recommending below 0.10%.

What should stay human in sales when AI agents are used?

Relationships with buyers and partners, pricing and contract negotiation, the choice of which segments and deals to pursue, and any communication where the company’s credibility is at stake should stay with people. Agents prepare the ground; people build the trust that closes the deal.

Which metrics show whether AI agents are working in sales?

The useful metrics are pipeline quality, conversion rate at each stage, cost per qualified opportunity, win rate, sales-cycle length, reply quality, spam-complaint rate and CRM data accuracy. Activity volume such as emails sent is an input and can rise while results fall.

Is it legal to use AI agents for prospecting under GDPR?

That depends on the lawful basis for processing prospect data, the national rules on unsolicited electronic marketing in each target country, the source of enrichment data and how objections are handled. Founders should settle these questions with qualified legal advisers before an agent contacts anyone; this is not legal advice.

Insights / Agentic AI in sales and marketing

By Niraj GemawatPublished 3 Nov 2026

Agentic AI in sales and marketing: what should scale-ups automate?

Scale-ups should automate the research, drafting and data work in sales and marketing, and keep relationships, negotiation, judgement and trust with people. In 2026 AI agents handle account research, first-draft outreach, CRM updates and forecast hygiene well when they work from clean data under human review. They do damage when they are used to multiply outbound volume, invent personalisation or act on poor data. The measure of success is pipeline quality and cost per qualified opportunity, not the number of emails sent.

Title graphic reading Agentic AI in sales and marketing, with a network of connected nodes on a navy background

What is agentic AI in sales and marketing?

Agentic AI is software that pursues a goal by planning and carrying out several steps on its own, using tools and data, rather than answering a single prompt. In sales and marketing that means an agent that can research an account, decide which contact to approach, draft a message, log the activity and schedule a follow-up, with a person setting the limits.

The difference from a chatbot or a writing assistant is autonomy: an assistant suggests, an agent acts. That is where both the value and the risk sit. Gartner’s June 2025 prediction that over 40% of agentic AI projects will be cancelled by the end of 2027 points to escalating costs, unclear business value and inadequate risk controls as the reasons. Gartner also warned of "agent washing", where existing chatbots and automation tools are relabelled as agents. Before buying, founders should ask what the tool decides on its own, and what happens when it gets that decision wrong.

How widely are sales teams using AI agents in 2026?

Use is widespread among individual sellers, but scaled deployment inside organisations is still the exception. Salesforce’s State of Sales report 2026, based on a survey of 4,050 sales professionals fielded in August and September 2025, found that 54% of sellers say they have used AI agents and nearly nine in ten plan to by 2027. Sellers in that survey expect agents, once fully implemented, to cut prospect research time by 34% and email drafting time by 36%.

Organisation-wide scale is a different picture. McKinsey’s The state of AI in 2025 survey found that 23% of respondents said their organisations were scaling an agentic AI system somewhere in the enterprise and a further 39% were experimenting. In any single business function, no more than 10% said they were scaling agents. For a scale-up, that means competitors are mostly experimenting too, and a disciplined rollout is still a source of advantage.

Which sales and marketing tasks can AI agents do reliably today?

Agents are reliable where the task is well defined, the inputs are structured and a mistake is cheap to catch. Four categories meet that test today:

Research agents. Building account briefs from company websites, filings, news and CRM history; mapping buying committees; summarising what a prospect has already seen from the company. This is the clearest time saving, and the output is reviewed before anyone acts on it.
SDR agents. Qualifying and routing inbound leads, booking meetings, sending agreed follow-ups and drafting first-touch outbound for a seller to approve. A sales development representative (SDR) agent works well inside narrow rules: a defined ideal customer profile, approved messaging and volume caps.
Content agents. First drafts of emails, landing-page variants, case-study summaries, localised versions for different markets and repurposing long-form content into shorter formats. Editing, claims-checking and the point of view stay with people.
CRM and forecasting agents. Turning call notes into structured CRM fields, flagging deals with no next step or a stalled stage, spotting duplicate records and highlighting gaps between the forecast and the evidence in the pipeline. These agents improve the data every other agent depends on.

The common thread is that agents take over preparation and administration, so sellers and marketers spend more time with customers. Agents earn their place by returning time to people, not by replacing the conversation.

What should stay human in sales and marketing?

Anything that depends on trust, negotiation or judgement should stay with people. That includes relationships with buyers and partners, pricing and contract negotiation, the choice of which segments and deals to pursue, and any message where the company’s credibility is on the line.

Buyers are moving in the same direction. Forrester’s 2026 B2B marketing, sales and product predictions, published in October 2025, report that 19% of buyers using generative AI applications feel less confident in their purchasing decisions because of inaccurate information, and predict that human expertise will rival generative AI in appeal as buyers seek deeper validation. The same predictions expect at least one in five B2B sellers to be compelled to respond to AI-powered buyer agents in 2026. When the buyer’s side is automated too, the seller’s judgement about terms, risk and fit matters more, not less.

This is consistent with how buyers now discover vendors: increasingly through AI assistants, before any conversation with sales (see generative engine optimisation). The human role shifts towards the moments that decide trust, a theme explored in where human insight meets AI.

What goes wrong when AI agents run sales and marketing?

Most failures come from using agents to scale activity before the underlying commercial system is sound. The recurring failure modes are:

Spammy outbound at scale. An agent that can send a thousand emails a day will, if allowed. Poorly targeted volume generates rejection, unsubscribes and a reputation for noise that is hard to reverse.
Deliverability damage. Google’s email sender guidelines require senders of more than 5,000 messages a day to Gmail accounts to authenticate with SPF, DKIM and DMARC, support one-click unsubscribe for marketing messages and keep the spam rate reported in Postmaster Tools below 0.30%, with 0.10% as the recommended ceiling. Outbound from the main company domain puts invoices, support replies and customer mail at risk if that reputation drops.
Hallucinated personalisation. Agents can invent a prospect’s initiative, misread a job change or cite a news item that does not exist. A single false detail undoes the purpose of personalising at all.
Poor data quality. Agents act on whatever the CRM says. In Salesforce’s 2026 survey, 51% of sales leaders with AI said disconnected systems were slowing their AI initiatives, and 74% of sales professionals said they were focusing on data cleansing.
Brand and governance risk. Agents that publish content or make claims without review can state things the company cannot support. Forrester’s 2026 predictions warn that ungoverned use of generative AI will lead to incidents that cost enterprise value through falling share prices, legal settlements and fines.

An agent multiplies whatever system it is given, including its weaknesses.

Which data protection questions should founders ask before deploying AI agents?

Founders should treat consent and data protection as design questions to settle with qualified legal advisers before an agent contacts anyone. This article is not legal advice; the rules differ between the EU, the UK, the US states and India, and between email, phone and messaging channels. The questions to answer are:

What is the lawful basis for processing each prospect’s personal data, and has it been documented?
Do the rules on unsolicited electronic marketing in each target country allow this contact without prior consent, and for which kinds of recipient?
Where did the enrichment data come from, and can the vendor show how it was collected?
How is the person told who is contacting them, why and how to object, and is every objection honoured across every tool?
Where does the agent vendor store and process the data, and under what contract?
Does the agent make any decision about a person, such as scoring or excluding them, that needs human review or explanation?
How long are prospect records kept, and who deletes them?

How should a scale-up adopt AI agents in sales and marketing?

Adopt in stages, starting where the risk is internal and moving outward as data and controls prove themselves. Gartner’s June 2025 release recommended pursuing agentic AI where it delivers clear value and rethinking workflows rather than adding agents to existing ones. A practical sequence:

Fix the foundation. Agree the ideal customer profile, stage definitions and qualification criteria; clean and connect CRM data. Without this, every later stage inherits the errors.
Start with internal agents. Deploy research, call-summary and CRM-update agents that never touch a customer. Measure time saved and data accuracy.
Add drafting with human approval. Let agents draft outbound, follow-ups and content, with a person approving every message. Track how often drafts are edited or rejected, and why.
Grant narrow autonomy. Allow agents to act alone on low-risk, well-defined workflows such as inbound routing, meeting booking and agreed follow-ups, with volume caps, a separate sending domain for outbound and a clear way to stop the agent.
Review and redesign quarterly. Expand what works, retire what does not, and redesign the workflow around the agent rather than layering it onto the old process.

Which metrics show whether AI agents are improving sales and marketing?

Measure outcomes in the pipeline, not activity volume. Emails sent, sequences launched and tasks completed are inputs; an agent can raise them sharply while pipeline quality falls. The metrics to watch are:

Pipeline quality. The share of new opportunities that match the ideal customer profile and progress past the first stage.
Conversion by stage. From first touch to meeting, meeting to qualified opportunity, and opportunity to closed-won, compared with the period before the agent.
Cost per qualified opportunity. Total sales and marketing cost, including tools and the people reviewing agent output, divided by qualified opportunities created.
Win rate and sales-cycle length. Whether opportunities sourced or supported by agents close at the same rate and pace as others.
Reply quality and guard rails. Positive versus negative replies, unsubscribe and spam-complaint rates, and domain reputation.
Data accuracy and time returned. CRM completeness, forecast accuracy, and how sellers use the hours agents give back.

If pipeline quality and cost per qualified opportunity are not improving, more activity is not progress.

How does TGC Capital Partners approach AI in go-to-market?

TGC Capital Partners, the strategic investment arm of Gateway Group of Companies, works with tech founders in Europe, North America and India as their Growth Engineering Partner. Gateway Group has been a software engineering company since 1997 and has more than 2,000 people in offices across 15 countries. Through 1X Revenue Engineering, TGC works with founders on the commercial system that agents depend on: ideal customer profile, positioning, pipeline definitions and the metrics above. Through 2X Software Engineering and Tech Excellence, it can support the data integration and engineering work that makes agents reliable. Revenue is the outcome TGC measures, not one it promises. The wider model is described on the platform page.

Agents change the cost of sales and marketing work; the commercial system decides whether that work becomes revenue.

Frequently asked questions

What is agentic AI in sales and marketing?: Agentic AI in sales and marketing is software that pursues a goal by planning and carrying out several steps on its own, such as researching an account, drafting outreach, logging activity in the CRM and scheduling a follow-up, within limits set by people. It differs from a writing assistant because it acts rather than only suggesting.
Which sales tasks can AI agents do reliably in 2026?: AI agents are reliable at account research, inbound lead qualification and routing, meeting booking, first drafts of outreach and content, turning call notes into CRM fields and flagging stalled or poorly evidenced deals. They work best with clean data, narrow rules and human review of anything a customer will see.
Should AI SDR agents send outbound email without human approval?: Most scale-ups should start with a person approving every outbound message and grant autonomy only for low-risk, well-defined workflows once edit rates, reply quality and spam-complaint rates are understood. Volume caps and a separate sending domain for outbound help protect the main company domain.
How can AI outbound damage email deliverability?: High-volume, poorly targeted outbound raises spam complaints and unsubscribes, which lowers the sending domain’s reputation. Google’s sender guidelines ask senders of more than 5,000 messages a day to Gmail accounts to authenticate their domain, offer one-click unsubscribe and keep the reported spam rate below 0.30%, recommending below 0.10%.
What should stay human in sales when AI agents are used?: Relationships with buyers and partners, pricing and contract negotiation, the choice of which segments and deals to pursue, and any communication where the company’s credibility is at stake should stay with people. Agents prepare the ground; people build the trust that closes the deal.
Which metrics show whether AI agents are working in sales?: The useful metrics are pipeline quality, conversion rate at each stage, cost per qualified opportunity, win rate, sales-cycle length, reply quality, spam-complaint rate and CRM data accuracy. Activity volume such as emails sent is an input and can rise while results fall.
Is it legal to use AI agents for prospecting under GDPR?: That depends on the lawful basis for processing prospect data, the national rules on unsolicited electronic marketing in each target country, the source of enrichment data and how objections are handled. Founders should settle these questions with qualified legal advisers before an agent contacts anyone; this is not legal advice.