Agentic GTM: turning customer signals into better sales decisions

The campaigns are running. The pipeline looks healthy. Too few opportunities are turning into customers.

Marketing wants better targeting. Sales wants better leads. Operations would quite like everyone to update the CRM.

Now there's another suggestion on the table: bring in an AI agent.

It could help. But before giving it a job, we need to understand what is happening. Are we attracting customers whose needs we can't meet? Are promising evaluations waiting for answers? Or are customers progressing while our records tell a different story?

Sending the next follow-up faster won't resolve all three.

This is where I think agentic go-to-market gets interesting. An agent could help investigate the problem, connect evidence across teams and support a useful response. The commercial ambition is better sales velocity, with stronger qualification and conversion doing much of the work.

That means recognising credible opportunities and helping those customers get enough confidence in the value to buy.

The plan is agreed. How well is it working?

Let's start with a B2B technology business that already has a go-to-market plan. The ideal customer profile, proposition, messaging and commercial targets are agreed. Marketing is generating interest and sales is working opportunities.

Some customers can try the product, achieve something useful and buy independently. Others need technical evaluation, stakeholder agreement and procurement support. A team that starts through self-service might later need help with a wider deployment.

Company size alone won't tell us which route is appropriate. A small business can have a difficult integration requirement. A team in a large enterprise might be able to make its initial purchase without speaking to sales.

Our starting point is a strategy in motion. Its assumptions remain open to challenge. If the target segment lacks the urgency or budget we expected, we'd want to discover that before investing more in the same approach.

What would better sales velocity look like?

Sales velocity brings four measures together:

Sales velocity = qualified opportunities × average deal value × win rate ÷ average sales-cycle length.

With cycle length measured in days, the result is a monetary value per day. It provides a way to compare pipeline performance; it isn't recognised revenue or a guaranteed daily sales forecast. HubSpot's explanation of sales velocity

The levers I want to focus on here are qualification and conversion. Are we finding customers with a problem worth solving and a credible route to purchase? Once we have their interest, are we helping them establish whether the product is right for them?

An agent could help recover a promising enquiry that never reached the right person. It could surface a requirement we can't meet, or identify the missing evidence holding up a buying decision.

Each finding changes where we put our effort. The commercial benefit depends on what we do with it.

Tighter qualification may leave us with fewer opportunities and a higher reported win rate. We haven't necessarily won any additional business. Better qualification earns its keep when it helps us recognise good opportunities, spend our time more effectively and improve the outcomes that follow.

Give the agent a clear job

By agentic GTM, I mean a workflow where an AI model can work towards a defined commercial task, choose from permitted tools, examine evidence and propose or take actions within agreed limits.

For example, we could ask it to investigate an apparently stalled evaluation. It might check the opportunity record, read relevant correspondence, inspect product activity and follow up a discrepancy before preparing a recommendation.

That ability to investigate is where I'd look for value. A polished account summary is useful, but I'd also want to know what the agent found that changes our next move.

Some jobs need very little reasoning. A rule can flag an overdue task or a missing owner. An agent becomes more interesting when the records disagree, the meaning of a customer exchange needs interpreting, or the next question depends on what it has just found.

Four places to look before turning up the activity

Poor conversion is a starting signal. It doesn't identify the cause. I'd look across four areas:

These problems can overlap. A customer has an integration question. Our handover fails. The question goes unanswered, and eventually the CRM describes the account as disengaged.

From the customer's perspective, they may simply be waiting for us to reply.

The opportunity that went quiet. Or did it?

Take a fictional company evaluating workflow software. The opportunity has no recent activity and the expected close date has passed. A re-engagement sequence looks like a reasonable next step.

Before drafting it, the agent checks the records it is allowed to access.

It finds product activity under a second account with a similar name. Recent correspondence says the customer is still evaluating. A support exchange contains a question about connecting the product to an existing system. Someone promised a technical response, but there's no recorded owner or next action.

That gives us a different explanation to investigate.

The agent still needs to check the connections. Similar names don't prove two records belong to the same organisation. The usage could come from another team, and the support question might have been answered elsewhere. Those gaps should be visible in its findings.

Once the relationships and outstanding question are confirmed, we can revisit qualification. What is the customer trying to achieve? Is the integration essential? Who will assess the result, and how would a successful evaluation lead to a purchase decision?

If there is a good fit and a credible reason to proceed, the next move is to assign a technical owner and agree what the evaluation must demonstrate. The agent could prepare the brief, bringing together the customer's requirement, relevant product evidence and the questions still to answer.

If we can't support the integration, an honest answer may save both sides a great deal of time. The agent has no licence to invent a workaround or promise something on the roadmap to keep the deal alive.

This is the connection between qualification and conversion. Establish whether we can deliver the outcome, then help the customer find out whether we can deliver it for them.

Another email asking if they're still interested would be a fairly awkward contribution.

This is an illustrative scenario, not a result from a deployed agent. It shows the kind of investigation I'd want to test.

We don't have to wait for perfect CRM data

Improving the information can be part of the agent's job. It could find inconsistencies, suggest missing account relationships and prepare corrections with links to the supporting evidence.

In our example, it could put a possible duplicate in front of the account owner, draft an opportunity update from the customer exchange and prepare a handover for the technical question.

The checks matter because these actions have different consequences. A possible duplicate needs confirming before records are merged. Renewed activity alone doesn't justify adding a deal to the forecast. A customer's preferred date isn't a delivery commitment we've accepted.

Decide what the agent may inspect, recommend and change. Assigning an internal task under tested routing rules may be straightforward to automate. Record merges, discounts, customer commitments and forecast changes need controls appropriate to their impact. Access should stay within the scope of the task too.

We can give an agent useful work to do while keeping those boundaries clear.

A completed demo isn't a buying decision

Marketing-qualified leads, sales-accepted leads, demos and proofs of value help teams organise work. The labels become less helpful when we assume they tell us how far the customer has progressed.

The demo happened. Did it address the problem that brought the customer to us? The proof of value has started. Have we agreed what success looks like, and who can approve a purchase if it succeeds?

An agent can pull stated requirements from conversations and highlight questions that still need asking. Its output should distinguish what the customer confirmed from what the salesperson assumed and what the model inferred.

Incomplete information is a reason to investigate. Automatically rejecting accounts with gaps would risk favouring tidy records over worthwhile opportunities.

Self-service signals need the same care. Frequent logins might mean enthusiastic adoption, repeated difficulty or several people exploring separately. A pricing-page visit tells us something, but very little about budget or authority to buy.

Elena Verna makes a useful point in her discussion of sales problems in product-led growth: treating every signup as a sales lead can interrupt the experience that attracted the user in the first place. Her examples concern PLG businesses developing enterprise sales motions. For our purposes, the question is whether sales involvement would help this customer now. Elena Verna: sales problems in PLG

Sometimes the customer is doing perfectly well on their own. That's a good outcome too.

What if the plan needs changing?

Resolving one missed handover helps one account. Patterns across accounts can tell us something more significant.

Suppose we repeatedly attract customers who need a capability we don't offer. Our messaging might imply that it exists. We might be reaching the wrong use case. There could also be a worthwhile product opportunity.

Before changing direction, examine the pattern across a defined group, including accounts that progressed successfully. Which segments are affected? What outcome do they want? What alternatives do they use? Are a few vocal prospects dominating the conversation?

An agent could assemble that evidence and help expose contradictions between our positioning and the expectations it creates. Product and commercial leaders would then have a firmer basis for deciding whether to change the message, the target audience or the offer.

An agreed strategy should give the team direction without making contrary evidence unwelcome.

Give the account owner something they can use

For the fictional evaluation, I'd want a short action brief covering:

• What we found: the confirmed customer exchange, relevant usage and unanswered request, with sources and dates.
• What it suggests: the evaluation appears to need technical assistance, with any competing explanation made clear.
• Why the opportunity remains credible: the customer outcome, essential requirements and known buying process, including gaps.
• What happens next: the proposed response, its owner and any approval required.
• What would count as progress: the evaluation demonstrates the agreed outcome and the customer confirms their next decision.

The integration's feasibility and any unresolved buying requirements should remain explicit. The account owner needs enough information to act and enough visibility to challenge the recommendation. A confidence score on its own wouldn't give them that.

Better qualification, stronger conversion. Can we show it?

Start by checking whether the agent's conclusions stand up. Give it cases where similar company names refer to different organisations, older correspondence has been superseded, or strong usage needs no sales intervention. Include cases where the right answer is to gather more information.

Compare it with a simpler approach, such as existing routing rules or a structured account review. Check accuracy, source use, proposed actions and the time people spend correcting its work. Fast preparation is worth less if the account owner has to redo the investigation.

Then follow the effects through to qualification and conversion.

For qualification, review both accepted and rejected accounts. Did the agent recognise credible opportunities? Did it overlook prospects with incomplete records? Where did the team redirect the time it saved?

For conversion, track whether customers completed the agreed evaluation and whether qualified opportunities became wins. An updated stage is something to inspect. The customer's decision provides stronger evidence.

Sales velocity helps bring those results together, provided the comparison is consistent. Keep segments and buying motions comparable, and separate new business from expansion where their economics differ. Self-service needs appropriate activation and purchase measures; every signup shouldn't become a sales opportunity for the sake of the calculation.

Keep the qualification criteria, cycle start and end points, and deal-value basis consistent. If a definition changes, show its effect separately. Preserve the original groups of opportunities and their outcomes so removing weak deals doesn't erase losses from the assessment.

This is particularly important when cleaning the CRM. Correcting duplicates and stale dates can change the numbers without changing what happened commercially. A higher win rate might also reflect a smaller denominator. Those are reasons to understand the movement before celebrating it.

Allow enough time for buying decisions to play out. Compare similar groups, using a controlled rollout where practical, and account for changes in pricing, campaigns, account mix and sales effort. Review all four sales-velocity components alongside actual wins, margin and cost to serve. A larger contract with disproportionate custom work, or a faster signature bought through a heavy discount, may be a poor trade.

The aim is to establish whether better qualification and customer support translated into better business. We should be able to explain the contribution beyond an improved dashboard.

Start with one decision worth improving

For the first implementation, I'd keep the brief focused: which apparently stalled evaluations remain credible opportunities, and what evidence or assistance do they need next?

Agree what the agent can access and do. Test realistic cases before allowing changes to live records or customer communications. Expand the scope when the results justify it.

Back in the pipeline meeting, marketing, sales and operations may each have spotted part of the problem. Connecting their evidence could reveal an overlooked opportunity, a failed handover or a mismatch between the promise and the product.

That's where I'd want agentic GTM to earn its place. Help us recognise customers we can serve well, give them the evidence and support to make a buying decision, and learn from the opportunities that don't convert.

If that improves sales velocity, we'll have a useful commercial story to tell. And a better explanation than simply saying we sent more emails.

Better qualification is only the beginning. In the continuation, I explore how AI could improve discovery and solution design, help qualified prospects buy and reduce the cost of sale.