Bespoke automated B2B prospecting

The B2B prospecting market is crowded: Apollo, Cognism, Lusha, Pharow, Lemlist, La Growth Machine and Waalaxy. Each excels within its own niche, but none covers the complete chain. Five structural limitations explain why no mass-market provider can do so without abandoning its economic model.

• First, detecting newly formed French companies.
  

  No market tool natively captures and analyses France’s company-registration stream every day. Yet these are precisely the companies that need everything: an accountant, a website, an electronic-signature solution, professional insurance, legal counsel, a communications agency, a CRM and a business bank account. The commercial window is open and brief, but no one monitors it systematically.

• Second, the depth of enrichment.
  

  Apollo-type platforms stop at the surface: name, sector, contact and declared company size, at best. The target remains a row in a spreadsheet, never a fully researched file. Where Apollo supplies a business card, we wanted to provide a corporate biography: accounts, actual headcount, the founder’s career history, past legal proceedings, related companies and public historical signals, all examined in depth before qualification.

• Third, message personalisation. The tone is poor.
  

  The context is superficial. The enrichment gathered upstream is almost never used downstream. Emails written by mass-market AI tools are not genuinely personalised; they imitate the form without the substance. There are two understandable reasons for this. Rebuilding an application every time a new language model is released is an undertaking no mass-market SaaS vendor can repeat indefinitely.

  Running a frontier model for every message would destroy its margin. Powerful local alternatives such as Mistral, Gemma and Llama will address the first constraint. The second—the cost—will remain. Deep enrichment will never become cheap enough to fit a mass-market subscription.

• Fourth, the perceived quality of the sending company deteriorates.
  

  A recipient who receives a standardised cold email does not merely read a message; they draw conclusions about the company that sent it. Poor writing, a mechanical tone and approximate relevance are three signals that can lower the sender’s standing by one or two levels in less than five seconds. Unlike the email, that diminished perception does not disappear. It remains.

  The recipient now associates the brand with a clumsy cold email, an unwanted message and a company that does not know how to address them properly. The halo effect works in reverse: the brand loses a degree of authority in the recipient’s mind and may need to earn back twice as much. Reputational capital has been spent on a message that should never have been sent.

  Poorly calibrated prospecting is not neutral; it is destructive.

  This damage is not incidental. In a market where executives receive dozens of B2B approaches every day, the difference between a message that elevates the sender and one that diminishes it becomes an unspoken filtering criterion. The first opens a door—perhaps later, perhaps never, but the door remains open. The second closes that door and others the sender has not yet knocked on.

  By encouraging volume at the expense of care, mass-market tools make their users lose precisely what they claim to help them gain.

  Our conviction is the exact opposite. The goal is not a campaign to one million email addresses, but a single, justified and defensible approach. We contact a particular prospect because our analysis has established that they face a concrete need we can meet, not because their name appears in a list generated by an approximate filter.

  We approach them properly, without being aggressive, applying pressure or inflicting the fifth follow-up in an automatically generated sequence. Prospecting becomes what it should always have remained: a surgical act, not an avalanche.

• Fifth, emails end up in spam.
  

  The issue here is no longer the recipient’s perception, but that of the mailbox itself. Modern filters such as Gmail, Outlook and Proofpoint no longer analyse only keywords; they recognise patterns. A sequence sent to ten thousand recipients, with the same structure, rotating hooks and generic dynamic images, can be identified as mass cold email within hours.

  The result is unforgiving: messages land almost automatically in spam, where they are neither read nor answered. The recipient never even sees them because the filter has already intervened. The consequence is twofold: delivery costs are incurred with no return, while the sending domain’s reputation deteriorates with every wave, making subsequent campaigns even less deliverable. It is a downward spiral.

  The more a vendor pushes volume, the more that volume becomes noise no one hears because no one receives it.

Four stages, one integrated workflow.

• Detection.
  

  Logiks Detect orchestrates five French government databases to capture newly registered legal entities every day, filtered by legal form, activity, geography and disclosure status. It is not a stock of records to refresh, but a stream to process as soon as a company enters the register.

• Qualification.
  

  Two distinct scores are used and never conflated. An opportunity score measures the target’s intrinsic commercial quality: legal form, capital ranked by percentile within its sector cohort, activity code, corporate purpose and recency. A contactability score measures something different: whether the company can and should be contacted, and through which quality of channel.

  A third structural safeguard applies France’s commercial-solicitation status, blocking outreach by default for companies that have asked not to be approached.

• AI fact-checking.
  

  This is where the tool stops being a prospecting tool and becomes a qualification engine. A chain of queries scans the prospect’s public LinkedIn profiles using their registry identity rather than the company name. A deterministic algorithm scores each candidate against ten criteria. Only the shortlist is submitted to the model, which must produce a structured profile across six validation dimensions.

  One hard, non-negotiable rule applies: if the model proposes a profile that is not on the shortlist, it is rejected. The AI does not choose; it verifies.

• Generation.
  

  A confidence score consolidates the evidence and determines one of five outreach modes, from complete blocking to authorised personalisation. The writing agent receives only the authorised context. It may not invent a fact, cite an unverified biography or use a prohibited signal. Drafting runs on frontier models that mass-market tools cannot afford to use.

  Saint-Exupéry expressed the principle in relation to flight: perfection is achieved not when there is nothing left to add, but when there is nothing left to remove. We filter rigorously upstream so that computing power is used only for targets that merit it.

The tool is particularly well suited to B2B companies that operate across several sectors and offer a broad range of services. This is precisely where personalisation becomes most demanding: varying the tone is not enough; the right offer must be matched to the right prospect without hesitation or approximation.

The pipeline receives the full set of use cases, services and segment-specific arguments as context. Guided by that catalogue, the AI selects the most relevant ones for each target.

Consider a concrete example. A property agency is newly established in Lyon. Its NAF code identifies the category unambiguously. The prospect’s career history, verified through fact-checking, shows previous experience in high-end property transactions.

The pipeline then assembles a calibrated approach: for this target, it highlights our services for property websites, local search visibility and branding for independent agencies—not those intended for a law firm or software vendor.

The message does not say, “we offer many different services”; it says, “this is what you will need in the coming weeks, and this is why we know how to deliver it.” The target receives an individually written proposal: an automated chain produces a bespoke draft, governed with the discipline of a workshop.

The level of automation is not set globally through a monolithic switch. It is configured step by step, with one click at every stage of the pipeline. Detection, qualification, enrichment, fact-checking, drafting and sending can each switch independently between three modes. Automatic mode runs the stage without intervention.

Rules-based mode runs it only when the specified conditions are met—for example, automatic qualification only for SASUs with more than ten thousand euros in capital and within selected NAF codes. Manual mode keeps a human in control of that specific stage.

The configuration we use for our own prospecting illustrates the philosophy. Detection, qualification, enrichment and fact-checking run automatically because they are sorting and investigation tasks at which machines excel. Drafting is also automatic because the upstream context is sufficiently controlled for a frontier model to produce a strong first version. Sending, however, remains manual. The AI creates a draft directly in our Gmail account.

We review it, adjust the wording where necessary and send it ourselves. This is a deliberate decision. We want to operate precisely, never spam and never allow a message to leave that we would not personally sign. The machine performs the research and preparation. The human retains the final say.

Our objective is neither to spam nor to pretend that we understand prospects we do not know. It is more modest and more demanding at the same time: to understand a company’s genuine need before making contact and propose precise solutions that deliver concrete, measurable value. That is all. Everything else is the preliminary research required to determine whether the approach is justified.

This discipline produces three direct effects.

• First, better targeting.
  We contact a company only when the analysis establishes that it faces a need our services genuinely address. All others remain in the database, scored but not approached. Contact volume decreases while the relevance of each contact increases.

• Second, preserved reputational capital.
  Where a poorly calibrated mass campaign creates a lasting association between the brand and a clumsy cold email, a surgical approach does the opposite: no negative trace, no automatic spam classification and no destructive halo effect. Mass-market tools produce volume that lands in spam; we produce messages that reach the inbox.

• Finally, documentable commercial value.
  For an equivalent delivery cost, response rates, conversion rates and the cost per qualified appointment all move in the right direction. This is not a question of volume, but of basic commercial arithmetic.

We are working to measure this difference precisely across controlled cohorts. If your company actively prospects and would like to take part in a comparative study—your current stack measured against Detect calibrated for your vertical, comparing response rate, conversion rate and cost per qualified appointment—we would be pleased to discuss it.