Jeff Brokaw

Your AI Tool Says "Qualified." Sales Calls BS.

Jeff Brokaw · CMO and Certified Chief AI Officer ·

Your marketing team buys an AI tool. Now the qualified-lead count looks fantastic.

Before renewing the subscription, ask Sales to open the records.

Can those companies buy what you sell? Is the contact involved in that purchase? What did the system find that makes a conversation worth having now?

If the rep has to work all of that out again, the automation has handed over unfinished research. Putting a score beside the name doesn't finish it.

Before I approve more budget, I want to know whether we found better prospects or made it easier to earn the label.

What should AI lead qualification establish?

AI lead qualification should check a contact or account against the criteria for a defined next step and preserve the evidence behind that decision. An account worth researching, a lead Sales accepts for follow-up and an opportunity with a confirmed need are different stages. The software should say which stage it supports, what remains unknown and who handles the next action.

You don't need proof of an imminent purchase before making a useful introduction. You do need to stop reporting a researched contact as though a buying conversation has happened.

Check what the AI was told to predict

HubSpot's AI lead-scoring documentation says the team selects a lifecycle-stage transition and a timeframe for evaluating contacts. HubSpot then recommends scoring criteria and points from that contact data, which the team can review and edit.

If your team assigns a sales-qualified label before Sales has confirmed the qualification criteria, training around that transition can reward the existing labeling process. It doesn't independently establish that the resulting contacts will buy.

Check who changes the stage and what evidence they need. An automatically assigned label, an accepted conversation and a won deal cannot stand in for one another.

Check the records that failed to advance, too. If a capable buyer was never contacted, the lack of progression may reflect follow-up rather than poor fit. Don't teach the system to reject someone because your team never called them.

An agency implementing the tool should explain this choice to the client. Agree on the outcome being prioritized before building a report around it.

Company intent is not a confession from the person you emailed

Apollo's buying-intent documentation says its intent signals concern companies. A person returned in the search may work at that company without being the person who researched the topic.

That gives me a reason to research the account. It doesn't give me permission to email a stranger claiming I know what they've been reading.

Suppose a company is researching inventory software. That gives a supplier a reason to investigate. The relevant question might concern a new warehouse, a replacement system or an integration project. The contact you found could have no responsibility for any of it.

The research needs to identify the likely purchasing function and preserve what is still unverified. Otherwise the sequence begins with an assumption the recipient has to correct.

Data from a company, evidence about a person and a need confirmed in conversation belong in separate fields. Combining them into one high score makes it harder for the rep to see what the system knows.

A procurement title still needs a relevant company

My specialist-manufacturing work includes account research using public signals such as job postings and procurement activity. The purpose is to find businesses worth investigating before they submit a form.

At that manufacturer, annual inbound web leads grew from 43 in 2023 to 203 in 2025 during the commercial rebuild. Those are inquiry counts. They still need a qualification check before anyone puts them in the opportunity forecast.

A new procurement leader can be relevant. Before turning that change into outreach, I want to know whether the company uses the materials or capabilities the supplier provides. A promising title at the wrong kind of business still wastes the seller's time.

I also want the message to fit the work that person might control. Someone buying production materials needs different evidence from someone sourcing a software platform. A generic procurement label hides that difference.

The outbound demo on my Builds page uses synthetic prospects to show how account research informs a working proof built in the prospect's own brand.

The decision comes before the personalized asset. If the account doesn't qualify for the next step, I don't want the system spending more on enrichment, copy or delivery to make the wrong prospect feel researched.

For the accounts that deserve attention, the rep should be able to open the source and understand why the system selected them.

Give Sales a record it can use

An alert that says "high intent" leaves the recipient guessing. A useful handoff explains what happened and why it relates to the offer.

For an outbound research queue, I would include:

What the rep needs What the record should show
Account fit The specific product, service or customer condition that makes the business relevant.
Reason to investigate now A dated event or observed action, with its source.
Contact relevance The person's current role and the basis for connecting it to the purchase.
Existing relationship Current customer, open opportunity, previous rejection or another owner already working the account.
Missing information Questions the research did not answer.
Next action What the assigned person should do and what would justify advancing the record.

The exact fields depend on the business. A small agency selling a fixed audit needs a different qualification check from a manufacturer preparing a complex quote.

For an inbound request, I would preserve what the prospect actually asked for. Someone requesting a demonstration has given you different information from someone whose company appeared in a research feed.

The sales leader should approve the criteria for accepting each kind of handoff. Marketing should be able to challenge a rejection, too. "Bad lead" needs a recorded reason. Sales doesn't get to dismiss the work without explaining what failed.

Keep HubSpot. Fix what reaches it.

Use the CRM you already own if it can hold the criteria, evidence and routing you need. A clear rule and a saved view may solve the first version of this problem without another platform.

Changing the CRM won't repair research nobody did.

MoltSets for contact data. MoltSets offers prospect search and enrichment through API and MCP connections, including use in n8n. Use that layer to look up professional contact details and fill gaps in a record. Test its coverage against the accounts the business wants before choosing a plan.

Its current FAQ says the email grade concerns sending risk. It does not establish that an address belongs to the person you requested. A good grade cannot answer whether that person still holds the role, controls the purchase or needs your offer.

In a September 4 update from the second week of beta, founder Adam Robinson described complaints about deliverable addresses that did not match a contact's current role. Check employment separately before using that email to reach someone about a job they may have left.

Exa Websets for account research. A title and industry filter may miss the condition that makes a company worth calling. Websets supports natural-language searches and custom research columns. A test could look for manufacturers that publicly describe a capability relevant to the supplier's offer. Keep the supporting pages available so the reviewer can check what earned the match.

n8n for the handoff. It can connect the ownership check, research requests and review queue. Its human-review tools can pause an AI tool action for approval. Sales should receive the reason for the recommendation and the source, with a way to accept it or record why it falls short.

HubSpot can hold the result and the later outcome. Its AI-assisted contact scoring requires Marketing Hub Enterprise, but buying that feature is not a prerequisite for defining acceptance criteria or recording a review.

I'd test these components against a specific gap. If the company research is already solid and the only problem is an outdated email, I wouldn't rebuild the entire prospecting process.

A small n8n example you can inspect before connecting anything

Suppose a specialist manufacturer wants procurement contacts at companies that use its capabilities. The workflow starts with the account criteria and checks whether a rep already owns the relationship. It then needs current contact information and evidence that the business fits.

For a connected implementation, MoltSets could supply contact research, while Exa could help find public evidence about the company. n8n would preserve their separate results and send a review record to the assigned person. A missing answer would stay unresolved. It would not earn points because the AI sounded confident.

There are mundane ways to get this wrong. MoltSets documents that an unsuccessful lookup can return HTTP 200 with a not_found status. A workflow checking only the HTTP response could mistake a completed request for a found contact. Exa's Websets API starts processing a search when a Webset is created. That submission is not the finished research.

The downloadable n8n sample uses seven fictional records to demonstrate the review rules. They include a high-scoring account in the wrong industry, a deliverable email with an unconfirmed role and a company with an existing opportunity. Only one reaches human review. None is reported as Sales-accepted or given a pipeline value.

The sample runs manually on the fictional records. Provider request templates are disconnected and disabled. Review the decisions before connecting paid services or customer records.

Find out whether the lead count improved because the rule changed

Before comparing the new workflow with the old one, save the qualification criteria used by each.

Did the new system accept smaller businesses? Did it count everyone at a company showing intent? Did an unanswered question previously cause a record to wait, while the new workflow assumes a favorable answer?

Any of those changes can affect the total without demonstrating better demand.

Review records from both methods using the same acceptance criteria. Where practical, hide the generating method from the reviewer. Keep the review focused on the evidence that was available at the time of the handoff. A deal that eventually closed may make an old record look obvious in hindsight.

Include rejected records, not only the ones the AI selected. A system can produce a tidy list by excluding buyers whose needs are harder to research. Missing those buyers is a different failure from sending Sales too many weak contacts.

Then check what Sales did. An untouched record, a late response and a researched rejection tell you different things. A qualification workflow cannot be assessed fairly if one queue gets immediate attention and the other waits a week.

Use the marketing decision record to preserve the criteria and the outcome. When the team changes a rule, the next report should show the change instead of presenting the new total as a comparable result.

Spend one week checking the handoff before scaling it

Choose one offer and one customer segment. Marketing and Sales should agree on what qualifies a record for the next step, who handles it and how quickly they will review it.

Take a manageable batch from the new workflow and a comparable batch from the existing process. Check source accuracy and the time Sales spends repairing the research. Record acceptance and rejection reasons. Keep unknowns visible instead of turning them into a positive answer to make the workflow finish.

Use that first week to check how much research Sales has to repair and which rejection reasons keep coming back. Compare the time and cost per accepted handoff. If only a few records make it through, keep the test running before making a spending decision. A simple exclusion rule may remove enough waste to postpone buying more software.

One week will not establish revenue lift for a long sales cycle. Keep following the records into opportunities that meet your agreed criteria and track their later outcomes. Compare acquisition and serving costs where you can, and leave immature results pending rather than assigning them a convenient lifetime value.

Last week's reporting article describes the brief I would use for that decision. I'd want to see the accounts reviewed, what Sales accepted, what it rejected and the work required to reach that point. The number of rows the AI processed can stay in the operating log.

Before approving a larger rollout, open ten of the leads in the budget meeting. Have Sales show which ones deserve a call and why. Fix what keeps failing before you pay for another thousand.

Frequently asked questions

Why can AI-qualified leads fail to become sales opportunities?

The workflow may use criteria that don't match the sales handoff, outdated information or a company signal incorrectly assigned to an individual. Missing follow-up and Sales capacity can also prevent progression. Review the evidence, criteria and actions separately before concluding that the tool or the lead quality caused the problem.

What is the difference between AI lead scoring and lead qualification?

Scoring assigns a rank or value using selected criteria or a trained model. Qualification checks whether a record meets the requirements for a defined next step. A score can help prioritize that review, but it doesn't establish a confirmed customer need unless the underlying evidence supports it.

Does buying-intent data identify the person ready to buy?

Not necessarily. Some products measure research activity at company level. Apollo explicitly says contacts returned from those companies may not be the individuals who conducted the research. Check the provider's unit of measurement before attributing an action to a person.

Can a team improve lead qualification without buying another AI tool?

Yes. Start by defining the next step and its acceptance criteria in the existing CRM. Preserve source evidence, current ownership and rejection reasons. Buy additional research or automation only when a specific missing capability prevents the team from doing that work reliably.

What can a one-week qualification test establish?

It can reveal inaccurate research, inconsistent criteria, review effort and whether Sales receives usable handoffs. It normally cannot establish revenue improvement for a long sales cycle. Follow the same records through their later outcomes before claiming a commercial return.

Sources

Product documentation checked October 27, 2026. Capabilities described here are vendor-documented, not a comparative performance test.