DEV Community: Alex Boissonneault

Your AI assistant can't read your pipeline — here's why that's a problem

Alex Boissonneault — Wed, 13 May 2026 15:18:25 +0000

You use AI every day for writing, summarising, and brainstorming. But ask it what's really happening in your pipeline right now — and it stares back at you blankly. That's not a prompt problem. It's a structural one.

The honest reality of AI and business data today

When you open Claude, ChatGPT, or any large language model and ask a business question, the AI is working from one of three sources:

Training data that ended months or years ago
Whatever you pasted manually into the chat window
Documents you uploaded in that specific session

None of those are your live CRM. None of them know which deals are stalling, which customers are about to churn, or which marketing channel is actually converting. The AI is flying blind.

What this looks like in practice

You: "Why are our Q2 deals taking so long to close?"

AI: "There are several reasons deals may take longer to close..."

You: [copy-pastes five pipeline screenshots]

AI: "Based on the screenshots you shared, it looks like..."

You: [repeat for every other business question]

This is not intelligence. This is pattern-matching on stale context. The AI doesn't know that Deal #47 has been sitting in "Proposal Sent" for 23 days.

Why this gap exists

Most AI tools were built to process text. Business data — CRM records, pipeline stages, customer segments — lives in structured databases, not documents.

To make AI genuinely useful for revenue operations, you need a bridge between the AI's reasoning engine and your structured business data. That bridge has a name. It's called MCP — Model Context Protocol.

What MCP changes (without the jargon)

Model Context Protocol is a standard developed by Anthropic that lets AI assistants call structured tools using plain language.

Instead of copy-pasting screenshots, you ask:

"Analyze my pipeline health"
"Who are my highest-risk accounts?"
"What's my dominant growth constraint right now?"

And the AI actually knows. Not because you told it. Because it has structured access to your data through purpose-built tools.

The bigger picture

Right now, most SMBs are using AI as a glorified autocomplete. Enterprise teams with large budgets are quietly building AI-native revenue systems. The gap is widening every quarter.

The tooling to close that gap is now open-source, free to install, and works in minutes. Next week I'll walk you through exactly how it works.

Before you go: How are you currently using AI in your sales or marketing workflow? Are you feeding it live business data, or still copy-pasting context manually?

CTA: Follow me on dev.to — next week: a plain-language breakdown of MCP and exactly how it bridges AI and your CRM.

RFM Segmentation for B2B SaaS: The 11-Segment Model That Changed Our Clients' Retention

Alex Boissonneault — Fri, 08 May 2026 16:27:19 +0000

Most B2B companies treat customer segmentation like a binary: you're either active or churned. That's like a doctor classifying patients as either "healthy" or "dead" — technically accurate, and completely useless for intervention.

RFM analysis — Recency, Frequency, Monetary — has been a retention staple in B2C e-commerce for decades. But the B2B SaaS application of it is different, and most implementations I've seen get it wrong. They use a 5-segment model (Champions, Loyal, At Risk, Hibernating, Lost) and treat it as a one-time exercise.

What we run at Artefact Ventures is an 11-segment, AI-native RFM model that produces different interventions for each segment and runs continuously against live CRM data. Here's how it works, why the extra segments matter, and what to actually do with the output.

What RFM Means in B2B Context

Before the mechanics: a quick note on how to interpret the three dimensions for B2B SaaS specifically, because they mean something different than in e-commerce.

Recency — When did this account last meaningfully engage? In B2C, this is "last purchase date." In B2B, it should be the most recent high-signal activity: a logged call, an expansion conversation, a support ticket that resolved successfully, a renewal. Not just login activity — that conflates usage with engagement.

Frequency — How often does this account transact, expand, or engage at a decision-making level? In B2B, a customer who buys once and stays for 3 years at the same contract value is very different from one who buys once and churns. Frequency should capture touchpoints that indicate active relationship investment.

Monetary — Total revenue contribution, including expansion. Critically, this should be weighted by margin contribution where possible — a large account on a heavily discounted legacy plan is worth less than a smaller account at full rate.

If your RFM model doesn't account for these B2B-specific interpretations, you'll misclassify accounts and send the wrong interventions.

The 11-Segment Model: Full Breakdown

Standard RFM uses a 1-5 score on each dimension, producing 125 possible combinations. We bucket these into 11 actionable segments. Here's the complete model:

Segment 1: Champions

Profile: High recency, high frequency, high monetary value. These accounts bought recently, buy often, and spend the most.

What they need: Recognition, early access, co-creation opportunities. Champions are your best reference customers and your best source of product intelligence. They should never receive a generic nurture email.

Intervention: Executive relationship check-in. Early access to new features. Referral program invitation. Case study or co-marketing proposal.

Warning signal: A Champion whose recency score drops suddenly is your highest-priority churn risk — because the fall from Champion to At Risk is faster and more expensive than any other transition.

Segment 2: Loyal Customers

Profile: High frequency, strong monetary value, moderate recency. They buy consistently but haven't engaged recently.

What they need: Re-engagement with new value, not a sales pitch. They already like you. Show them something they haven't seen yet.

Intervention: Product update briefing. Invite to a user community or event. Expansion conversation anchored in their specific use case.

Segment 3: Potential Loyalists

Profile: Recent customers with above-average frequency, not yet in Champion territory.

What they need: Velocity. They're on the right trajectory — the goal is to accelerate the pattern without over-engineering the relationship.

Intervention: Onboarding optimization check-in. Feature adoption nudges. Loyalty program introduction if you have one.

Segment 4: Recent Customers

Profile: Bought recently, low frequency, lower monetary value. New accounts in early lifecycle.

What they need: A successful first experience. Everything else is secondary. If they don't get value in the first 90 days, they will not become Potential Loyalists.

Intervention: Active onboarding support. Success milestone tracking. First-90-days check-in with a human, not an automated email.

Segment 5: Promising

Profile: Recent purchase, low frequency, low monetary value. Early-stage, low commitment.

What they need: A proof point. One clear win that makes the relationship feel worth continuing.

Intervention: Use-case-specific success story. Quick-win workflow or template. Low-friction expansion offer (not upsell — proof-of-value first).

Segment 6: Need Attention

Profile: Above-average scores across all three dimensions historically, but recency is declining. These accounts were strong and are starting to drift.

What they need: Proactive contact before they self-identify as disengaged. This is the intervention that most companies miss because the account still looks healthy in a revenue dashboard.

Intervention: Proactive QBR or success review. Direct outreach from account owner, not CSM automation. ROI recalculation to re-anchor value.

Segment 7: About to Sleep

Profile: Below-average recency and frequency, but not yet lost. They're fading.

What they need: A reason to stay that they haven't heard before. Generic renewal reminders will not work here.

Intervention: Personalized re-engagement campaign based on their specific product usage history. Limited-time expansion offer. Direct conversation about fit — sometimes it's better to right-size than to retain at the wrong tier.

Segment 8: At Risk

Profile: High historical monetary value but declining recency and frequency. High-value accounts showing churn signals.

What they need: Urgent, executive-level attention. Not a CSM — the account executive or a founder, depending on company size.

Intervention: Executive sponsor check-in within 5 business days. Competitive displacement assessment. If they're evaluating alternatives, you need to know now, not at renewal.

Segment 9: Cannot Lose Them

Profile: Made large purchases historically but recency is very low. These accounts were significant and have gone quiet.

What they need: A genuine reconnection, not a retention script. Something went wrong — find out what before making any offer.

Intervention: Honest conversation about the relationship. Service recovery if applicable. Re-scoping the engagement to match current needs.

Segment 10: Hibernating

Profile: Low recency, low frequency, low monetary. Minimal engagement across all dimensions but not technically churned.

What they need: A decision. Either re-engage with a compelling reason or let the relationship end cleanly. Maintaining hibernating accounts in your pipeline creates false coverage.

Intervention: Sunset campaign with a clear value proposition. If no response after two touchpoints, move to offboarding and clean the CRM.

Segment 11: Lost

Profile: Lowest scores across all three dimensions. Churned or effectively churned.

What they need: A clean exit, a post-mortem, and a future win-back path if appropriate.

Intervention: Exit survey (keep it short — 3 questions max). Flag for win-back sequence in 6-12 months if the churn reason was situational rather than product-fit. Feed insights into ICP refinement.

The Scoring Mechanics

Here's how we score accounts for placement:

Step 1 — Score each dimension 1–5:

5 = top 20% of your customer base on that dimension
4 = 60th–80th percentile
3 = 40th–60th percentile
2 = 20th–40th percentile
1 = bottom 20%

Step 2 — Combine into an RFM string: An account scoring R=4, F=3, M=5 is a "435."

Step 3 — Map to segment: Use a lookup table to assign each RFM combination to one of the 11 segments. Champions are typically 554, 544, 545, 455, 454. Lost accounts are 111, 112, 121.

Step 4 — Run interventions per segment: This is the step most companies skip. The scoring is not the output — the intervention is.

Running This with AI: How the Artefact MCP Server Does It

Manual RFM scoring is a spreadsheet exercise that gets done once a quarter and immediately goes stale. The Artefact MCP RFM Analysis Engine runs this continuously against live HubSpot data.

A natural language query like:

"Run RFM analysis on my customer base and show me which segments have the highest churn risk this month"

Returns a structured output with:

Current segment distribution across all 11 buckets
Accounts that have moved segments since the last analysis (the transitions are the most important signal)
Pre-built intervention recommendations per segment
Priority ranking by revenue at risk

Install:

pip install artefact-mcp
claude mcp add artefact-mcp

Pro tier adds live HubSpot integration with custom RFM thresholds — meaning you can adjust the recency, frequency, and monetary weightings to match your specific business model rather than using generic defaults.

The Transition Matrix: What Really Matters

The segment score at a single point in time is less important than the direction of movement. Here's the transition matrix you should monitor:

From	To	Priority
Champion → At Risk	Any drop	🔴 Immediate
Need Attention → About to Sleep	Declining recency	🔴 Immediate
Loyal → Need Attention	Frequency drop	🟠 High
Potential Loyalist → Need Attention	Stalled frequency	🟠 High
Recent → Promising	Frequency increase	🟢 Positive
Promising → Potential Loyalist	Sustained engagement	🟢 Positive

A company that knows its segment distribution but doesn't track transitions is reading a photograph instead of watching a film.

Common RFM Mistakes in B2B

Mistake 1: Using login data as a proxy for recency. Login frequency measures access, not engagement. An account with daily logins and zero expansion conversations for 6 months is not a Champion — they're a habitual user who may be actively evaluating alternatives.

Mistake 2: Treating all segments with the same communication cadence. Lost accounts do not need weekly newsletters. Champions do not need basic feature education emails. Segment-specific cadence is not optional.

Mistake 3: Running RFM on a too-short time window. B2B purchase cycles are long. A 30-day RFM window will misclassify most of your customer base. For most B2B SaaS companies, use a 12-month look-back for Frequency and Monetary, and a 90-day look-back for Recency.

Mistake 4: Ignoring segment transitions at renewal time. If a customer was a Champion 12 months ago and is now in the "Need Attention" segment, renewing them at the same price with the same pitch will fail. The conversation needs to be different.

FAQ

What is RFM segmentation?
RFM segmentation is a customer analysis framework that scores customers on three dimensions — Recency (how recently they engaged or purchased), Frequency (how often they engage or purchase), and Monetary value (how much revenue they generate). The scores are combined to classify customers into segments that each require different retention strategies.

How many RFM segments should a B2B company use?
Most implementations use 5 segments. A more effective B2B model uses 11 segments — Champions, Loyal Customers, Potential Loyalists, Recent Customers, Promising, Need Attention, About to Sleep, At Risk, Cannot Lose Them, Hibernating, and Lost — because each requires a materially different intervention and the 5-segment model collapses important distinctions.

How often should I run RFM analysis?
Monthly at minimum for active intervention. Weekly if you have an automated system. The transitions between segments — not the current snapshot — are the most actionable signal, and those transitions happen continuously.

Can RFM analysis be automated with AI?
Yes. The Artefact MCP Server's RFM Analysis Engine runs against live HubSpot data and returns segment distributions, transition alerts, and pre-built intervention recommendations through a natural language interface. Install with pip install artefact-mcp.

What's the difference between B2C and B2B RFM?
B2C RFM uses purchase date, purchase count, and purchase value — all cleanly transactional. B2B RFM requires a more nuanced interpretation: recency should capture high-signal engagement, not just last purchase; frequency should reflect decision-level interactions; monetary should account for expansion and margin, not just contract value.

Alex Boissonneault is the founder of Artefact Ventures, a Québec-based firm building AI-native GTM systems for SMBs. He has 15 years of experience across seven major enterprise revenue transformations.

I Built the First GTM-Native MCP Server — Here's What I Learned

Alex Boissonneault — Wed, 29 Apr 2026 14:50:15 +0000

When Anthropic released the Model Context Protocol (MCP) spec, most developers immediately saw database connectors, file-system tools, and API wrappers. I saw something different: a direct bridge between an AI assistant and the revenue chaos that kills most SMBs.

I've spent 15 years fixing broken go-to-market operations — seven major enterprise transformations, including work that contributed to $800M in revenue at Québecor's retail network. Every single engagement had the same root problem: the data existed, but no one could turn it into a decision fast enough.

MCP changes that equation. So I built the Artefact MCP Server — the first MCP server purpose-built for GTM revenue intelligence — and published it as open source.

Here's what I learned building it, and why the architecture decisions matter if you're thinking about AI-native revenue operations.

Why GTM Data Is Uniquely Hard for AI
Before I explain what I built, it's worth explaining why this problem is harder than it looks.

Most MCP servers solve a data-access problem: "let the AI read my calendar" or "let the AI query my database." That's valuable, but it's table stakes. The harder problem in GTM is not access — it's interpretation.

When a sales manager asks "how is my pipeline doing?", they don't want a row count. They want to know:

Which deals are at risk of stalling before month-end
Which accounts are showing expansion signals
What constraint is currently limiting revenue growth — traffic, conversion, deal size, or velocity

This requires methodology embedded alongside data access. Raw HubSpot data + an LLM produces hallucinated optimism. HubSpot data + structured analytical frameworks + an LLM produces actual intelligence.

That's the gap I built Artefact MCP to fill.

The 7 Tools: What Each One Does and Why It Exists
The Artefact MCP Server exposes seven structured tools through Anthropic's Model Context Protocol. Each one is a conversation between your AI assistant and a specific revenue framework.

1. Pipeline Signal Detection
What it does: Classifies pipeline activity into 6 signal types with evidence-backed scoring.

The 6 signal types:

Momentum Shift— acceleration or deceleration in deal progress
Stall Pattern — deals not advancing past expected timelines
Conversion Anomaly — unexpected conversion rate changes at any stage
Engagement Spike — sudden increases in prospect or customer activity
Risk Indicator — early warning signs of deal loss or churn
Expansion Signal — indicators of upsell or cross-sell opportunity

Why it exists: Most pipeline reviews are vibes dressed up as data. A rep says a deal is "progressing well" because the prospect replied to an email. This tool turns behavioral signals into scored, categorized intelligence that doesn't depend on rep optimism.

Example query: "Detect signals in my current pipeline and flag anything that needs attention this week."

2. Constraint Analysis
What it does: Uses the Revenue Formula to identify the single dominant constraint limiting your growth.
The Revenue Formula breaks down like this:

Revenue = Traffic × Conversion Rate × Average Deal Size × Velocity

Every company has one bottleneck that, if fixed, would unlock disproportionate growth. This tool identifies whether that bottleneck is:

Traffic (not enough pipeline entering the top)
Conversion (deals not progressing through stages)
Deal size (closing too small)
Velocity (taking too long to close)

Why it exists: I've watched companies spend $200K on lead generation when their real problem was a broken demo-to-proposal conversion rate of 8%. Fix the constraint, not everything.

Example query: "Analyze my pipeline data and tell me what's the primary constraint on my revenue right now."

3. Value Engine Analysis
What it does: Assesses the health of three value engines — Growth, Fulfillment, and Innovation — and reveals which one is underpowered.

The three engines:

Growth Engine — how you acquire and expand customers
Fulfillment Engine — how you deliver value and retain customers
Innovation Engine — how you build new capabilities and stay ahead

Most SMBs over-invest in the Growth Engine and ignore the other two, then wonder why churn is high or why they keep losing deals to better-featured competitors.

Example query: "Evaluate my value engines and tell me where to invest next quarter."

4. GTM Commit Drafting
What it does: Drafts structured GTM change proposals — like version control for your go-to-market strategy.

Each commit includes:

Intent — what you're changing and why
Evidence — the data supporting the change
Impact surface — what parts of the business are affected
Rollback criteria — what would trigger reverting the change

Why it exists: Most GTM changes are made on instinct and reversed on panic. Treating strategy changes like code commits — with explicit evidence requirements and rollback plans — forces rigor and creates an institutional memory.

Example query: "Draft a GTM commit for changing our ICP from SMBs under 50 employees to 50-200 employees, based on our last 6 months of deal data."

5. ICP Triangulation
What it does: Builds an ideal customer profile using three dimensions, and generates a composite fit score per account.

The three dimensions:

Company attributes — firmographics: size, industry, location, revenue band
Behavioral signals — engagement patterns, purchase history, support interactions
Growth indicators — funding rounds, hiring trends, expansion signals

Most ICPs stop at firmographics. That's like hiring based only on job title — you'll miss the best candidates and recruit expensive disappointments.

Example query: "Triangulate my ICP using my last 90 days of closed-won deals and score my current open opportunities."

6. RFM Analysis Engine
What it does: Classifies customers across 11 segments based on Recency, Frequency, and Monetary value, with pre-built retention strategies per segment.

The 11 segments:

Why it exists: Every segment needs a different intervention. Sending a win-back campaign to your Champions is not just wasteful — it's actively damaging. This tool generates segment-specific strategies automatically.

Example query: "Run RFM analysis on my customer base and tell me which segments need attention this month."

7. Pipeline Health Scoring
What it does: Generates a 0–100 pipeline health score by analyzing velocity, stage distribution, conversion rates, and deal aging.

The score factors in:

Velocity through each stage
Distribution balance (too many deals stacked in one stage = a problem)
Conversion rates stage by stage
Deal aging relative to your average sales cycle
Exit criteria compliance

Example query: "Score my pipeline health and tell me what's dragging the number down."

How to Install and Use It
The install is deliberately simple — one command, no dependency management required:

bash
pip install artefact-mcp

Then configure it in your AI assistant's MCP settings:
bash claude mcp add artefact-mcp

It works with Claude Desktop, Claude Code, and Cursor. No API key required to get started — the server ships with sample data so you can explore every tool immediately.
For live HubSpot data, upgrade to Pro (the configuration takes about 2 minutes).

Full documentation and the source code are at: github.com/alexboissAV/artefact-mcp-server

The package is published on PyPI: pypi.org/project/artefact-mcp

What I Got Wrong the First Time
The first version of Artefact MCP was basically a HubSpot API wrapper with some prompts. It was useless. An AI assistant with read access to your CRM is not intelligence — it's just a slower way to look at data you already had.
The breakthrough was embedding the methodology inside the tool definitions. Instead of returning raw pipeline data and hoping the LLM would know what to do with it, each tool returns structured, scored, categorized output that tells the AI what it means — not just what it is.
That shift — from data access to data interpretation — is what makes the difference between a party trick and a revenue advisor.

What's Next
The current release is v0.3. On the roadmap:

Multi-CRM support (Salesforce, Pipedrive) for Enterprise tier
Custom signal configurations — define your own signal taxonomy
Integration with the Artefact CRO Platform for continuous monitoring

If you're building AI-native GTM tooling or working on MCP integrations for revenue operations, I'd genuinely enjoy the conversation. Reach out at alex@artefactventures.com.

FAQ
What is the Artefact MCP Server?
The Artefact MCP Server is an open-source Model Context Protocol server that transforms AI assistants like Claude into revenue intelligence advisors. It provides seven structured tools — Pipeline Signal Detection, Constraint Analysis, Value Engine Analysis, GTM Commit Drafting, ICP Triangulation, RFM Analysis, and Pipeline Health Scoring — accessible through natural language queries.

How do I install the Artefact MCP Server?
Run pip install artefact-mcp to install, then claude mcp add artefact-mcp to configure it with Claude Desktop or Claude Code. No API key is required for sample data mode.

What AI assistants does Artefact MCP work with?
Artefact MCP is compatible with Claude Desktop, Claude Code, and Cursor. It implements Anthropic's Model Context Protocol standard.

Is Artefact MCP free?
Yes. The free tier includes all 7 tools running on sample data. The Pro tier ($149/month) adds live HubSpot data, custom RFM thresholds, and custom ICP mapping. Enterprise ($499/month) adds multi-CRM support and dedicated onboarding.

What is the difference between Artefact MCP and the HubSpot MCP Server?
HubSpot's MCP server (which reached GA in April 2026) handles CRM operations — reading and writing to contacts, deals, companies, and other objects. Artefact MCP handles revenue intelligence — analysis, scoring, signal detection, and strategic frameworks. They are complementary, not competing.

Alex Boissonneault is the founder of Artefact Ventures, a Québec-based consulting and product firm specializing in AI-native GTM systems and revenue intelligence for SMBs.