A Lead Is Not Qualified Until Sales Can Verify the Handoff

A marketing-qualified label should authorize a review, not force a sales call. A useful handoff contains enough evidence for a representative to decide whether the account fits, whether the person’s professional role is relevant, what happened, what remains unknown, and whether the next contact is permitted. If one of those facts is missing, the correct output may be hold or recycle, even when a platform score crosses its threshold.

This distinction matters because a lead can be commercially interesting and still be unsafe to contact, while a legally contactable record can be commercially irrelevant. The operational answer is a small evidence card and a deterministic disposition: accept, recycle, reject, or hold. The 2026 France B2B lead-generation overview supplies the channel-planning baseline; this guide adds a reviewable handoff test that separates qualification evidence from permission to act.

The example asset below is a synthetic 12-case test set authored on 4 November 2026. It contains no real leads, CRM exports, response rates, or conversion measurements. Its purpose is to expose ambiguous rules before a team routes live records.

The decision in one sentence

Send a lead to a named seller only when there is enough evidence of account fit, role relevance, a useful business context, an accountable owner, and a permitted next action. Otherwise preserve the record with a reason and a next review condition. “More leads” is not the objective of this gate. The gate aims to make the next action explainable and to prevent an automation score from hiding missing facts.

A qualification label should describe what the organization knows, not what it hopes. “Visited a pricing page” is an observed event. “Ready to buy” is an inference. “The operations director at a 200-person logistics company requested a demo for a named workflow” combines role, account context, and an explicit request. These are not interchangeable statements, and a system should store them separately.

The disposition does not predict a sale. It decides who may do what next. An accepted record can still fail to convert; a recycled record can later become actionable; a rejected record may be a duplicate or outside the target market; a held record requires a human or compliance check. Those outcomes should not be collapsed into a single score or treated as proof that marketing or sales performed badly.

Why a score alone is a weak handoff

A numerical score can help rank records when its inputs and use have been tested. It cannot explain whether a person is the right contact, whether the recorded source is reliable, or whether an objection has been honored. Two records can share a score while having radically different evidence: one may contain a direct demo request and a known role; another may contain repeated page views from an unidentified visitor. Ranking them equally conceals the very distinction a seller needs.

An academic study in Junior Management Science examines how marketing and sales collaboration relates to perceived lead quality and follow-up, and reports that joint planning and information exchange are associated with greater acceptance. The result supports the value of shared definitions, but it does not establish that any particular score or handoff card causes more revenue in every organization. See Hilke, “When Does Marketing & Sales Collaboration Affect the Perceived Lead Quality?”, published in 2024. Its construct is perceived quality and organizational collaboration, not a universal conversion benchmark.

A 2026 preprint proposes a hierarchical preference-ranking approach for lead scoring using data from one automotive brand. Its authors report model and online-test outcomes, but the setting is a single company and sector; the work is not independent validation for French B2B consulting or a general accuracy guarantee. See the arXiv preprint. The useful inference is modest: prioritization can be modeled, but a model’s reported performance must stay attached to its population, label definition, and test design.

The four dispositions

Disposition	Meaning	Required next action	Typical reason
Accept	Evidence supports a human sales review now	Route to one named owner with a due condition	Relevant request, fit, role, and usable context
Recycle	Potential fit, but timing or evidence is incomplete	Return to a named nurture path with a review trigger	Need plausible, timing absent, permission gate passed
Reject	Evidence shows the record is outside scope or invalid	Suppress from the sales queue; retain only necessary reason	Duplicate, wrong geography, irrelevant offer, invalid record
Hold	A required fact or control is unresolved	Block outreach and assign a reviewer	Unknown source, objection conflict, legal or identity uncertainty

“Hold” is deliberately distinct from “reject.” A missing data-origin field is not proof that the person is unsuitable; it is a block on action until the organization can resolve the missing information. Conversely, an explicit opt-out should not be placed into a routine nurture sequence while someone investigates commercial fit. A suppression decision must be applied before any campaign or task is created.

“Recycle” also needs a specific return condition. “Try again later” is not a condition. “Revisit after the budget cycle the contact named in the discovery form” is. If a company cannot identify a responsible owner or a review date, it has not designed a recycle path; it has created a parking lot that can silently become stale.

Keep commercial qualification separate from contact permission

Marketing qualification and permission to contact answer different questions. Qualification asks whether the record appears relevant to a commercial process. Contact permission asks whether the proposed processing and communication can lawfully proceed under the applicable jurisdiction, channel, source, notice, and objection controls. A high fit score must never override a suppression flag.

For France, the CNIL’s guidance dated 10 June 2026 says that electronic B2B prospecting may rely on legitimate interest when the solicitation relates to the professional activity of the person contacted. It also says professionals must be informed about the use of their contact details and able to object simply and freely; where data are already held or acquired from third parties, the organization should ensure the person has been informed about the possible use for prospecting and can oppose it. Each message must identify the sender and provide a simple way to refuse further solicitations. Read the CNIL guidance on email, SMS/MMS, and automated calls alongside its electronic communications rules.

That statement is bounded. It is about the described French rules and channels; it does not establish that every business address, scraped profile, purchased list, call, postal message, or cross-border campaign is permissible. The campaign owner must check the actual channel and collection context, current law, notices, and organization policy. This article is an operational guide, not legal advice. When the record’s provenance or lawful basis cannot be established, use hold, not a guessed answer.

The European Data Protection Board’s Guidelines 1/2024 describe legitimate interest as a case-specific three-part assessment: identify a lawful, clearly articulated, real interest; assess necessity; and balance that interest against the person’s interests, rights, and freedoms. The balancing must consider context, impact, reasonable expectations, and mitigation. See the EDPB Guidelines 1/2024. A checkbox labelled “B2B” does not perform that assessment, and a prospecting justification for one channel does not automatically authorize every downstream use.

A minimum evidence card

A compact card should answer the questions a seller would otherwise have to reconstruct. Keep direct observations, inferred judgments, and unknown values visibly distinct. Preserve timestamps and source references for events; do not store a long narrative where a structured field is sufficient. The purpose is not to gather every possible data point. It is to supply the minimum context required for a justified next step and an auditable reason when the gate blocks.

Field	Example of acceptable evidence	Weak substitute	Gate use
Account	Company, market, size band, geography, and source	“Enterprise” with no definition	Fit and territory
Person and role	Role relevant to the stated workflow; source and date	Seniority guessed from a title fragment	Professional relevance
Trigger	Requested demo, named problem, or explicit question	Page view interpreted as intent	Timing and next action
Provenance	Form, event, referral, or supplier record and collection date	“Imported”	Notice and source review
Contact controls	Notice version, lawful-basis review pointer, consent where required, suppression result	Global “marketing allowed” flag	Stop or release
Seller context	Product/use case, prior conversation, and missing fact	Raw event log with no summary	Useful first contact
Ownership	One person or queue, due date, and SLA definition	Shared inbox with no service target	Accountability
Disposition	Accept, recycle, reject, or hold plus reason code	Score only	Audit and learning

A record should not include sensitive details merely because a form collected them. Data minimization applies to the card: retain what is necessary for the defined business purpose, avoid copying unneeded personal narrative, and restrict access according to the job. If a seller needs a small excerpt to understand a request, preserve the relevant excerpt and its source rather than copying an entire message chain into every downstream tool.

A field called source is not enough if it only says “website.” Distinguish first-party form, event registration, partner referral, enrichment vendor, public business directory, and manual entry where those categories alter expectations or review. Record when the source was obtained and what notice was displayed. If the information came from a third party, record enough lineage to answer what the person was told and where the data travelled. Unknown should be a valid value that blocks an action; it should not be silently converted to “yes.”

The decision matrix

Use gates in order, not a weighted average. First check suppression and permission prerequisites; then validate identity and provenance; then assess commercial fit and context; finally assign an owner and next action. A failed early gate cannot be compensated for by a strong score on a later gate. This “fail closed” sequence is especially important when importing data from more than one source or when separate tools maintain their own opt-out state.

Check	Yes	No	Unknown
Suppression or objection clear?	Continue	Reject/suppress outreach	Hold and reconcile lists
Source and notice traceable?	Continue	Hold	Hold
Proposed channel and purpose reviewed?	Continue	Hold or reject per review	Hold
Account matches a defined segment?	Continue	Reject or route elsewhere	Recycle if resolvable; otherwise hold
Person’s role relates to the stated offer?	Continue	Reject or identify a role-based route	Hold; do not infer from company alone
A real business context is documented?	Continue	Recycle or hold	Recycle only with a valid trigger
Owner and next action exist?	Accept to named queue	Do not accept	Hold

A “yes” is not automatically a fact. It should point to a field, source, and timestamp. “Fits” could mean the account is in a supported geography and has the workflow the product serves; “role relevant” could mean the person owns, evaluates, or influences that workflow. Write definitions that can be applied consistently by two reviewers. If the organization cannot define these terms without reference to a particular person’s intuition, the model is not ready for automation.

A business may add thresholds after it has enough trustworthy outcomes to evaluate them. Before then, arbitrary point weights create the appearance of measurement without calibration. If the team uses a model, version the input set, target label, training window, evaluation population, exclusions, and human override reasons. Track false positives and false negatives separately; “accepted by sales” measures acceptance, not customer value. Revisit the threshold after a defined observation window and with a review of drift, not because one executive dislikes this week’s queue.

A synthetic 12-case test asset

The following test set is a worked example, not research on actual customers. Twelve scenarios were hand-authored on 4 November 2026 to test the order of gates and expose ambiguous records. The sample size is exactly twelve fictional cases; there were no people, companies, emails, calls, sales outcomes, or CRM rows in the exercise. Exclusions include real campaign performance, jurisdiction outside the France-focused examples, legal review of any named organization, and model validation. Therefore no pass rate, accuracy percentage, or commercial uplift should be inferred from the examples.

The cases intentionally include traps: a perfect company fit with a recorded objection; an explicit demo request with unknown third-party provenance; a relevant role that never stated a need; and a complete record with a named owner. Their outputs are expected dispositions under the rules in this guide, not empirical labels. Teams can replace the fiction with anonymized fixtures after confirming lawful use, access controls, and a safe test environment; production personal data should not be copied into a public sample.

// Synthetic acceptance-gate fixtures, authored 2026-11-04.
// No real people, organisations, outreach, or outcome data are represented.
const cases = [
  { id: "S01", suppressed: true, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "demo_request", owner: true, expect: "REJECT" },
  { id: "S02", suppressed: false, sourceKnown: false, noticeKnown: false, channelReviewed: true, fit: true, role: true, context: "demo_request", owner: true, expect: "HOLD" },
  { id: "S03", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: false, fit: true, role: true, context: "demo_request", owner: true, expect: "HOLD" },
  { id: "S04", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: false, role: true, context: "demo_request", owner: true, expect: "REJECT" },
  { id: "S05", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: false, context: "demo_request", owner: true, expect: "RECYCLE" },
  { id: "S06", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "none", owner: true, expect: "RECYCLE" },
  { id: "S07", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "demo_request", owner: false, expect: "HOLD" },
  { id: "S08", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "demo_request", owner: true, expect: "ACCEPT" },
  { id: "S09", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "named_problem", owner: true, expect: "ACCEPT" },
  { id: "S10", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "timing_unknown", owner: true, expect: "RECYCLE" },
  { id: "S11", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "demo_request", owner: true, expect: "ACCEPT" },
  { id: "S12", suppressed: false, sourceKnown: true, noticeKnown: true, channelReviewed: true, fit: true, role: true, context: "named_problem", owner: true, expect: "ACCEPT" }
];

function decide(x) {
  if (x.suppressed) return "REJECT";
  if (!x.sourceKnown || !x.noticeKnown || !x.channelReviewed || !x.owner) return "HOLD";
  if (!x.fit) return "REJECT";
  if (!x.role || x.context === "none") return "RECYCLE";
  if (x.context === "timing_unknown") return "RECYCLE";
  return "ACCEPT";
}

for (const item of cases) {
  const actual = decide(item);
  if (actual !== item.expect) throw new Error(`${item.id}: ${actual} != ${item.expect}`);
}
console.log(`${cases.length}/${cases.length} synthetic rule checks passed`);

The expected output is 12/12 synthetic rule checks passed. That output describes the consistency of this short code example against its hand-authored expected results only. It says nothing about production recall, precision, fairness, conversion, lawful basis, or whether a person is actually interested. The rule is intentionally easy to inspect and too simple to replace human judgment on complex cases. Real teams should add fixtures for duplicates, subsidiaries, partner-originated data, changed job roles, incomplete notices, channel restrictions, and conflicting suppression records before any automation change.

The fixture’s channelReviewed boolean is an abstraction. A production implementation should avoid using one universal flag that cannot show which country, channel, message type, purpose, notice, and policy version were considered. Similarly, sourceKnown should refer to a provenance record, not a free-text guess. A held record needs a reason code and reviewer. If those fields do not exist in the CRM, the safest immediate output is to block the route while the schema is corrected.

The handoff workflow

flowchart TD
  A[Capture event and source lineage] --> B{Suppression or objection?}
  B -->|Yes| C[Block outreach and record suppression]
  B -->|No| D{Notice, channel and purpose review complete?}
  D -->|Unknown| E[HOLD: compliance or data steward]
  D -->|No| E
  D -->|Yes| F{Account and role fit evidenced?}
  F -->|No fit| G[REJECT or route to another segment]
  F -->|Unknown| H[RECYCLE only with a review trigger]
  F -->|Fit| I{Need or explicit next step documented?}
  I -->|No or timing missing| H
  I -->|Yes| J{Named owner and useful context?}
  J -->|No| E
  J -->|Yes| K[ACCEPT to named seller queue]
  K --> L[Seller accepts, rejects, or returns with reason]
  L --> M[Review outcomes by source and cohort]

The flow makes the order visible: contact controls precede sales convenience. It also makes data quality operational. A line can be accepted only after a named owner exists; otherwise it cannot be observed as a fair SLA failure because no one owns the clock. The final review loops back to definitions and routing, not to automatic score inflation. Any line from HOLD to ACCEPT must correspond to a documented resolution of its hold reason.

A routing system should keep separate timestamps for event creation, qualification, queue entry, first human review, first permitted contact, and disposition. Those timestamps answer different questions. A campaign can generate a prompt response but still have a long delay before a seller sees it; a seller can review quickly but decide that fit is weak. Combining those intervals into one “speed to lead” measure obscures where the process failed.

Service levels without fake universal benchmarks

A team needs a response expectation, but it should define one from its own coverage, operating hours, urgency, and channel obligations instead of borrowing a universal number from a vendor post. State when the clock starts, what counts as a staffed period, how weekends and public holidays work, what qualifies as first response, and whether an automated receipt counts. A response target is a local service design, not a law of conversion.

Set a pilot SLA as a hypothesis: for example, “during staffed hours, the assigned queue reviews an accepted record by the next agreed service interval.” Choose an interval the team can actually meet, label it as a pilot target, and measure it. If a single fixed target would punish different time zones or request types, use separate service classes. Do not make someone wait for a non-urgent newsletter click as if it were a direct request for a product consultation.

Measure at least four separate quantities: queue acceptance rate, share of accepted records actually reviewed inside the defined interval, share of reviewed records returned with a reason, and share of accepted records progressing to a mutually agreed commercial conversation. Use explicit denominators and group records by source, segment, campaign, and time period. A dashboard showing counts without denominator, exclusions, and cohort rules is not decision evidence.

Avoid claiming that a shorter interval caused a higher conversion rate merely because the two move together. High-intent requests may both receive faster follow-up and convert more often; team capacity, product fit, geography, and seasonality can affect both. To assess a change, use a predeclared comparison or randomized operational experiment where ethical and feasible, protect urgent requests from delayed treatment, and report the allocation, period, exclusions, and outcome definition. If a causal design is unavailable, call the result an association and say what else may explain it.

What the evidence does and does not establish

The target page describes channel strategies and presents a company case. Publicly visible material does not provide enough sample construction, time period, attribution rules, or control details to reproduce its reported uplift independently. This is a verification limit, not evidence that the case is false. A reader should treat it as a claim from the publisher until the underlying measurement design is supplied. That is why this article does not repeat the number as a benchmark.

The 2024 academic paper supports a narrower point: collaboration and information exchange can influence sales’ perceived quality of marketing-generated leads and follow-up. The outcome is organizational and perception-oriented; it does not settle which evidence fields are sufficient for every team. The EDPB source supports a legal-method claim about assessing legitimate interest, but it does not prescribe a CRM schema or prove that a particular campaign’s balancing test passes. The CNIL source describes France-specific rules for the listed communications and the professional-relevance condition; it does not generalize to consumer marketing or other jurisdictions.

McKinsey’s 2026 B2B sales commentary reports survey-based observations about how growth-oriented organizations connect market mapping, AI, and sales workflows. It is consultancy research and reported practice, not a randomized causal test of this handoff protocol or a French SME benchmark. See McKinsey’s B2B sales analysis. The recommendation here is therefore to use its operating-model perspective as context, not borrow a promised productivity percentage.

Questions practitioners actually ask

Public discussions about lead qualification repeatedly surface practical questions that generic channel lists do not settle: what should a team do when it has no historical conversion data; how should a seller return a lead without starting a dispute; what is a defensible review window; and why does a record look good in a score but stall in conversation? These are useful prompts for designing the process, not representative survey findings. Forum anecdotes cannot establish prevalence or causation.

The missing operational unit is often not another scoring field but a shared disposition contract. Marketing needs to know what evidence a seller requires. Sales needs to return a specific missing fact or mismatch. Data operations needs a stable reason code. Privacy or compliance needs a separate path for source, notice, lawful-basis and objection questions. If each function receives only the “MQL” label, the organization cannot tell whether the disagreement concerns fit, timing, proof, ownership, or permission.

A practical taxonomy can begin with a small number of reasons: wrong segment; wrong role; duplicate; invalid contact; no demonstrated need; timing later; insufficient context; no owner; source or notice unresolved; objection/suppression conflict. Do not use “bad lead” as a catch-all. Each reason should have a designated owner and a clear consequence. For example, “timing later” returns to nurture with a contactable status and review date; “source unclear” enters a compliance hold and must not be placed in a marketing sequence.

Roles, permissions, and audit trail

A lead lifecycle usually touches multiple tools: form provider, analytics, enrichment service, marketing automation, CRM, and sales engagement software. One tool’s delete or suppression state may not automatically update every other copy. Map where personal data are created, enriched, synchronized, and used. A system owner should test how an objection flows through that map, including queues, exports, and vendor connectors, before calling suppression complete.

Assign authority carefully. Marketing can describe a campaign event and the notice shown at collection. A data steward can validate provenance and field lineage. Privacy or legal staff can review the applicable basis and channel policy. Sales can assess account context and the usefulness of a next conversation. RevOps can own definitions, routing, and reporting. One person may hold several roles in a small organization, but the evidence and decision should still be distinguishable.

Log the rule version used, who resolved a hold, what changed, and when the record became eligible for a new action. Avoid storing sensitive legal narratives in a broad sales note. The audit record should show the outcome and source of the review with access limited to those who need it. Retention periods should be specified for the card and event history; do not retain an indefinite trail simply because storage is cheap.

Where an automated model ranks or rejects records, keep a human review and appeal path for affected operational decisions. Review performance across segments that matter to the organization, check whether missing data systematically lowers a group’s score, and document overrides. Model inputs should be limited to relevant, explainable signals. A person’s job title can be stale; a company’s size band may be an estimate; behavioral tracking may be noisy. None should be treated as certainty without provenance and freshness checks.

Failure modes and edge cases

A click becomes intent. Repeated page views can be generated by research, internal testing, shared devices, or accidental visits. Record observed behavior, but do not claim a buying project unless a person supplied that context. A direct request can justify a human review; a passive event alone may be a weak reason for an intrusive sequence.

A good fit overrides an objection. This is a control failure. Suppression must run before scoring, routing, and campaign enrollment. If two tools disagree, hold the record and reconcile the identifiers. Never “temporarily” bypass an objection to allow the seller to verify fit.

The person left the role. An old title can make an account look relevant even after responsibilities changed. Store the source date, verify material role facts through a permissible process, and route uncertainty to hold or a role-based contact path. Do not infer personal contact permission from an account-level contract or another employee’s request.

A partner sends a list without lineage. “Partner lead” is not a complete source record. The receiving organization should verify what notice and information were provided, what onward use was described, and what objection mechanism exists. If those facts cannot be confirmed, do not treat the source as clean because the partner says it is.

A recycle queue becomes permanent storage. A recycle disposition without owner, date, and trigger can create an uncontrolled database. Define when the record is reviewed, what event reopens it, and when it is removed or suppressed under policy. A trigger should not silently authorize a message if contact controls have changed.

A service-level target is gamed. If an automated acknowledgement counts as a seller response, the dashboard may improve while the person receives no useful human action. Define the event precisely and sample the actual work. Measure first meaningful review separately from receipt delivery, and audit a small set of records against the underlying timestamps.

The model learns a biased proxy. Past “accepted” labels can reflect team habits, territory, or campaign access rather than eventual value. A model optimized against acceptance may reproduce the same disagreement. Compare labels with downstream outcomes and examine missingness, but avoid claiming fairness from a single aggregate metric.

A practical pilot plan

Start with one segment and one inbound source, not every channel. Write the segment boundary, permitted use, event definition, evidence card, disposition reasons, owner, and review interval. Have marketing and sales jointly review a small batch of anonymized or synthetic records. Ask each reviewer to decide independently, then compare the reason for disagreement. The goal is to discover ambiguous definitions before the process is wired into automation.

Next, run the gate in shadow mode. The system records its proposed disposition while the existing process continues under appropriate safeguards. Review conflicts with a human, especially suppression and provenance mismatches. Do not let shadow mode trigger outreach. Check whether the proposed outputs can be explained from source fields and whether every hold has a named resolver. If the source lacks an explicit notice or permission attribute, fix collection and lineage rather than inventing a value to clear the queue.

Only after the definitions and controls work should the team route accepted records automatically. Keep hold and suppression paths deterministic and fail closed. Give sales one clear action: accept with owner, return with a reason, or escalate a control issue. Do not let the system silently reassign rejected lines to another campaign. Reassess the process after enough records have been observed to review operational failure modes; do not pretend that a tiny pilot proves impact.

A useful pilot report includes population, date window, source and segment, the number of records entering each disposition, duplicate handling, missing-field rate, seller acceptance, timely review under the local SLA, return reasons, and any objection or provenance incidents. Include the number excluded and why. If the team cannot publish even an internal report with denominators and exclusions, it should not make external performance claims.

Measurement without confusing outputs

Use an event model with separate fields for observed action and derived state. For example, form_submitted_at is an event; account_fit_status is a judgment with a rule version; sales_disposition is a human or system decision; first_meaningful_review_at is an operational outcome. Never overwrite the original source event with a later score. Preserve the dates needed to reconstruct the state at handoff.

A ratio such as accepted / submitted is interpretable only if its cohort and denominator are explicit. Does “submitted” include spam, duplicates, consumer inquiries, and out-of-region records? Are records deduplicated by person, account, or request? Are late-arriving events included? A different denominator can tell a different story. Publish the definition beside the figure, and retain a query or versioned report that reproduces it.

Report both friction and leakage. A very restrictive gate can reduce poor handoffs but miss legitimate opportunities; a permissive gate can raise seller workload and expose people to irrelevant contact. Compare returned reasons, time spent reviewing, progression to agreed conversations, and the share blocked for unresolved controls. Treat any result as specific to the segment, period, and definitions. When volume is low, show counts and uncertainty instead of percentages that appear more stable than the evidence.

If testing a new evidence card against an old process, predefine what changes, what remains constant, and the primary outcome. Avoid changing the marketing campaign, sales staffing, scoring threshold, and outreach copy at the same time if the goal is to understand which change mattered. Where random assignment is inappropriate, a carefully documented before/after comparison can still inform operations, but confounders limit causal interpretation. State that limitation plainly.

Proof pack

Claim	Source	What it supports	Verification limit
France B2B electronic prospecting requires professional relevance, information, and simple objection capability in the described cases	CNIL, guidance updated 10 June 2026	The France-focused contact-control gate described above	Scope is the listed channels and circumstances; verify current law and campaign facts
Legitimate interest requires a three-part, case-specific assessment and balancing	EDPB Guidelines 1/2024	Why a generic “B2B” flag is insufficient	Does not decide whether a particular controller’s test passes
Marketing-sales collaboration is related to perceived lead quality and follow-up	Hilke, 2024 academic article	Shared planning and information exchange are relevant process variables	Findings concern the study’s constructs and sample; no universal revenue benchmark
Model ranking results depend on context and evaluation design	Authors’ 2026 arXiv preprint	Example of a proposed model evaluated in one automotive setting	Preprint and single-company context; not independent validation for French B2B services
B2B playbooks increasingly connect market mapping, AI and sales operations	McKinsey, 2026	Consulting perspective on connected commercial workflows	Survey/consulting claims are not causal proof of this handoff design

The proof pack separates a rule source, research, and business interpretation because they answer different questions. Legal guidance does not validate commercial effectiveness. A research paper does not create legal permission. A consultancy report does not establish that the example’s operating model fits a small French firm. Keep each claim close to the source that supports it, and retain its boundary in the sentence rather than hiding all caveats in the final paragraph.

Method, scope, and limitations

Research was conducted on 4 November 2026. Topic selection used a targeted Google Search Console page-and-query review for the exact French source page, after a broader position check. The search-result review covered the target article and competing operational guides; multi-angle searches included official privacy guidance, collaboration research, model-scoring research, consulting commentary, and practitioner questions. The internal search data selected an intent; its private impressions and position are not evidence of lead-generation performance and are not reproduced here.

The target page was read as the primary editorial baseline. Its channel and strategy coverage is useful for orientation; its public presentation of an internal company result does not expose enough underlying sampling and attribution detail for independent reproduction. This is a verification limit, not evidence that the case is false. Search-result snippets and page views were not counted as evidence of outcomes. Public forum questions were used only to identify unresolved practitioner tasks, not as a representative survey. A webinar agenda surfaced a similar handoff question, but a transcript was not verified and no factual claim relies on it.

The original asset is the twelve hand-authored synthetic cases and the rule function included above. The method was scenario design against the gate order, followed by checking that code output matches each expected label. Sample size is twelve fictional scenarios. Exclusions are all live records, campaign send data, CRM outcomes, model outputs, and legal advice. Limits: the examples cover only a small set of cases, use simplified boolean fields, and do not estimate performance, error rates, fairness, or legal compliance. A passing test confirms only code/example consistency.

Validation checklist before routing

• Is the source and collection date known, with the notice or lineage pointer available?
• Is the proposed country, channel, purpose, and message type identified?
• Has the required policy or legal review been completed for that exact context?
• Is the person’s role relevant to the offer, supported by a dated source?
• Is the account within a written segment definition rather than an intuitive label?
• Is the need or next step an observation, a clearly labelled inference, or still unknown?
• Has suppression been checked across relevant systems before any score or campaign rule?
• Does every accept have a named owner and a useful first action?
• Does every recycle have an owner, review date, and trigger?
• Does every hold have a resolver and a reason that blocks outreach?
• Can a seller return a record without using “bad lead” as the only explanation?
• Are metrics defined with denominators, cohorts, timestamps, and exclusions?
• Is any model version, threshold, and human override recorded?
• Are retention, access, and deletion rules stated for the evidence card and event history?

If any required control answer is unknown, pause the outbound action and resolve it. Do not increase the lead score to compensate for a missing source, a stale role, or an unprocessed objection. A productive system is one in which the blocked state is visible, owned, and recoverable; an invisible exception is simply a future incident.

Closing decision

Use the evidence card as a contract between teams: marketing records what was observed and how the person entered the process; sales records whether the context supports a useful conversation; data and privacy owners resolve source, notice, channel, and suppression questions. The handoff is complete when a person can understand the reason for the next action without reverse-engineering a score from raw events.

Begin with the synthetic cases, replace them with carefully governed local fixtures, and keep the permission gate ahead of commercial ranking. Choose accept, recycle, reject, or hold from evidence and explicit rules. Never describe a scenario test as a measured customer result, and never treat a model score as a lawful basis. The practical promise is narrower and more useful: every routed record has an explainable next step, an owner, and a trail of what remains uncertain.