What it solves. A purchased list describes a market everyone has already contacted. A built base describes yours, with the criteria you actually decide on.

What I do

• Define the segment through verifiable criteria: industry code, size, job title, network membership.
• Pick the source that actually carries that segment, not the most convenient one.
• Extract, normalise, deduplicate across campaigns and across tools.
• Enrich only what informs a decision, never every available field.
• Qualify through rules written in a spec file, not from memory.
• Explicitly exclude accounts that must not be contacted, current clients included.
• Export to the import schema of the tool taking over, tagged segment-priority-source.

How

• Source cost follows campaign stage: paid credits are reserved for post-conversion work, the top of funnel runs on public sources.
• Deduplication is cross-tool: never write to an address already contacted elsewhere.
• A campaign is read back from its spec file, which makes it reproducible by someone else.

With what

• Python — the processing pipeline and its tests
• PostgreSQL — the prospect base and sequence history
• INSEE / recherche-entreprises — official French company data
• Pappers — legal enrichment, deliberately rationed
• LinkedHelper — profile and company extraction
• Tavily — targeted web research and extraction

Proof

• 18,175 prospects, 1,101 qualified measured 1,101 consolidated contacts across three segments: 330 M&A boutiques, 545 HR and strategy firms, 226 business buyers, exported to the LinkedIn tool import schema.
• documented case Two exclusions capture the method: the current client is removed from targeting, and paid data credits are reserved for post-conversion work.

What it solves. The data that drives targeting almost always exists publicly somewhere. The work is not finding it once, it is collecting it monthly without breaking it.

What I do

• Identify the source that is authoritative for the segment, not the one easiest to read.
• Extract in the shape of the source: an API where one exists, a browser rendering where the page is built in JavaScript.
• Normalise identifiers on entry, because an identifier existing in two forms becomes a silent trap.
• Put a known witness in every extraction, to tell “nothing to find” from “extraction broken”.
• Keep the source and date attached to every collected field.

How

• What the server sends is not what the visitor sees: as soon as a claim is about a clickable link, measurement goes through a browser.
• An extractor returning an empty list on failure manufactures credible zeros: it must return an error.

With what

• Outscraper — geographic data, listings and reviews
• Tavily — search and page extraction
• Playwright — real rendering of JavaScript-built pages
• LinkedHelper — structured profile export
• Python — normalisation, deduplication, witnesses

Proof

• 18,175 prospects, 1,101 qualified measured 1,101 consolidated contacts across three segments: 330 M&A boutiques, 545 HR and strategy firms, 226 business buyers, exported to the LinkedIn tool import schema.

What it solves. The first message does not sell the product. It sells the time needed to understand the other side’s problem.

What I do

• Segment before writing: one message per segment, never one message for everyone.
• Set up sending infrastructure: dedicated domains, SPF, DKIM, DMARC, progressive warm-up.
• Keep transactional mail separate from outreach so the main domain’s reputation is not exposed.
• Write a short sequence where each follow-up adds something new.
• Personalise at the level the data can actually sustain.
• Treat positive and negative replies alike: both qualify.

How

• I sell the conversation before the solution: the goal of the message is the meeting, not the sale.
• A single approved sending runner, so no parallel script ever double-sends.
• The sending cap stays low until segment, data and promise all hold.

With what

• Mailjet — campaigns and customer activation
• HubSpot — sequences tied to the CRM
• Mailchimp — newsletter and nurturing
• Resend — transactional mail and alerts, kept apart from outreach

Proof

• documented case A single approved sending runner, and deduplication that forbids writing to an address already contacted by another tool.
• documented case The message does not sell the solution, it sells the conversation: the page justifies the meeting rather than closing.

What it solves. In a market where nobody knows the brand, audience and conversation are built in the same place.

What I do

• Target through standard and sales search depending on the precision needed.
• Grow the company page through sequenced invitations.
• Run a contact sequence after acceptance, not at invitation time.
• Export profiles to enrich elsewhere, and reconstruct address formats where data allows.
• Map the org chart by hand on large accounts, and build the relationship before the ask.

How

• Automation stops where account value justifies human time.
• Company page and personal profile are grown together: one gives credibility, the other opens the conversation.

With what

• LinkedHelper — invitations, sequences, extraction
• Sales Navigator — fine targeting when needed
• Make — moving data across to other tools
• Tavily — additional research on an account

Proof

• 0 → 2,000+ followers observed Company page from 0 to 2,000+, personal profile from 500 to 5,000, newsletter holding above 25% open rate.

What it solves. Without a shared system, marketing and sales tell two different stories about the same month.

What I do

• Architect contacts, companies and deals starting from the questions the team must answer weekly.
• Set lifecycle stages and the criteria for moving from one to the next.
• Define the score that turns a marketing contact into a sales one.
• Build the pipeline, nurturing and dashboards that genuinely serve the weekly review.
• Base forecasting on stages that have a definition, not on end-of-month intuition.
• Connect the product’s free tier to the CRM so usage becomes a commercial signal.

How

• Start from the smallest model that answers this week’s questions, and add a field only when a decision depends on it.
• A system nobody feeds produces a false reading, which is worse than no reading.

With what

• HubSpot — the CRM, built from zero and administered
• Mailchimp — the newsletter by segment
• Mailjet — customer activation campaigns
• Make — bridges between tools

Proof

• documented case A system shared by marketing and sales, plus a free tier of roughly 1,000 users feeding conversions and meetings.

What it solves. A proposal written before discovery answers an assumed problem. It then gets negotiated on price, for lack of having been negotiated on need.

What I do

• Look for the problem the organisation is living with, not the one the product knows how to solve.
• Identify champion, decision-maker, user and procurement — four roles rarely held by one person.
• Qualify budget, urgency and relative priority, the last of which kills the most deals.
• Disqualify early when the need does not match, and say so.
• Capture the customer’s vocabulary, objections and verbatims to feed them back into the messaging.

How

• The Échiquier method, a French qualification approach whose logic is compatible with MEDDPICC, applied to the sector’s core problem rather than recited as a grid.
• The demo doubles as discovery material: what the prospect comments on says more than what they answer.

With what

• HubSpot — the record of stages and stakeholders
• Démonstration produit — the discovery material

Proof

• 7 to 10 conversations a week observed This cadence was held with no media budget at all. Verbatims fed back into positioning, value proposition and sales messaging.
• 6 to 10 months, six or more stakeholders observed Exclusivity won with the federation, a pilot run with the public-sector body, deployment secured with the large group.

What it solves. The paid pilot is the way in: it is sold and signed first. What is at stake afterwards is making it succeed — and that is what decides the annual contract, which no buyer signs on a promise.

What I do

• Design scope and sessions against this client’s context, never a standard pack.
• Write the success criteria and the decision date before kickoff.
• Run a kickoff with every stakeholder, sponsor and managers included.
• Train the teams, then activate end users through dedicated campaigns.
• Track real usage during the pilot and correct mid-course.
• Run mid-pilot and end-pilot reviews, and turn them into the case for the annual contract that follows.

How

• The pilot is paid: what is free never gets decided.
• The managers’ role is a pilot condition, on par with scope.

With what

• HubSpot — deal tracking and follow-ups
• Mailjet — activation campaigns towards end users

Proof

• 2 to 6 months, 30 to 300 users observed Usage was steered during the pilot, not after: weekly reporting to the sponsor, mid-pilot and end-pilot reviews.
• observed The pilot did not convert. Sponsor and managers are part of the adoption system, not its backdrop.
• 6 to 10 months, six or more stakeholders observed Exclusivity won with the federation, a pilot run with the public-sector body, deployment secured with the large group.

What it solves. An account is rarely lost on the product. It is lost on usage that never started and that nobody looked at in time — and it is that usage, not the relationship, that decides the annual contract. Where the paid pilot is the way in, this function does not follow the sale: it produces it.

What I do

• Take over the account when the pilot is signed, keeping the sponsor relationship and delegating execution.
• Onboard the teams: train the administrators, then get the end users hands-on.
• Activate end users through dedicated email campaigns, designed account by account.
• Measure activation on usage events, not on impressions.
• Hold a weekly video report with the client: most active users, progression, usage peaks, most followed content — and what is not working.
• Re-engage inactive users through the channel that reaches them, and through the sponsor when needed.
• Run mid-pilot and end-of-pilot reviews with the sponsor and the managers.
• Measure satisfaction through a survey of the users themselves, not only of whoever signed.
• Carry adoption blockers back to the product team, and return to the account when the change ships.
• Turn a successful pilot into an annual contract, then into a wider scope.

How

• The paid pilot is the way in, not one land motion among several: the annual contracts are pilots that succeeded and were then renewed. Customer success is therefore not an after-sales function, it is the commercial engine.
• The pilot is sold and signed before it starts. The job is not to drive adoption in order to get a signature: it is to make what was signed succeed, because that success is what sells the annual contract.
• The weekly report also shows what is not working: a report that only says the positive is useless for deciding.
• An adoption problem is often a management problem, not a product or training one.
• Satisfaction is measured with end users, not only with whoever signed.
• The scope actually held: a handful of enterprise accounts owned end to end, not a book of several hundred accounts with an industrialised function, a tooled health score and a churn rate tracked down to its denominator.

With what

• Mailjet — per-client activation campaigns
• HubSpot — account and renewal tracking
• Pouls — satisfaction measurement, turned into a product

Proof

• 2 to 6 months, 30 to 300 users observed Usage was steered during the pilot, not after: weekly reporting to the sponsor, mid-pilot and end-pilot reviews.
• Roughly 1 in 2 accounts renewed observed Roughly one account in two renewed. Stated as an order of magnitude, not as a measurement.
• observed The pilot did not convert. Sponsor and managers are part of the adoption system, not its backdrop.
• documented case The anonymity rule goes as far as refusing to publish a whole breakdown: hiding one row is not enough, subtraction gives it back. www.solutio.work/pouls

1. Discovery — Understand the organisation, its real pain points and who decides, before proposing anything.
2. Bespoke pilot — A scope and a session design built against this client’s context, never a standard package.
3. Training & activation — Train the teams, then activate end users until usage holds without me.
4. Review & measurement — Weekly usage reporting to the sponsor, a mid-pilot review, satisfaction measured with the users themselves.
5. Renewal — Build the renewal case from evidence gathered during the pilot, not promises made at the end.
6. Expansion — Extend proven value to new populations, use cases or contract structures.

What it solves. A site can gain impressions for months without winning a single customer. The work is as much about spotting what will never convert as about producing pages.

What I do

• Read positions by intent and query pattern, never aggregate volume.
• Build the architecture and templates that carry real demand, at scale.
• Maintain technical health: redirects verified in production, structured data validated, internal linking with no orphan page.
• Make each intent owned by a single page before writing new ones.
• Instrument measurement and verify the instrument before drawing conclusions from it.
• Explicitly rule out what cannot be won, and say why.

How

• The buying vocabulary is not the problem vocabulary: the converting query is rarely the industry one.
• A supposedly empty niche is checked in the target language and in the real results page.
• Zero clicks on page one is not always a title problem: look first at the query’s language and shape.

With what

• Search Console — the source of truth on positions, owned through the API
• GA4 / Tag Manager — audience measurement and its verification
• Ahrefs — authority and assistant citations
• Astro — the templates producing pages at scale
• Playwright — performance measurement in a real browser

Proof

• 720+ products, 2,154 pages measured 2,154 pages built, 1,540 URLs declared in the sitemap, zero orphan pages. trouvemonsaas.fr
• median position 9 versus 53 measured Long conversational questions land at median position 9, generic queries at 53. The conclusion holds without its beachhead: 53% in the top 10 versus 4%.
• guide at position 1.7 measured The guide holds position 1.7, the category page fades. Two categories out of five were deliberately left out for lack of a measurable gap.
• 6 sites, 8 collectors, 84 tests measured The verification step caught five false figures, all shaped like correct measurements, including two sites counted under a single name.
• First ten clients in one month observed Ten first clients acquired in one month through local search alone, with no advertising.

What it solves. A single operator only sustains several products if repetitive work no longer depends on being there.

What I do

• Describe the process by hand before automating it, and fix it while it is still readable.
• Version the jobs and their tests, so switching tools is not a rebuild.
• Put a witness on each job, able to tell a failure from an absence of results.
• Keep a human stop where the decision is irreversible or visible to a customer.
• Watch the date of the last item produced, not the number of jobs installed.

How

• I adapt the stack to the problem, not the problem to the stack: value sits in the procedures, not the vendor.
• What stays manual is a decision, not a backlog: editorial validation, sending thresholds, identity-verified submissions.

With what

• Claude Code — orchestration of jobs and agents
• n8n / Make — bridges between off-the-shelf tools
• PostgreSQL / Supabase — shared state across automations
• MCP — tooled access to services, including a purpose-built server
• Telegram — the day-to-day control surface

Proof

• documented case Plus a deliberate line on what stays manual: editorial validation, sending thresholds, identity-verified submissions.
• 6 sites, 8 collectors, 84 tests measured The verification step caught five false figures, all shaped like correct measurements, including two sites counted under a single name.
• 18,175 prospects, 1,101 qualified measured 1,101 consolidated contacts across three segments: 330 M&A boutiques, 545 HR and strategy firms, 226 business buyers, exported to the LinkedIn tool import schema.
• seven business applications, one client observed Seven business applications, delivered as a v1 to a single consulting firm, then fully reworked into a v2.
• nineteen pages, no active mutating button measured Nineteen pages walked in production, no active mutating button. demo.solutio.work