hey squad
shall we settle in?

01 · AUT · 03/03

Equipped AI.

Not a showcase chatbot. Not a generic GPT. AI plugged into your data, supervised.

You set up AI tools for your team: lead qualification, conversation summaries, ticket classification. With **MCP** connectors to your CRM, helpdesk and product base. Mandatory human validation before any external send.

By Jérôme · Automation & data

Get my AI tool quotedSee the Ecostal mission

02 · OPERATION NOTE · OPS EQUIPPED AI

Equipped AI, decoded

The field report

Your team spends an hour a day qualifying incoming contacts, summarising calls, or classifying support tickets. A well-briefed AI does that work in a few seconds. That is often the moment, especially when request volume climbs.

The cost of waiting

Without equipped AI, you pay humans for tasks the machine does better and faster: triaging 200 incoming emails, summarising 30 Intercomthe customer support chat tool conversations, scoringautomatically rating contact quality 50 contacts per day. The opportunity cost climbs with your volume. And humans wear out on repetitive work instead of focusing on judgement.

In black and white

Qualified time given back to your team for tasks that need judgement. Consistency in contact and ticket processing. An AI layer that leans on your real data via a custom MCP Claude bridge (connecting Claude to your private data), not on a generic model trained on the internet.

The manoeuvre

We start by looking at what eats your team time, before writing a single AI instruction (prompt). The target use case is defined in writing. We code MCP custom Claudethe bridge connecting Claude to your private data connectors to GA4the Google audience measurement tool, HubSpotthe marketing and sales CRM and your product base. The Claude APIthe Anthropic AI model on inferencewhen the AI replies. Mandatory human review before any external send to the client. Notion doc handed over to your team.

The lock that breaks

A layer of intelligence on your catalogue and customer interactions. To combine with a server-side tracking layer so the AI works on clean data.

03 · WHAT WE WON'T WRITE IN AN RFP

Equipped AI is a supervised productivity tool. Not a replacement for your teams.

We code for your team, not against it. The rule we hold: the AI proposes, the human validates before any external send. A sales email never leaves to a client without a proofread. A contact score feeds the decision, it does not replace it. The value is in the data fed through custom MCP (the bridge connecting Claude to your private data), not in model sophistication.

  • 01

    Qualified time given back

    High-volume repetitive tasks (qualifying incoming contacts, summarising calls, classifying tickets) leave the human calendar. Your team focuses on judgement, nuance and client relationship.

  • 02

    Processing consistency

    All contacts scored on the same grid, all tickets classified on the same typology. No more Monday vs Friday variation, no more tired human judgement at end of day.

  • 03

    AI plugged into your data

    Through MCP connectors, the AI works on your catalogue, your CRM, your conversation history. Not on a generic model trained on the internet. Reply relevance follows data quality.

  • 04

    Systematic human supervision

    No external send without human validation. All the AI's outputs go through a review interface before leaving to the client. Hallucinations stay team-side, not in the inbox.

04 · THE MISSION THAT PROVED IT

We deployed it for these clients.

Ecostal Group team handling a solar panel in their warehouse

Ecostal Group

B2B photovoltaic e-commerce

6-month MCP custom Claude mission.

Read the Ecostal mission
Brasero estate, corporate retreats and weddings in Liège province

Brasero Group

Hospitality & events

AI qualification of corporate vs wedding inquiries.

Read the Brasero mission
Amecam workshop, fitting work vehicles in Wallonia

Amecam Van Conversion

B2B work-vehicle conversions

3-year form-lead pre-qualification.

Read the Amecam mission

05 · THE PLAY-BY-PLAY

Four steps. Six weeks on average. No tool handed over without training.

  1. STEP 01

    We look at what eats your team time.

    Audit of repetitive tasks with heavy cognitive load: qualifying incoming contacts, summarising calls, triaging tickets, personalising emails. We note frequency, human time and error risk. Tools: direct observation, Notion for mapping, 2-week sampling.

  2. STEP 02

    We pick the use case that really pays back.

    Target list of 1 to 3 use cases by ROI order. Scope frozen in writing, client sign-off. If a task requires too much contextual judgement, we say no, we do not code it. No silent additions outside the quote, no general AI promise.

  3. STEP 03

    We wire the AI through MCP.

    Development of MCP custom Claude connectors (the bridge connecting Claude to your private data) to your source tools: HubSpot (opens in a new window), GA4 (opens in a new window), BigQuery (opens in a new window) or your product base. Claude API (opens in a new window) on inference with versioned instructions (prompts). Mandatory human validation interface before any external send. Logging of every inference.

  4. STEP 04

    We train and we document.

    Gain measurement: human time saved, AI suggestion acceptance rate, residual error rate. 4-hour training for your team on usage and proofreading. Detailed Notion doc per use case. Prompt versioning for internal takeover.

06 · THE FLOW AT A GLANCE

The path each request takes between your source tools and Claude inference.

CLAUDE APISONNET / OPUSversioned promptslogged inferenceMCP CUSTOM CLAUDEtunnel to private dataGA4eventsBIGQUERYdata lakeCRM HUBSPOTleads · dealsHELPDESKtickets · product basePRIVATE DATAHUMAN VALIDATIONLEAD SCORINGunified scoring grid↳ after human reviewCONVERSATION SUMMARIEScalls · chats · emails↳ after human reviewTICKET CLASSIFICATIONconsistent typology↳ after human reviewPERSONALISATIONemails per segment↳ after human review« cherry on top, after the rest holds »— Jérôme

The custom MCP tunnels your private data into Claude. Inference comes out, but never leaves the agency without a human review. No external send without a gate.

07 · NOT YET FOR YOU IF

Three cases where equipped AI is not the top priority.

  • Your data is scattered and inconsistent.

    An AI plugged through MCP onto chaos gives chaotic replies. If your CRM contains duplicates and your tickets have no typology, we lay down cleanliness before the AI. Otherwise the model hallucinates on your data and team trust collapses fast.

  • Your monthly request volume is under 100.

    Below this threshold, the entry cost of an equipped AI tool does not pay back. At these volumes, humans remain faster to deploy and more precise on context. The cost-benefit ratio becomes debatable. We would rather say it than walk you into the dark.

  • You want an AI that decides for you.

    That is not what we code. The rule we hold: systematic human review before any external send. If you are looking for an AI that replies to clients on its own without validation, we are not the right provider. Supervision matters as much as the model.

08 · THE QUESTIONS WE ACTUALLY HEAR

Questions whispered after the second meeting. Honest answers.

On the Belgian SME market in 2026, ranges: 1 to 3 human hours per day on heavy-cognitive-load tasks (qualifying contacts, summarising conversations, classifying tickets). The delta depends on your request volume and expected precision level. Over 12 months, that easily equals a half-time role. The cost-benefit ratio gets assessed after scoping. The broader agency vs freelance vs in-house debate sits here.

Not the same. The Claude API (Anthropic's model) via custom MCP (the bridge connecting Claude to your private data) lets you wire precise connectors to your CRM, product base or helpdesk, with fine control over the data sent to the model. ChatGPT Enterprise remains a chat interface without that structured connector layer. For team usage on private data, MCP wins on precision and auditability (being able to trace who saw what). For free-form writing, ChatGPT stays competitive. The AI layer then leans on cross-tool workflows to automate what the AI qualified.

Not to our knowledge as of late 2026 on the AI channel itself. Documented sanctions hit unframed data transfers outside the EU (Anthropic and OpenAI having valid standard contractual clauses since 2024) and the lack of legal basis per purpose. A clean mechanic with team opt-in, an up-to-date processing register and a log of each inference remains compliant.

Belgian SME 2026 market range on a 1 to 2 use case setup: 1 to 4 human hours per day recovered over time. The delta depends on request volume, expected precision and the human proofread time kept. For your specific case, we quote after a scoping call. The proof on a custom MCP wired to Mailchimp + catalogue: Ecostal mission.

Claude API side: 100 to 500 € HTVA per month depending on inference volume (the number of times the AI replies). Maintenance side: 1 dev day per quarter to follow Claude evolutions (new models like Sonnet in fast version or Opus in powerful version, API changes) and refine instructions. If we leave, you keep the Notion doc, the MCP admin access, the versioned prompts and the internal agency relay procedure. For content-side copywriting, we frame the same method in briefed AI copywriting.

Yes. We do this on half our AI missions. The rule: who steers use case definition and post-go-live validation. If your in-house dev holds the infra, we frame the prompts and the validation interface. If we hold the mechanic, we document for the next owner. We never step on the other team without saying. To plug the AI onto already-tuned sequences, see also Mailchimp / Brevo newsletter scenarios.

Field note

Equipped AI is supervised productivity. Not a human replacement. Across the missions we have run, what works is 1 or 2 precise use cases, plugged into clean data via MCP, with mandatory human review. It is the cherry on top of a data setup that already holds. We rarely start with AI.

Jérôme · Strategy & data
JérômeStrategy & data · HeySquad

Other sub-services MARKETING AUTOMATION

We look at what eats your team time. We tell you what we equip, and what stays human.

Get my AI tool quotedBack to the Marketing automation service