Agentic selling in practice

What does an AI sales agent actually do?

AI watching for what needs attention

Forget the imaginary AI employee. The easiest way to understand agentic selling is to see what happens when AI is given responsibility for a specific piece of sales work.

An enquiry arrives. A quote needs following up. A meeting is tomorrow. The CRM needs updating. A prospect asks for something unusual.

Those are jobs. And some of those jobs can now be handled differently.

Here are eight examples.

Explore the examples ↓
A job arrives THE AGENT HANDLES DEFINED WORK Read Check Prepare Update Act You take theexceptions
Built by Creative Sauce AI

These are examples of agentic sales workflows, not case studies. Real numbers depend on your business, your systems and the work you choose to improve.

First, one important thing

These aren't eight products we think every business should buy.

They're examples of agentic workflows. You might need one. You might combine several. You might discover that ordinary automation is enough.

The starting point should always be:

Ask this
What work are we trying to improve?
Not this
How many AI agents can we add?
01

The Website Enquiry Agent

When a new lead arrives

From form submission to useful next action.

New enquiry Understand Check Identify gaps Prepare Update Act or escalate
Before
  • Someone completes your website enquiry form.
  • An email lands in an inbox.
  • Someone eventually sees it.
  • They read it.
  • Open the CRM.
  • Search for the customer.
  • Check what service they're asking about.
  • Look for the relevant information.
  • Write a response.
  • Create or update the CRM record.
  • Make a note to follow up.
  • Hopefully.
With an agentic workflow
  • New enquiry
  • UnderstandWhat does this person actually want?
  • CheckExisting customer? Previous conversation? Relevant service?
  • Identify gapsIs anything needed before we can respond properly?
  • PrepareCreate the appropriate response.
  • UpdateCreate/update CRM and next action.
  • Act or escalateRoutine enquiry? Continue within permitted limits. Something unusual? Send to a person.
Human handover
Human handover
⚠ Needs your attention
Customer has requested a non-standard service and asked for pricing that isn't in the approved information.
[Review enquiry]
Possible value
  • Faster initial response.
  • Less administration.
  • More consistent handling.
  • Fewer enquiries forgotten.
Sensible starting authority
Read→ Recommend→ Prepare→ Human approval
[Can AI respond to website enquiries? →]
02

The Lead Qualification Agent

Before the sales call

Find out whether there's an opportunity and what's still missing.

Incoming enquiry "Could somebody call me?" AI QUALIFIED LEAD Service: Commercial installation Timescale: Within 6 weeks Decision maker: Yes Locations: 3 Budget: Not confirmed Missing: Site dimensions Recommended: Discovery call
Before
  • A salesperson receives an enquiry containing: "Hi, we're interested in your service. Could somebody call me?"
  • That's about all they know.
  • They call.
  • The person isn't available.
  • Emails go backwards and forwards.
  • Eventually they discover the customer isn't looking for something the business actually provides.
With an agentic workflow
  • Lead arrives
  • InterpretWhat can we already understand?
  • CompareDoes it broadly fit the business's agreed criteria?
  • GatherAsk only for the important missing information.
  • OrganisePut the answers into a useful lead record.
  • RecommendQualified / More information needed / Human review / Not currently suitable
  • RouteSend it to the appropriate next step.
Human handover

Instead of: "Someone filled in the form." The salesperson gets:

New qualified enquiry
Service:Commercial installation
Locations:3
Timescale:Within 6 weeks
Budget:Not confirmed
Decision maker:Yes
Existing customer:No
Missing:Site dimensions
Recommended next step:Discovery call
Possible value
  • Less time gathering basic information.
  • Better prepared first conversations.
  • Faster routing.
  • More consistent qualification.
Sensible starting authority
Read→ Recommend→ Prepare

The business defines the qualification criteria. The AI doesn't invent them.

[Can AI qualify sales leads? →]
03

The Sales Follow-Up Agent

After the quote, call or email

Make sure "I'll chase that next week" actually happens.

Quote sent Record next action Wait / check Prepare follow-up Continue or stop
Before
  • A quote goes out.
  • Monday passes.
  • Tuesday passes.
  • The salesperson gets busy.
  • Next Friday somebody asks: "Whatever happened to that £12k enquiry?"
  • Nobody followed up.
With an agentic workflow
  • Quote sent
  • Record next actionFollow up after agreed period.
  • WaitNo pointless activity while nothing needs doing.
  • CheckHas the customer replied? Has somebody else contacted them? Has the opportunity changed?
  • PrepareCreate an appropriate follow-up using the actual context.
  • ActSend within defined rules or request approval.
  • Continue or stopReply received? Hand back into the process. No reply? Follow agreed rules. Stop condition reached? Stop.
Human handover
Human handover
2 follow-ups ready
Acme Ltd
Quote sent 6 days ago. No response. Follow-up prepared.
Smith & Co
Customer previously asked to be contacted this week.
[Review both]
Possible value
  • More consistent follow-up.
  • Less reliance on memory.
  • Fewer opportunities disappearing.
  • Less time checking what needs chasing.
Sensible starting authority
Prepare→ Human approval

Routine follow-up may later move to ACT WITHIN LIMITS if appropriate.

04

The Meeting Preparation Agent

Before someone joins the call

Turn scattered information into useful context.

CRM historyPrevious emails Old notesCompany research TOMORROW'S MEETING BRIEF Who / History / Opportunity Commitments / Open questions Research / What to discuss next
Before
  • Meeting in 15 minutes.
  • Open CRM.
  • Search email.
  • Read old notes.
  • Open company website.
  • Try to remember what happened last time.
  • Discover three minutes into the call that somebody else spoke to the customer yesterday.
With an agentic workflow

The meeting appears on tomorrow's calendar. The agent prepares:

  • WhoWho are we meeting?
  • HistoryWhat has happened previously?
  • OpportunityWhat are they interested in?
  • ConversationsWhat has already been discussed?
  • CommitmentsWhat have we said we'll do?
  • Open questionsWhat's still missing?
  • ResearchWhat external information is genuinely relevant?
  • NextWhat might need discussing?
Human handover
Meeting tomorrow, 10:30
Company:Acme Ltd
Last contact:14 September
Opportunity:Website redevelopment
Quote:Not yet prepared
Main concern:Migration timescale
Previous commitment:Sarah to confirm API compatibility
Still unresolved: API compatibility
[Open full brief]
Possible value
  • Less preparation time.
  • Better-informed conversations.
  • Fewer missed commitments.
  • Less searching between systems.
Sensible starting authority
Mostly Read→ Prepare

Low autonomy. Potentially high usefulness.

05

The Proposal Agent

After the conversation

Get from meeting to first draft without rebuilding everything.

Meeting notes Pricing Services and case studies PROPOSAL DRAFT
Before
  • The salesperson finishes a good meeting.
  • Now somebody needs to: find the proposal template,
  • copy customer information,
  • find pricing,
  • read the meeting notes,
  • write the requirements,
  • select relevant services,
  • find case studies,
  • create the document,
  • and check everything.
  • The proposal waits until tomorrow.
  • Or Friday.
With an agentic workflow
  • Meeting complete
  • Understand requirementsUse approved meeting notes/transcript.
  • Check informationCustomer, service, pricing, relevant business information.
  • Identify gapsAnything required before a proposal can be prepared?
  • AssemblePrepare the appropriate proposal sections.
  • Flag decisionsAnything requiring commercial judgement goes to a person.
  • Prepare draftReady for review.
Human handover
Proposal draft ready
✓ 7 sections completed.
2 decisions required
→ Final implementation price
→ Whether expedited delivery can be offered
[Review proposal]
Possible value
  • Faster proposal preparation.
  • Less copying.
  • More consistent documents.
  • Human time focused on commercial decisions.
Sensible starting authority
Read→ Prepare→ Human approval

The AI prepares. The person decides.

06

The CRM Agent

After the work happens

Keep the system updated without making people feed it all day.

CRM RECORD Last activity: 27 August Meanwhile, seven things have happened. CRM UPDATED ✓ Meeting summary added ✓ Next action created ✓ Follow-up date added
Before
  • Great customer call.
  • The salesperson has another call in five minutes.
  • They promise themselves they'll update the CRM later.
  • They don't.
  • Three weeks later the CRM says: Last activity: 27 August
  • Meanwhile, seven things have happened.
With an agentic workflow
  • Interaction happens
  • CaptureUse appropriate available information.
  • UnderstandWhat changed?
  • Prepare updateSummary, next action, agreed fields.
  • Check rulesWhich fields can be updated automatically?
  • UpdateMake permitted changes.
  • FlagAnything uncertain goes to a person.
Human handover
CRM updated
✓ Meeting summary added
✓ Next action created
✓ Follow-up date added
Needs review
Opportunity stage may have changed.
[Review stage]
Possible value
  • Better CRM information.
  • Less manual administration.
  • More reliable next actions.
  • Better context for people and AI.
Sensible starting authority

This can vary by field.

Routine notes: Act within limits
Important commercial fields: Recommend or approval

That's an important design point. An agent doesn't need identical authority across an entire system.

07

The Pipeline Attention Agent

Every morning

Instead of another dashboard, tell me what needs me.

This is one of my favourite small-business examples. Not because it's technically spectacular. Because it could actually be useful.

Before
  • Open CRM.
  • Look at pipeline.
  • Try to work out: What changed?
  • What hasn't moved?
  • Who replied?
  • Who needs chasing?
  • What's overdue?
  • What am I supposed to do today?
With an agentic workflow

The agent watches defined sales activity and starts the day with:

Possible value
  • Better prioritisation.
  • Less CRM archaeology.
  • Important work surfaces earlier.
  • Fewer things rely on memory.
Sensible starting authority
Read→ Recommend→ Prepare

An agent doesn't have to autonomously contact anybody to be valuable.

Human handover
Good morning
3 new enquiries
2 ready for review
1 needs more information
2 quotes
Follow-up due today
1 opportunity
Customer replied overnight
1 stalled deal
No next action recorded
4 routine actions
Already prepared
2 THINGS NEED YOU.

That's it.

08

The Sales Coordinator Agent

Across the whole workflow

The interesting bit may be what happens between your systems.

This is the more advanced example. Imagine the business uses:

Website.Email.CRM.Calendar.Documents.Proposal software.

None of those systems is necessarily the problem. The work between them is.

Website AIUnderstand CRM Business info Email Calendar AIMeeting brief Meeting AIActions Follow-up
An agentic workflow might coordinate:
  • WebsiteNew enquiry arrives.
  • AIUnderstands it.
  • CRMFind/create customer.
  • Business informationRetrieve relevant information.
  • EmailPrepare response.
  • CalendarOffer/book appropriate meeting.
  • AIPrepare meeting brief.
  • MeetingHuman has conversation.
  • AIPrepare actions and proposal.
  • CRMUpdate opportunity.
  • Follow-upMake sure next action happens.
Human handover
Human handover
Humans enter where their judgement matters.
The workflow handles more of the coordination around them.
Possible value
  • Less copying between systems.
  • Fewer broken hand-offs.
  • Less administrative coordination.
  • A clearer end-to-end process.
Sensible authority
Mixed

That's important. Different stages should have different permissions.

This isn't
AI runs sales.
It's
AI helps run the process around sales.
Worth noticing

Notice what these agents aren't doing.

None of these examples requires pretending AI has become a brilliant human salesperson. They're mostly doing things businesses already need done:

ReadingGatheringCheckingPreparingUpdatingRememberingCoordinatingEscalating

Sometimes acting. Sometimes not.

That is why agentic selling can be useful without the science-fiction version of autonomous selling being true.

A fair question

What about outbound prospecting?

Of course AI can also be used around outbound sales. It can potentially help with:

prospect researchaccount prioritisationpersonalisationoutreach preparationsequence managementresponse interpretationand follow-up

But there's a reason we haven't made: SEND 10,000 AI EMAILS one of our flagship examples.

Automating volume is easy to understand. Creating communication people actually want to receive is harder.

For many businesses, there may be much more immediate value in improving how they handle the demand and relationships they already have.

Where to begin

Which AI sales agent should you build first?

Not necessarily the one with the biggest list of features. Look for the job with the best combination of:

Frequency

How often does it happen?

Friction

How much time or effort does it consume?

Value

What improves if we do it better?

Information

Do we have what the agent needs?

Checkability

Can we tell whether it did the job properly?

Reversibility

What happens if it gets something wrong?

Measurement

Can we see whether the new process actually worked?

Then start with one.

An honest possibility

You may not need an agent.

There's another possibility. You map the process and discover: this doesn't need AI at all. Excellent.

Perhaps you need:

a CRM rulean email automationa better forman integrationa reminderor a simpler process

The objective isn't to justify an AI agent.

The objective is to fix the work.

A quick shortcut

Find the example closest to your business.

Enquiries are slow
Start with the Website Enquiry Agent.
Salespeople waste time on poor leads
Look at the Lead Qualification Agent.
Leads disappear after quotes
Look at the Sales Follow-Up Agent.
People spend ages preparing for calls
Look at the Meeting Preparation Agent.
Proposals take too long
Look at the Proposal Agent.
Nobody updates the CRM
Look at the CRM Agent.
It's hard to know what needs attention
Look at the Pipeline Attention Agent.
People are constantly moving information between systems
Look at the Sales Coordinator Agent.

That's where we'd begin the conversation.

From example to your process

What would an AI sales agent look like in your business?

It probably won't look exactly like one of these. Good. Your customers, systems, information and sales process are yours. We start with how the work happens today.

Then we work out:

Build an agentic sales workflow See how we build it →