What Sales Tasks Should You Automate With AI?

AI can automate more sales work than most businesses currently use it for. That does not mean you should automate all of it.

The better question is: Which parts of your sales process are repetitive enough to automate, require enough understanding for AI to help, and carry low enough risk for the level of authority you are considering?

A useful starting framework is:

• RULE → AUTOMATION
• UNDERSTANDING → AI
• RESPONSIBILITY → AI AGENT
• JUDGEMENT → HUMAN

Most good sales processes will use a combination of all four. The objective is not maximum automation. It is a better sales process.

Start with your sales process. Not a list of AI tools.

A common approach to sales AI looks something like this: Find an AI tool. Look through its features. Try to work out where those features could be used. That is backwards.

Start with what happens today. Follow an enquiry. From the moment it arrives to the moment it becomes: A customer. A closed opportunity. A future opportunity. Or something you decide not to pursue.

Look at every step. Who does something? What do they do? What information do they need? Where does that information come from? What decision do they make? What happens next? Where does somebody wait? Where does somebody copy information? Where does somebody have to remember something? Where does somebody repeatedly interpret information before taking a predictable next step?

Those are the places worth examining.

Look for five signals.

Then decide what kind of system the task needs.

Not every automation problem needs AI. And not every AI problem needs an agent. Use this framework.

Rule → Automation

Use normal automation when the decision is predictable.

If the logic can reliably be expressed as: IF this happens, THEN do that you may not need AI.

Examples: When a website form is submitted, create a CRM record. When an opportunity reaches an agreed stage, create a task. When a meeting is booked, send the calendar information. When a field changes, notify the appropriate team. When a known condition occurs, move information between systems.

Traditional automation is: Predictable. Fast. Relatively easy to test. Relatively easy to understand. Often cheaper and simpler than introducing AI. Do not replace a reliable rule with AI simply because AI is newer.

Understanding → AI

Use AI where information needs interpretation.

Some work is not a simple rule. An enquiry might arrive saying: "We're looking at replacing the way our regional team currently handles incoming commercial enquiries. We use HubSpot but most of the process still happens through email. Could you help?"

A normal automation can move the message. AI can potentially help understand it. For example: What is the person asking? Which service does it relate to? What systems did they mention? What information is missing? Does an existing account appear to exist? What context might the salesperson need? The AI can interpret the information and prepare something useful. It still does not necessarily need authority to act.

Responsibility → AI Agent

Use an agent when a defined piece of work needs to keep moving.

An AI agent becomes more relevant when the system has responsibility for a job rather than simply responding to a single prompt. For example: "Make sure every new website enquiry reaches the appropriate salesperson with the relevant account context."

That job might require the agent to: Notice a new enquiry. Understand it. Identify the organisation. Check existing records. Gather relevant context. Apply routing rules. Recognise missing information. Prepare the handover. Carry out approved internal actions. Escalate exceptions. That is more than generating content. The system is responsible for moving a defined piece of work towards an outcome.

Judgement → Human

Keep people where judgement matters.

Some sales work involves context that should not simply be handed to AI. For example: Negotiation. Important commercial decisions. Sensitive conversations. Complex objections. Relationship judgement. Unusual customer situations. Commercial exceptions. Strategic account decisions. Significant commitments.

AI may still help. It can gather information. Prepare options. Summarise context. Identify previous decisions. Surface relevant documents. But the person remains responsible for the judgement.

The same workflow can use all four.

It does not have to be "AI" or "not AI".

Imagine a new sales enquiry moving through a single AI sales workflow.

One workflow. Different technologies. Different responsibilities. That is usually more sensible than trying to make everything "agentic".

1. New enquiry handling

Often a good place to look.

A new enquiry may require somebody to: Read it. Understand the request. Identify the company. Search the CRM. Check previous history. Determine whether information is missing. Decide who should handle it. Prepare a response. Create or update a record. That is a mixture of: Rules. Interpretation. Information gathering. Action. It can be a strong candidate for AI assistance or an agent.

What should stay controlled? Customer-facing responses. Commercial commitments. Unusual enquiries. Unclear routing. Sensitive situations. The AI should have an escalation path.

2. Lead qualification

AI can apply your criteria. It should not invent them.

Lead qualification is often discussed as though AI should simply decide whether a lead is "good". That is too vague. Your business needs to define what matters. For example: Relevant service need. Geography. Company type. Project type. Required capability. Timing. Known constraints. Existing relationship.

AI can help gather information and compare it with those criteria. But qualification should remain explainable. Instead of a single opaque number, a useful output gives the salesperson something they can understand.

That gives the salesperson something they can understand.

3. Lead routing

Rules first. AI when the routing needs interpretation.

If routing is: Postcode A → Sarah, Postcode B → James you probably need automation. If routing requires understanding: What is the enquiry about? Which service is relevant? Does an existing relationship exist? Which specialist is appropriate? Is the request unusual? then AI may help. The objective is not sophisticated routing. It is getting the enquiry to the right place with useful context.

4. Meeting preparation

A strong low-authority use case.

Before a meeting, AI can potentially gather: Relevant CRM history. Previous meeting context. Outstanding actions. Relevant email context. Known opportunity information. Important documents. Approved external information. Then prepare a concise briefing. The salesperson reviews it. No systems need changing. No customer needs contacting. No autonomous action is required. This is an important point: A useful AI system does not need a lot of authority.

5. Meeting follow-through

Turn conversation into action.

After a meeting, AI may help: Identify decisions. Extract commitments. Separate customer actions from internal actions. Identify unanswered questions. Prepare CRM updates. Prepare tasks. Prepare follow-up. Track agreed future actions. A person can approve important outputs before anything changes. This can reduce the gap between: We had the meeting and: The process actually moved forward.

6. CRM administration

Automate the reconstruction, not the judgement.

Salespeople often have to recreate information in the CRM after work has already happened elsewhere. AI can potentially prepare: Meeting summaries. Next actions. Relevant field updates. Contact information. Opportunity notes. Tasks. But do not automatically give AI permission to change every CRM field. A meeting summary and an opportunity value do not carry the same consequence. Define permissions field by field where appropriate.

7. Sales follow-up

Context matters more than the calendar.

A simple follow-up automation might say: Seven days passed. Send an email. AI can potentially ask: What happened last? Who owes the next action? Is the customer waiting for us? Did we promise something? Did the customer reply somewhere else? Is the opportunity still active? Does contacting them make sense? A good follow-up workflow is not simply a more sophisticated email sequence. It is a way of understanding: What needs to happen next?

8. Account research

Gather what matters.

Salespeople may repeatedly search: CRM. Email. Meeting history. Documents. Internal knowledge. Approved external sources. AI can help assemble relevant context. This can often begin as a read-only workflow. The AI gathers and prepares. The salesperson decides. That is useful automation without autonomous action.

9. Internal handovers

Move context with the work.

When an opportunity moves between people, AI can help prepare: Relevant history. Requirements. Decisions. Commitments. Open questions. Documents. Next action. Ownership. The new person should not have to reconstruct the entire account from scratch.

10. Sales attention

Help people see what needs attention.

Sales teams often have dashboards full of information. The problem is knowing what matters today. An AI workflow might identify: Active opportunities without a next action. Overdue internal commitments. New enquiries without an owner. Upcoming meetings without preparation. Opportunities where something appears to have stalled. Records with conflicting information. It may not need to take action. It can simply surface the right work. Sometimes the most useful agent does not sell anything. It notices.

What should you NOT automate first?

Some work may be technically automatable but still be a poor starting point.

Do not start with rare work. If something happens twice a year, the effort required to design, integrate, test and maintain the workflow may not be justified.

Do not start with unclear work. If five people describe the process five different ways, understand the process first. AI cannot reliably automate a rule nobody has defined.

Do not start with maximum consequence. Your first AI workflow probably does not need authority to make significant commercial commitments on behalf of the company. Start somewhere easier to observe and control.

Do not start with maximum access. Do not connect every company system simply because you may eventually want the AI to use them. Give the workflow the information required for its defined job.

Do not start by replacing the human. Ask which work around the person could improve first. Often the easiest opportunities are: Preparation. Research. Administration. Information movement. Monitoring. Next-action tracking. The boring task may be the best one.

Consider two potential AI projects.

Project A sounds more impressive. Project B may be much easier to: Define. Test. Control. Observe. Measure. Improve. That makes it a very sensible place to begin.

The most impressive agent is not necessarily the best first agent.

Use Frequency, Friction and Consequence.

A simple way to compare opportunities.

For each task, ask three questions.

You are looking for tasks with: Meaningful frequency. Meaningful friction. Manageable consequence. That combination often gives you a sensible first project.

Then decide the authority.

Automation is not one switch.

Once you find a suitable task, decide what the AI should actually be allowed to do. Use the Agentic Selling Authority Ladder.

The correct level depends on the action. Authority belongs to the action. Not the agent. For the full model, see our guide to AI agent governance and authority.

Imagine an Enquiry Agent. It might: READ the enquiry. READ relevant CRM information. RECOMMEND the correct salesperson. PREPARE a response. ACT WITHIN LIMITS by adding an internal tag. ACT WITH APPROVAL before sending an external response. ESCALATE an unusual commercial request. One agent. Several authority levels. This is much more useful than asking: "Is the agent autonomous?"

Autonomy should earn its place.

Start with the authority required to make the workflow useful. You can begin with: Read. Recommend. Prepare. Observe what happens. Does the AI find the correct information? Are its recommendations useful? Do people regularly change its work? Where does it struggle? What exceptions occur? Then decide whether any additional authority would improve the process.

And if performance deteriorates? Move authority back down. Authority should not only move in one direction.

How do you know if a task is suitable for AI?

Ask: Is there a clearly defined job? If you cannot explain what the system is responsible for, keep working on the definition. Does the task happen often enough? Automation needs a reason to exist. Is there repeated manual friction? Searching, copying, checking, preparing, interpreting or remembering. Does the task require understanding? If not, conventional automation may be enough. Can the AI get the information it needs? Without inappropriate access. Can success be recognised? You need some way of knowing whether the workflow is useful. Can exceptions be identified? The AI needs somewhere to go when normal logic stops working. Can the consequences of mistakes be controlled? Through permissions, approval, validation or escalation.

If the answers are reasonably clear, you may have a good candidate.

What if conventional automation can do it?

Use conventional automation.

There is no prize for putting AI into every workflow. Traditional automation is excellent at: Moving information. Triggering predictable actions. Creating records. Changing known fields. Sending fixed notifications. Running scheduled processes. Connecting systems. Use it. Then introduce AI where the workflow genuinely needs: Understanding. Interpretation. Context. Flexible preparation. Reasoning between defined options. Use the simplest thing that works. If you are weighing up the options, our guide to AI agent vs automation compares them in detail.

What if existing software already does it?

Use the existing software.

Before building anything custom, check whether: Your CRM already supports it. Your email platform already supports it. Your meeting software already provides it. Your automation platform already handles it. An existing product solves enough of the problem. Custom development should focus on what is genuinely specific to your process. Buy the commodity. Build the difference.

What if the process itself is broken?

Fix that first.

Suppose follow-up is unreliable because nobody agrees who owns it. You could build an AI agent to monitor the CRM. But the ownership problem still exists. Suppose qualification is inconsistent because nobody has agreed the criteria. AI cannot make undefined criteria consistent. Suppose your CRM contains unreliable information. Automating more updates does not automatically make the information trustworthy.

AI does not fix a broken sales process. It can make the broken process happen faster.

A simple exercise

Example

A website enquiry arrives.

Now you can see where different technologies fit.

The goal is not an autonomous sales process. The goal is a better sales process.

A good result might be: The enquiry is understood faster. The salesperson gets better context. The CRM requires less manual work. Meeting preparation becomes easier. Commitments are less likely to disappear. Follow-up becomes more relevant. Handovers preserve information. People spend less time searching. Exceptions reach the right person. None of those requires removing the salesperson. It requires designing the work properly.

Frequently asked questions

Start with one piece of work.

Do not begin with: "How do we automate sales with AI?" That is too big. Ask: What does our team repeatedly have to read? What do they repeatedly have to find? What do they repeatedly have to copy? What do they repeatedly have to prepare? What do they repeatedly have to remember? What do they repeatedly have to interpret? Then ask: Does this need a rule? Automation? AI? An agent? A person? Or some combination? That is how you find the work worth automating.