How to agree on what makes a qualified lead

Why “poor-quality lead” explains nothing

If the cause is unknown, preserve that uncertainty in the data instead of replacing it with a convenient explanation

Does the person fit the business? Did the team reach them and understand their task? Did they buy? The questions are connected but not interchangeable. A suitable prospect may not answer. Someone with an unsuitable request may have a long, friendly conversation. A purchase can fail after the prospect has already qualified.

That makes “poor-quality lead” too broad to manage. It mixes customer characteristics, handling outcomes and sales results. First record what is actually known about each inquiry. If the cause is unknown, preserve that uncertainty in the data instead of replacing it with a convenient explanation.

Which criteria can be verified rather than guessed

Choose characteristics tied to your ability to serve the customer: the task, product, location and acceptable purchase conditions. When selling to another business, the person’s role in choosing a supplier may matter. An appointment-based service needs a different set of conditions. Each criterion needs evidence, such as the customer’s answer or an agreed field in the contact record.

Who records the rating matters too. HubSpot's default lifecycle stages, for example, separate a contact that marketing has qualified as ready for sales from one that sales has qualified as a potential customer. If both teams fill in one “qualified” field in your CRM, decide whose judgment it records.

How to test the rules on the same inquiries

The following is an editorial review protocol to test on your inquiries and agree with the sales lead. The Metrica documentation cited below supports offline-data import, not this assessment procedure.

The key condition: both people rate the same records without seeing each other's ratings. Where practical, hide the advertising source so an expectation that “this channel always sends poor leads” does not determine the assessment.

1. Use all inquiries from a defined period or a sample selected by a documented method, without picking the worst conversations.
2. Use an equivalent period after first contact, since recent inquiries may still be in progress.
3. Hide the inquiry source where possible.
4. Ask a marketing representative and a sales representative to record status, reason and evidence independently.
5. Combine the ratings in one table and mark the disagreements.
6. Review the disagreements together, clarify the rules and re-rate the disputed records.

An illustrative review: 20 inquiries, two ratings

It is too early to conclude that the channel produces a specific proportion of unsuitable inquiries

Consider an illustrative review of 20 inquiries with disagreement on six. In three, one person treats an unanswered call as evidence of poor quality while the other marks the status unknown. In two, they interpret the permitted location differently. In one, there is no evidence of the buyer’s budget. These are invented teaching figures, not company results.

In total, marketing counted 10 suitable, 4 unsuitable and 6 unknown. Sales counted 7, 9 and 4. That is where the “half” and the “roughly a third” at the meeting came from. It is too early to conclude that the channel produces a specific proportion of unsuitable inquiries.

How much agreement is enough: percent agreement and kappa

The simplest measure is the share of inquiries where the ratings match. Here it is 14 of 20, or 70%. But some matches may be chance: if both teams use one status often, they will agree on it even without a shared rule.

Cohen's kappa adjusts for that. It compares actual agreement with the agreement expected by chance given each rater's status shares, using (actual - chance) / (1 - chance). In the example, chance agreement is 32.5% and kappa is 0.56. Zero means agreement no better than chance; one means complete agreement.

On the scale Cohen proposed, 0.56 counts as moderate agreement. In a 2012 review, Mary McHugh calls that scale too lenient and classifies 0.40 to 0.59 as weak agreement in her own table. She also notes that percent agreement can overstate agreement, while kappa's assumptions may understate it. So report both figures.

What changes once the rules are clarified

The team agreed three rules. An unanswered call means “unknown for now”, not “unsuitable”. The permitted locations are written down as a list. Budget counts as confirmed only when the customer states it. After the disputed records were re-rated, 19 of 20 matched, or 95%, and kappa rose to 0.92. One disagreement remains: there is still no budget information.

Retain the earlier rating, revised rating and reason for change so the result remains auditable. Only now can channels be compared: both teams are counting the same thing.

Turn the findings into changes to forms, calls and tracking

Rules first, data transfer second

If the criterion is clear but the information is not collected, adjust a form field or conversation question. If the offer attracts the wrong task, refine its message. If a next step is never agreed, review the handling process. Assign one owner and an observable result to each change.

Sending confirmed sales stages from the customer-record system into analytics can connect advertising with events after the inquiry. Yandex Metrica documents this type of offline-data import and lists classifying form submissions among its uses. An offline conversion needs a visitor identifier, a goal of the “JavaScript event” type and the event time; the data appears in reports within 3 hours of upload. Meta describes a similar CRM integration through its Conversions API.

The integration transfers the statuses you define; it does not resolve an ambiguous qualification rule. Rules first, data transfer second.

1. Create a Metrica goal of the “JavaScript event” type for the agreed “suitable” status.
2. Make sure the CRM stores the visitor identifier from the visit: ClientID, UserID or yclid.
3. Upload confirmed statuses as a file or through the API.
4. Check in the Offline conversions report what share of records matched a visit.

What to show the owner, and how to avoid harm

A compact report can cover how inquiries were selected, how many could be assessed, where uncertainty remains, why ratings differed and which rules changed. Add percent agreement, kappa and a few anonymised examples with supporting evidence. A rejection count without an assessment method provides little help for a business decision.

Once the rules are aligned, compare equivalent groups by source, offer and language. Show underlying counts, the share still unknown and the observation window. A small sample can reveal a process problem without providing a reliable market estimate.

Use disagreement to improve the rule rather than rank employees. Under UK data protection guidance, personal data should be adequate, relevant and limited to what is necessary for the purpose, so bring only the customer details needed for the rating and limit access to the people in the meeting. This journal is published with the involvement of WEBNOSTRA; the case linked below is its own work.

Key decisions

1. separate customer fit, handling and the eventual sale
2. test the rules independently on the same set of inquiries
3. compare channels after the criteria are aligned

What to discuss with your team

• Criteria reflect the business task and verifiable information
• Unanswered contacts are separate from confirmed poor fit
• Both teams independently rate the same inquiries
• Percent agreement and kappa are calculated
• Selection method and observation window are recorded
• Changes have an owner and a verifiable outcome

Sources

1. Yandex Metrica: importing offline datayandex.com · checked 31/10/2026
2. Yandex Metrica: sending and processing offline conversionsyandex.com · checked 31/10/2026
3. McHugh M. L. Interrater reliability: the kappa statistic. Biochemia Medica, 2012pmc.ncbi.nlm.nih.gov · checked 31/10/2026
4. HubSpot Knowledge Base: Use contact and company lifecycle stagesknowledge.hubspot.com · checked 31/10/2026
5. ICO: Principle (c) Data minimisationico.org.uk · checked 31/10/2026
6. Meta: lead generation tools and CRM integrations, November 2023about.fb.com · checked 31/10/2026