Modern Pazarlama Operasyonları Yığını: Referans Mimarisi
The average B2B company uses 91 marketing tools. That's not a statistic to be proud of — it's a symptom of a stack that grew by accretion rather than design.
Every tool solves one problem. Every integration between tools introduces two more. The result: marketing ops teams spend 60% of their time moving data between systems and 40% actually using it. The stack was supposed to make operations faster. Instead, it became the operation.
This is a reference architecture for a marketing operations stack that works. Not a tool list. Not a comparison chart. An architecture — meaning the layers, the data flows, the integration patterns, and the decision criteria for what goes where.
It's opinionated. It's based on stacks we've built and rebuilt for B2B operators doing $2M-$50M ARR. Adjust for your scale, but the architecture holds.
Why Your Current Stack Is Probably Wrong
Three patterns produce dysfunctional stacks:
Tool-first decisions. Someone reads a blog post about a new marketing automation platform. The team signs up for a trial. It works well in isolation. It gets adopted. Six months later, the team discovers it doesn't integrate with the CRM, can't handle the lead scoring logic they need, and requires manual CSV exports to feed the analytics layer. But by then, 8 months of workflow history lives in the tool. Migration feels impossible. So they bolt on more integrations.
Departmental silos. Marketing buys HubSpot. Sales buys Salesforce. Customer success buys Intercom. Each team optimizes their own tooling without considering the data flow between them. The result: three sources of truth for the same customer, three conflicting activity timelines, and no single view of the customer journey.
Vendor lock-in by inertia. The stack works "well enough." Nobody wants to be the person who proposes a migration. So the team works around the tool's limitations instead of solving them. Workarounds accumulate. Tribal knowledge grows. The cost of staying slowly exceeds the cost of switching, but nobody measures the cost of staying.
The fix isn't better tools. It's better architecture.
The Reference Architecture
A functional B2B marketing operations stack has five layers. Data flows down from acquisition to insight. Decisions flow up from insight to action.
┌─────────────────────────────────────────┐
│ ACQUISITION LAYER │
│ Website · Ads · Content · Events · SDR │
├─────────────────────────────────────────┤
│ CRM / SOURCE OF TRUTH │
│ (HubSpot / Salesforce / etc.) │
├─────────────────────────────────────────┤
│ AUTOMATION LAYER │
│ Email · Sequences · Workflows · AI │
├─────────────────────────────────────────┤
│ ANALYTICS LAYER │
│ Attribution · BI · Dashboards · Alerts│
├─────────────────────────────────────────┤
│ DATA LAYER │
│ Warehouse · ETL · Reverse ETL · CDP │
└─────────────────────────────────────────┘
Layer 1: Acquisition
Everything that puts a lead into the system. Website forms, paid ads, organic content, events, SDR outreach, referrals. Each acquisition channel feeds into the CRM through a structured pipeline.
The critical principle: every lead enters through one door. Not directly into the email tool. Not into a spreadsheet that gets uploaded later. Into the CRM, with source attribution, immediately. If a lead's first touchpoint isn't recorded in the CRM within 60 seconds of conversion, your attribution data is already compromised.
For web-based acquisition, this means:
- Contact forms POST to an API that writes to the CRM and triggers notifications simultaneously
- UTM parameters are captured and stored with the lead record
- Page-level attribution (which page converted them) is recorded alongside channel attribution (which campaign brought them)
- Duplicate detection runs before creating a new contact (merge existing if match found)
Layer 2: CRM (The Source of Truth)
The CRM is not a sales tool. It's the source of truth for every customer interaction across every department. If it's not in the CRM, it didn't happen.
HubSpot vs Salesforce vs Pipedrive
The three dominant CRMs serve different operator profiles. The choice matters less than you think, as long as you commit fully to one.
HubSpot. Best for operators who want marketing + sales + service in one platform. The free tier is genuinely useful. The paid tiers ($800-$3,600/month for Professional/Enterprise) include marketing automation, content management, and reporting. Strengths: ease of use, native marketing features, excellent API. Weaknesses: custom object limitations, reporting depth, gets expensive fast at scale.
Salesforce. Best for operators with complex sales processes, multiple business units, or enterprise compliance requirements. Professional ($80/user/month) to Enterprise ($165/user/month). Strengths: unlimited customization, AppExchange ecosystem, enterprise features. Weaknesses: requires an admin, implementation takes 3-6 months, the UI fights you.
Pipedrive. Best for sales-led operators who need pipeline visibility without complexity. $14.90-$99/user/month. Strengths: pipeline UX, simplicity, affordability. Weaknesses: limited marketing features, basic reporting, weaker automation.
The decision framework:
- Under 20 employees, selling one product: Pipedrive
- 20-200 employees, marketing-led growth: HubSpot
- 200+ employees or complex enterprise sales: Salesforce
- If in doubt: HubSpot. The migration path to Salesforce is well-documented when you outgrow it.
For implementation patterns, read our CRM integration guide.
Layer 3: Automation
Automation handles the work that's important enough to do consistently but not complex enough to require human judgment.
The scope of marketing automation in 2026:
Email sequences. Lead nurture campaigns, onboarding flows, re-engagement campaigns, event follow-ups. Tools: HubSpot (native), ActiveCampaign, Customer.io. The differentiator isn't the tool — it's the segmentation logic and copy quality.
Lead scoring. Assigning numerical scores based on fit (company size, industry, role) and engagement (pages visited, emails opened, content downloaded). Scores determine when a lead moves from marketing to sales. Bad lead scoring wastes sales time. Good lead scoring increases conversion by 30-50% by ensuring reps talk to qualified prospects.
Workflow automation. Internal process triggers: when a deal reaches $50K+, notify the VP of Sales. When a trial user completes onboarding, add them to the upsell sequence. When a support ticket mentions a competitor, alert the account manager. Tools: native CRM workflows, Zapier/Make for cross-tool triggers, n8n for self-hosted.
AI-powered automation. In 2026, this means: AI-generated email subject line variants (measurable 10-15% open rate lift), AI lead scoring that factors behavioral patterns beyond simple point systems, and AI-powered chatbots that handle first-response qualification before routing to humans. It does not mean: fully autonomous sales processes. We're not there yet, and operators who pretend otherwise lose deals. See our voice AI agents guide for what's actually production-ready.
Layer 4: Analytics
Analytics answers three questions:
- What happened? (Descriptive — dashboards, reports)
- Why did it happen? (Diagnostic — attribution, cohort analysis)
- What will happen? (Predictive — forecasting, trend analysis)
Most marketing teams have too many dashboards showing descriptive metrics and zero systems answering diagnostic questions.
Attribution. The most important and most broken part of marketing analytics. Single-touch attribution (first-touch or last-touch) is wrong by design — it credits one interaction for a journey that involved 7-12 touchpoints over 3-6 months. Multi-touch attribution models (linear, time-decay, position-based, data-driven) are better but require clean data across the entire funnel.
The pragmatic approach:
- Use multi-touch attribution with a 90-day lookback window
- Weight first-touch (how they found you) and last-touch (what converted them) more heavily
- Accept that attribution will never be perfectly accurate
- Focus on directional trends rather than precise numbers
- Supplement with self-reported attribution ("How did you hear about us?" — often more accurate than any model)
Dashboards. One executive dashboard. One marketing operations dashboard. One per-channel dashboard (paid, organic, email, events). That's it. Every additional dashboard dilutes attention without adding signal. Use Looker, Metabase, or the CRM's native reporting. Read more in our dashboard problems article.
Alerting. Set threshold alerts for metrics that require action: lead volume drops below 80% of weekly average, cost per acquisition exceeds budget by 20%, website conversion rate drops below 2%. Dashboards show trends. Alerts catch anomalies.
Layer 5: Data
The unsexy layer that makes everything else possible.
Data warehouse. All marketing data — CRM records, web analytics, ad spend, email performance, event data — should flow into a central warehouse. Options: BigQuery (Google, pay-per-query, scales infinitely), Snowflake (enterprise-grade, expensive), PostgreSQL (self-managed, cheapest at small scale). For operators under $10M ARR, BigQuery's free tier is usually sufficient.
ETL / ELT. Extract, Transform, Load. Getting data from source systems into the warehouse. Tools: Fivetran ($1/MAR for CRM connectors), Airbyte (open-source), Stitch (HubSpot-owned). The key metric: data freshness. For marketing operations, hourly sync is sufficient. Real-time sync is overkill and expensive.
Reverse ETL. Taking processed data from the warehouse back into operational tools. Example: a lead scoring model runs in the warehouse, and the resulting scores are pushed back into HubSpot to trigger workflows. Tools: Census, Hightouch. This pattern enables analytics sophistication without moving processing logic into the CRM.
Customer Data Platform (CDP). A unified customer profile that combines data from all sources. Sounds like a CRM, but the scope is broader — CDPs include anonymous web visitors, ad impressions, and cross-device identity resolution. Tools: Segment, RudderStack (open-source). For most B2B operators under $20M ARR, a well-configured CRM + data warehouse replaces the need for a separate CDP.
The Integration Problem
The stack only works if data flows between layers. Integration is where most stacks break.
Native integrations. HubSpot ↔ Salesforce, HubSpot ↔ Google Ads, Salesforce ↔ Slack. These work reliably but are limited in scope. They sync standard fields and objects. Custom fields, custom objects, and complex logic require custom configuration.
Middleware. Zapier, Make (formerly Integromat), n8n. These handle event-driven integrations — "when X happens in tool A, do Y in tool B." Useful for bridging gaps between tools that don't have native integrations. Limitations: rate limits, error handling, and monitoring are primitive compared to custom code.
Custom integrations. API-to-API code running on your infrastructure. Highest flexibility, highest reliability, highest maintenance cost. Use for: high-volume data flows, complex transformation logic, business-critical integrations where Zapier's reliability isn't sufficient.
The decision tree:
- Does a native integration exist and cover your needs? → Use it
- Is the integration event-driven and low-volume (<1,000 triggers/day)? → Zapier/Make
- Is the integration high-volume, complex, or business-critical? → Custom code
- Do you need bi-directional sync with conflict resolution? → Always custom code
Building vs Buying
The build-vs-buy question applies at every layer.
Buy when:
- The category is mature and competitive (CRM, email, analytics)
- Your requirements are standard (lead capture, email sequences, basic reporting)
- The total cost of ownership including maintenance is less than building
- Time-to-value matters more than customization
Build when:
- Your requirements are genuinely unique (custom lead scoring models, industry-specific workflows)
- Integration complexity makes buying more expensive than building
- You need full control over data and processing
- The component is a competitive advantage, not a commodity
For most B2B operators, buy 80% and build 20%. The CRM, email platform, and analytics tool should be bought. The lead scoring model, the attribution logic, and the customer-facing AI features should be built.
The worst decision: building a CRM. The second worst: buying an AI solution when your data requirements are unique. Know which category each problem falls into.
The Migration Playbook
If your current stack is broken — and if you're reading this, it probably is — here's how to fix it without disrupting operations.
Phase 1 (Week 1-2): Audit. Map every tool, every integration, every data flow. Document what each tool does, what data it holds, and what would break if you removed it. This audit usually reveals 3-5 tools that are redundant and 2-3 integrations that are broken.
Phase 2 (Week 3-4): Design. Draw the target architecture using the five-layer model above. Identify which tools stay, which get replaced, and which get eliminated. Define the data model: what objects exist, what fields matter, how they relate.
Phase 3 (Month 2-3): Foundation. Set up the CRM as the source of truth. Migrate historical data with cleaned field mappings. Configure the core integrations (website → CRM, CRM → email). Test with a subset of real leads before going live.
Phase 4 (Month 4-6): Automation. Build workflows in the CRM. Set up lead scoring. Configure email sequences. Start with the 5 workflows that cover 80% of your operational needs.
Phase 5 (Month 7-12): Optimization. Add the analytics layer. Build attribution models. Create the dashboards (three, maximum). Set up alerting. Connect the data warehouse if volume warrants it.
Total timeline: 6-12 months. Total cost: $20,000-$80,000 depending on stack complexity and whether you use internal resources or hire external help. Payback period: 6-18 months through improved conversion rates, reduced manual work, and better lead quality.
FAQ
How much should we spend on our marketing stack?
The industry benchmark is 5-12% of marketing budget allocated to technology. For a company spending $500K/year on marketing, that's $25K-$60K/year on tools. This includes CRM, automation, analytics, and all supporting tools. If you're spending more than 15%, your stack is too complex. If you're spending less than 3%, you're probably under-investing in infrastructure.
Should we consolidate everything into HubSpot?
If your team is under 50 people and your sales process is straightforward, yes. HubSpot's all-in-one platform reduces integration complexity and lowers total cost. The tradeoff: you sacrifice best-of-breed capabilities in each category for simplicity. For most operators, that's the right trade. When you hit limitations, extend with the API rather than adding external tools.
How do we handle the transition without losing leads?
Run both systems in parallel for 30-60 days. New leads enter the new system. Existing leads and active deals stay in the old system until they close or age out. This avoids the "big bang" migration risk where everything switches at once and breaks. The parallel period costs more in tooling fees but eliminates the risk of lost leads during transition.
What's the role of AI in marketing ops in 2026?
AI is most valuable in three specific areas: lead scoring (behavioral pattern analysis beyond simple rule-based scoring), content personalization (adapting email and web content based on engagement patterns), and first-response qualification (AI chatbots that handle initial lead qualification before routing to humans). AI is least valuable for: strategic decisions, brand voice, and complex account management. Use AI to accelerate human work, not replace human judgment.
How do we measure if the stack is working?
Three metrics: (1) time-to-lead-response (target: under 5 minutes for inbound leads), (2) marketing-to-sales handoff quality (measured by sales acceptance rate — target: >70% of marketing-qualified leads accepted by sales), and (3) cost per opportunity (total marketing spend divided by opportunities created — should decrease over time as the stack matures). If these three metrics are improving quarter over quarter, the stack is working.
The stack isn't the strategy. It's the infrastructure that makes strategy executable. Build it as architecture, not as a tool collection. Every tool should justify its place in the system by the data it produces, the automation it enables, or the insight it provides. Everything else is overhead disguised as capability.