Custom Attributes: Create Your Own B2B Data Enrichments

Standard enrichment tools give you the same attributes as everyone else: email, phone, job title, company size. But what happens when your ICP is more specific? When you need scoring based on your unique criteria, or custom fields that don’t exist in any tool?

That’s exactly what custom attributes solve — personalized attributes you create yourself to enrich your data according to your precise needs. Instead of being limited to 50 standard fields, you can generate hundreds of custom columns: personalized scoring, custom industry classification, profile summaries, buying signal detection, and much more.

Create Your Own Attributes in Google Sheets

Use Claude and ChatGPT directly in your sheets to generate custom fields in seconds.

Try for Free →

What is a custom attribute and why you need one

A custom attribute is enriched data you create yourself to meet a specific business need — data that doesn’t exist in standard enrichment tools.

Take Sarah, Head of Sales at a SaaS startup selling to e-commerce companies. Standard tools give her the prospect’s position, company size, and “Retail” sector. But Sarah needs much more:

• Does this company sell on Shopify or WooCommerce?
• How many products are in their catalog?
• Are they already using a CRM?
• Are they in growth or stabilization phase?

This information doesn’t exist in any standard field. Sarah must create custom attributes to get it.

Most commonly used custom attribute types:

Now that you understand what a custom attribute is, let’s see why traditional methods no longer suffice and how AI changes the game.

Why standard attributes aren’t enough in 2026

Traditional enrichment tools work on a simple model: they have a fixed database with predefined fields. You get what exists in their catalog, period.

The problem? Every business is unique.

The limits of standard enrichments

Imagine Mike, Growth Marketer at an online training platform targeting B2B companies. With a classic tool, he gets:

• First name, last name, email ✅
• Position: “VP Sales” ✅
• Company: “TechCorp” ✅
• Size: 50-200 employees ✅
• Industry: “Software” ✅

But Mike needs to know:

• Does TechCorp already train their sales teams? ❌
• How many SDRs do they have? ❌
• Did they launch an onboarding program recently? ❌
• Are they using an LMS or not? ❌

This information is critical for qualification and personalization, but no standard tool provides it. Mike must either:

1. Do manual research → 10 minutes per lead = impossible to scale
2. Go without this info → conversion rate divided by 3
3. Create custom attributes → automate extraction of this data

The explosion of unstructured data

According to a 2026 Gartner study, 80% of data relevant for B2B prospecting is unstructured: LinkedIn posts, “About” pages, blog articles, job descriptions.

This data doesn’t fit into any predefined field. You need intelligence to:

• Extract relevant info
• Interpret it in your business context
• Transform it into actionable attributes

That’s exactly what AI-generated custom attributes enable.

Every ICP is different

Two companies selling to the same sectors never have the same Ideal Customer Profile. Take two SaaS companies targeting marketing agencies:

• SaaS A (reporting tool) looks for agencies with 10+ clients and at least 1 data analyst
• SaaS B (cold email platform) looks for agencies in scaling phase (active hiring) and without existing automation tool

Standard criteria “sector = Marketing” and “size = 20-50 employees” can’t distinguish these two ICPs.

Custom attributes let you create scoring and segmentation that reflects YOUR unique ICP, not that of all your competitors using the same tool.

How generative AI revolutionizes custom attributes

Until 2026, creating custom attributes required code, complex APIs, or data scientists. The arrival of generative AI (ChatGPT, Claude) changed everything.

AI as an enrichment engine

Before AI: You had to:

1. Write a rigid rule (“IF position CONTAINS ‘VP’ THEN score = 10”)
2. Manually map every possible case
3. Maintain dozens of if/else conditions

With AI: You simply describe what you want in natural language:

“Score this lead from 0 to 100 based on: seniority (30%), industry fit (25%), tech stack (20%), buying signals (15%), engagement (10%)”

AI understands context, interprets data, and generates a coherent score — even on profiles it’s never seen.

The 3 AI revolutions for custom attributes

1. Contextual understanding

AI doesn’t search for exact keywords. It understands meaning:

• It knows “Business Development Manager” = “Sales Manager”
• It detects that a company hiring 5 SDRs is in scaling phase
• It identifies subtle buying signals in a LinkedIn bio

Concrete example: You want to detect “hyper-growth” companies. Instead of coding rigid rules, you ask Claude:

“Is this company in hyper-growth? Analyze: employee count, evolution over 12 months, open jobs, recent funding rounds. Answer Yes/No + confidence level.”

AI analyzes all dimensions and gives you a nuanced verdict.

2. Structured content generation

AI can transform unstructured text into structured data:

• Summarize a 500-word LinkedIn profile into 2 actionable sentences
• Extract the 3 pain points mentioned on the “About” page
• Categorize a vague industry (“Retail” → “Luxury fashion e-commerce”)

Concrete example: From a LinkedIn bio, generate:

Seniority: Executive
Buying Power: Decision-maker
Industry Focus: B2B SaaS
Pain Points: Team scaling, automation, data quality
Fit Score: 92/100

3. Multi-step reasoning

AI can follow complex logic to score or qualify:

1. Analyze position → detect seniority level
2. Cross with company size → validate potential budget
3. Check tech stack → identify possible integrations
4. Scan recent posts → detect current pain points
5. Synthesize → final score + justification

This level of reasoning was impossible with classic if/else rules.

How to create your custom attributes step by step

Creating custom attributes happens in 3 phases: define what you want, configure the enrichment logic, and test/optimize. Let’s look at each step in detail.

Step 1: Define your custom attribute needs

Start by identifying business questions you can’t answer.

Ask yourself:

• What information am I missing to qualify a lead?
• What do I spend time on in manual research?
• How do I differentiate my best clients from average clients?
• What signals indicate a prospect is ready to buy?

Example: Define custom scoring

Emily, Sales Ops at a B2B recruiting platform, analyzes her last 50 won deals and identifies patterns:

• 80% have 10+ employees
• 95% are actively hiring (open jobs)
• 70% don’t yet use a modern ATS
• 60% are growing (headcount +20% over 12 months)

She creates custom scoring with these 4 weighted criteria.

Reflection template:

Expected result: A clear list of 3-7 custom attributes you want to generate.

Step 2: Configure AI generation

Now that you know what you want, you need to tell AI how to generate it.

2.1 — Choose the right AI model

Two main options in 2026:

• Claude (Anthropic): Best for complex reasoning, nuanced analysis, multi-criteria scoring
• ChatGPT (OpenAI): Very versatile, excellent for rapid extraction and classification

Tip: Test both on 10-20 examples and compare results.

2.2 — Write the optimal prompt

Your custom attribute quality depends 80% on prompt quality.

Recommended structure for a good prompt:

[CONTEXT]
You are a B2B qualification expert. Your role is to score leads for a SaaS company selling to marketing agencies.

[AVAILABLE DATA]
You will receive:
- Prospect's position
- Company size
- Detected tech stack (CRM, marketing tools)
- Recent LinkedIn activity

[WHAT YOU MUST GENERATE]
A score from 0 to 100 based on these criteria:
1. Seniority (30 points max): VP/Director = 30, Manager = 20, Specialist = 10
2. Company size (25 points max): 20-50 = 25, 10-20 = 15, <10 = 5
3. Tech stack (20 points max): HubSpot/Salesforce = 20, other CRM = 10, none = 0
4. Engagement signals (25 points max): posts about marketing automation = 25, general marketing posts = 10

[RESPONSE FORMAT]
Respond ONLY with:
Score: [number]
Reason: [1 short sentence explaining the score]

[EXAMPLE]
Input: VP Marketing, 35 employees, HubSpot, recent posts about automation
Output:
Score: 95
Reason: High seniority + ideal size + mature stack + automation engagement

Golden rules for custom attribute prompting:

1. Be specific: Give concrete examples of what you want
2. Constrain format: Force a response structure (score + reason, or JSON)
3. Weight criteria: Indicate relative weights of each factor
4. Give context: Explain your business and your ICP
5. Include edge cases: What to do if info is missing?

2.3 — Configure in your tool (Derrick example)

With Derrick, configuration happens directly in Google Sheets:

1. Select columns to analyze
   • Select cells containing your data (position, company, etc.)
2. Choose AI function
   • Ask Claude for complex reasoning
   • Ask OpenAI for rapid extraction/classification
3. Paste your prompt
   • Adapt the template above to your need
4. Launch enrichment
   • Derrick automatically generates your custom attribute for all leads
5. Retrieve results
   • A new column appears with your personalized attribute

Expected result: Your custom attribute is generated for your entire list in seconds.

Step 3: Test and optimize your attributes

Your first custom attributes will never be perfect on the first try. You need to iterate.

3.1 — Validate on a sample

Test first on 50-100 leads whose quality you already know:

• 30 leads you converted (your best clients)
• 30 leads you lost
• 20 unqualified leads

Check:

• Do good leads score high?
• Do bad leads score low?
• Are results consistent?

Validation example: Mike generates a fit score for 50 leads. He notices:

• ✅ Top 20% of score = 80% of past won deals → GOOD
• ❌ 15% of won deals score <50 → NEEDS IMPROVEMENT
• ❌ 25% of unqualified leads score >70 → PROMPT TO ADJUST

3.2 — Adjust prompt based on errors

Common errors and corrections:

Iterative improvement template:

V1 prompt → Test on 50 leads → Analyze errors
↓
V2 prompt (adjustments) → Test on 50 new leads → Compare V1 vs V2
↓
V3 prompt (final) → Deploy on entire database

3.3 — Automate for the future

Once your custom attribute is validated, automate it for all new leads.

Automation options:

1. Real-time enrichment
   • Each new lead in Google Sheets is automatically enriched
   • Via Zapier, Make, or webhooks
2. Weekly batch
   • Every Monday, enrich the week’s new leads
   • Via Google Sheets script or Derrick workflow
3. CRM trigger
   • When a lead enters the CRM, trigger custom enrichment
   • Via HubSpot, Salesforce, Pipedrive integration

Expected result: Your custom attributes generate automatically without manual intervention.

The 5 most powerful custom attribute use cases

Now that you know how to create custom attributes, let’s look at concrete applications that generate the most results.

1. Personalized scoring based on your unique ICP

The problem: Generic lead scoring tools don’t know your specific ICP.

The custom attribute solution: Create a fit score based on YOUR ideal customer criteria.

Real example — Lisa, SDR at a B2B ticketing platform:

Lisa targets professional event organizers. Her ICP:

• Organize 5+ events/year
• 200+ participants per event
• Already use a ticketing tool (but not Eventbrite Enterprise)

She creates a custom attribute “Event Fit Score” that analyzes:

• Number of events mentioned on site
• Average audience size (via LinkedIn posts)
• Detected ticketing tech stack

Prompt used:

Analyze this event organizer profile and website.
Score from 0-100 based on:
- Event volume (40 points): 10+ events = 40, 5-9 = 30, <5 = 10
- Event size (30 points): 500+ = 30, 200-500 = 25, <200 = 10
- Tech sophistication (30 points): Eventbrite Basic = 30, other tool = 20, manual = 5

Format: Score: X / Justification: [1 sentence]

Result: Lisa now prioritizes leads scoring >70, saving 12h/week of manual research.

2. Micro-industry classification

The problem: LinkedIn/Clearbit categories are too broad (“Retail”, “Software”).

The custom attribute solution: Create an ultra-precise custom taxonomy.

Real example — Anthropic (Clay user):

As mentioned in a Clay case study, Anthropic used Claude to recategorize prospects:

• Instead of “Retail” → “Luxury fashion e-commerce” vs “Food marketplace”
• Instead of “Software” → “B2B CRM SaaS” vs “B2C gaming mobile app”

Impact: 3x improvement in enrichment rate and 2x more personalized messages.

Typical prompt:

Categorize this company into ONE of these micro-industries:
- Fashion e-commerce
- Food e-commerce
- B2B marketplace
- Services marketplace
- CRM SaaS
- Analytics SaaS
- HR SaaS
- B2B Fintech
- B2C Fintech

Base on: company description, products, website.
Respond ONLY with category name.

3. Buying signal detection

The problem: The best moments to prospect are invisible with standard data.

The custom attribute solution: Generate a “Buying Signals” attribute that detects triggers.

Detectable buying signals:

• Recent funding round
• Mass hiring (5+ open jobs)
• Leadership change (new VP Sales)
• Geographic expansion (office opening)
• Product launch (announcement on LinkedIn)
• Strategic pivot (new keywords on website)
• Tech migration (mention “migrating from X to Y”)

Real example — Mike, Growth at a training platform:

Mike wants to detect companies actively training their teams. His “Training Signal” custom attribute searches for:

• Open jobs mentioning “training”, “onboarding”, “learning”
• LinkedIn posts about team training
• Career page mentioning “skills development”

Prompt used:

Analyze this company's data and identify active training signals.

Signals to search for:
- Jobs mentioning training/onboarding
- Posts about learning & development
- Career page discussing skill development
- Training programs mentioned

Respond:
Signal: Yes/No
Urgency: Low/Medium/High
Detail: [List of signals found]

Result: Mike contacts at the right time and doubles his response rate.

4. Smart profile summaries

The problem: Reading 50 LinkedIn profiles per day takes 3-4 hours.

The custom attribute solution: Generate an actionable summary in 2-3 sentences.

What a good summary should contain:

1. Seniority level and decision-making power
2. Main expertise and area of responsibility
3. Probable pain points
4. Suggested approach angle

Real example — Sophie, outbound SDR:

Instead of reading each full profile, Sophie generates a “Quick Summary” custom attribute:

Input: Full LinkedIn profile (500 words)

Output:

“VP Sales at 80-person SaaS, 5 years scaling XP. Manages team of 15 SDR/AE. Probable pain points: rep onboarding, fragmented tech stack, revenue forecasting. Approach: sales team efficiency.”

Prompt used:

Summarize this LinkedIn profile in 2-3 sentences for an SDR.

Include ONLY:
1. Position + seniority + team size
2. Key responsibilities
3. 2-3 probable pain points
4. Suggested approach angle

Style: concise, factual, actionable. Max 50 words.

Result: Sophie goes from 4h to 45min of research per day.

5. Personalization angle generation

The problem: Personalizing 200 emails per day is impossible manually.

The custom attribute solution: Generate a personalized “hook” for each prospect.

Real example — Mark, Founder doing outbound:

Mark generates a “Personalization Hook” custom attribute that analyzes:

• Prospect’s recent LinkedIn posts
• Company news
• Common points (alumni, events, connections)
• Detected pain points

Prompt used:

Generate a personalized opening line for this prospect.

Available data:
- Position and company
- Last 3 LinkedIn posts
- Company news
- Your ICP: [your business]

Rules:
- Max 15 words
- Explicit reference to specific element (post, news, context)
- No generic flattery
- Create natural connection

Format: [your opening line]

Example output:

“Saw your post on sales automation — we solve exactly that problem.”

Result: Mark sends 200 personalized emails/day and goes from 3% to 12% response rate.

Mistakes to avoid with custom attributes

Creating custom attributes is powerful, but certain mistakes can ruin your results. Here are the most common pitfalls and how to avoid them.

Mistake 1: Wanting to enrich everything at once

Symptom: You create 15 custom attributes at the same time and drown.

Impact: Impossible to validate quality, poorly adjusted prompts, exploding costs.

Solution: Start with 2-3 max attributes, validate them, then add progressively.

Prioritization framework:

1. Identify the 3 pieces of info you need MOST
2. Create ONE custom attribute for each
3. Test on 50 leads
4. Validate → Deploy → Move to next

Example: Emily only creates her “Fit Score” first. Once validated (1 week), she adds “Buying Signals”. Then “Personalization Hook” the next month.

Mistake 2: Prompts too vague

Symptom: AI returns inconsistent or too generic results.

Examples of bad prompts:

• ❌ “Score this lead”
• ❌ “Is this a good prospect?”
• ❌ “Tell me if they’re interesting”

Why it doesn’t work: AI doesn’t know what’s “good” or “interesting” for YOUR business.

Solution: Be ultra-specific about criteria and response format.

Transformation ❌ → ✅:

Before (vague):

Is this a good prospect?

After (specific):

Score this prospect from 0-100 based on:
- Seniority (40 pts): C-level = 40, VP = 30, Manager = 20, Specialist = 10
- Budget (30 pts): >100 employees = 30, 50-100 = 20, <50 = 10
- Tech stack (30 pts): HubSpot/Salesforce = 30, other = 15, none = 0

Format:
Score: [number]
Category: [Hot/Warm/Cold]
Reason: [1 sentence]

Mistake 3: Not validating before deploying

Symptom: You generate custom attributes on 10,000 leads without checking quality.

Impact: You discover too late that 40% of results are wrong or useless.

Solution: Always test on a representative sample BEFORE deployment.

3-step validation process:

Step 1: Generate on 50 varied leads (good clients, bad leads, edge cases)

Step 2: Manually validate results

• How many are correct?
• What types of errors?
• Pattern in errors?

Step 3: Adjust prompt based on detected errors

Minimum quality threshold: 80% correct results before large-scale deployment.

Mistake 4: Ignoring cost per enrichment

Symptom: You generate 50 custom attributes per lead and your bill explodes.

Impact: Cost of $0.50 per lead when your LTV doesn’t justify this investment.

Solution: Calculate ROI of each custom attribute before deploying.

ROI calculation framework:

ROI = (Generated Gain - Enrichment Cost) / Enrichment Cost

Where:
Generated Gain = Improved conversion rate × Nb leads × LTV
Enrichment Cost = Nb leads × AI cost per attribute

Concrete example:

Sophie generates a “Fit Score” custom attribute that improves her conversion rate from 3% to 5%.

• Monthly leads: 1000
• Average LTV: $500
• Enrichment cost: $0.03 per lead

Calculation:

• Gain = (5% – 3%) × 1000 × $500 = $10,000/month
• Cost = 1000 × $0.03 = $30/month
• ROI = (10,000 – 30) / 30 = 33,233%

→ Ultra-profitable custom attribute.

Rule of thumb: If ROI is <500%, question this attribute’s usefulness.

Mistake 5: Forgetting to maintain attributes over time

Symptom: Your custom attributes worked well in January but give inconsistent results in June.

Impact: Scoring and segmentation become obsolete, decisions based on outdated data.

Solution: Quarterly audit of your custom attributes’ performance.

Maintenance checklist:

□ Every 3 months: Check that results are still consistent □ At each ICP change: Adjust scoring criteria □ If performance drops: Re-test on sample and adjust prompt □ Documentation: Keep history of prompt versions

Example: Mike notices his “Event Fit Score” no longer detects good prospects. Analyzing, he realizes his ICP evolved (he now targets smaller events). He adjusts thresholds in the prompt.

Tools and technologies for generating custom attributes

Creating custom attributes requires the right tools. Here’s the complete 2026 ecosystem.

Derrick: Generative AI in Google Sheets

Why Derrick for custom attributes?

Derrick integrates Claude and ChatGPT directly into Google Sheets, allowing you to generate personalized attributes without leaving your spreadsheet.

Key features for custom attributes:

Typical workflow in Derrick:

1. Classic enrichment
   • Email Finder, LinkedIn Scraper, Phone Finder
   • → You have base data
2. Custom attribute generation
   • Ask Claude or Ask OpenAI on enriched columns
   • → You generate scoring, classification, signals
3. Export to CRM
   • Automatic push to HubSpot, Salesforce, or Pipedrive
   • → Your custom attributes arrive in your sales workflow

Derrick advantages vs alternatives:

Other ecosystem tools

Clay: Powerful but complex platform

• Strengths: 100+ data integrations, advanced workflows, custom formulas
• Weaknesses: Steep learning curve, high price ($349/month min)
• When to use: If you need very complex workflows with 10+ data sources

Zapier + ChatGPT API: DIY approach

• Strengths: Total flexibility, integration with 5000+ apps
• Weaknesses: Requires technical setup, variable costs
• When to use: If you already have a mature Zapier stack

Make (Integromat): Visual automation

• Strengths: Visual interface, cheaper than Zapier
• Weaknesses: Long initial setup, maintenance required
• When to use: To automate custom CRM to CRM enrichment

n8n: Open-source alternative

• Strengths: Free if self-hosted, total control
• Weaknesses: Requires technical skills, self-hosted = maintenance
• When to use: If you have a tech team and want maximum control

Direct APIs (Claude API, OpenAI API)

• Strengths: Maximum flexibility, per-token cost
• Weaknesses: Requires development, no interface
• When to use: If integrating into your product or custom workflow

Technical libraries and resources

For teams with developers:

Python + pandas

import anthropic
import pandas as pd

# Load your leads
df = pd.read_csv('leads.csv')

# Generate custom attribute with Claude
client = anthropic.Anthropic(api_key="your_key")

def generate_fit_score(row):
    prompt = f"Score this lead based on your ICP: {row.to_dict()}"
    response = client.messages.create(
        model="claude-sonnet-4-20250514",
        max_tokens=100,
        messages=[{"role": "user", "content": prompt}]
    )
    return response.content[0].text

df['fit_score'] = df.apply(generate_fit_score, axis=1)

Google Sheets + Apps Script

• Create custom functions directly in Sheets
• Call Claude or OpenAI API
• Generate attributes on demand

Useful documentation:

• Claude API Docs
• OpenAI API Docs
• Google Sheets Apps Script

Best practices: golden rules for custom attributes

To maximize the value of your custom attributes, follow these 7 proven principles.

1. Start simple, complexify progressively

Principle: One simple custom attribute that works > 10 complex half-reliable attributes.

Recommended roadmap:

Week 1: Create ONE basic scoring attribute

• Test on 100 leads
• Validate it works
• Deploy if >80% accuracy

Week 2-3: Add 2-3 complementary attributes

• Industry classification
• Buying signal detection
• Profile summary

Month 2: Optimize and automate

• Refine prompts based on sales feedback
• Automate generation for new leads
• Integrate into CRM

Month 3+: Experiment with advanced cases

• Personalization angle generation
• Churn prediction
• Re-engagement scoring

2. Document your prompts and versions

Principle: A custom attribute without documentation = impossible to maintain.

Documentation template:

## Custom Attribute: Event Fit Score

**Objective**: Prioritize event organizers with 200+ participants

**Data used**:
- Company website
- LinkedIn company page
- Job postings

**Scoring criteria**:
- Event volume (40%): 10+ = 40 pts, 5-9 = 30 pts, <5 = 10 pts
- Event size (30%): 500+ = 30 pts, 200-500 = 25 pts, <200 = 10 pts
- Sophistication (30%): Eventbrite = 30 pts, other = 20 pts, manual = 5 pts

**Prompt (v2.3 - 2026-01-15)**:

[insert exact prompt]

**Validated accuracy rate**: 87% on 150 leads **Changelog**: – v2.3 (2026-01-15): Added tech sophistication criterion – v2.2 (2026-01-10): Lowered event size threshold from 500 to 200 – v2.1 (2026-01-05): Added explicit weighting – v2.0 (2026-01-01): First validated version

Storage: Notion, Confluence, or simple shared Google Doc.

3. Create feedback loops with sales

Principle: SDRs/AEs are the best judges of a custom attribute’s quality.

Feedback process:

Weekly:

• Ask sales: “Do well-scored leads convert?”
• Identify false positives (high score but bad fit)
• Identify false negatives (low score but good fit)

Monthly:

• Analyze won deals: average score?
• Analyze lost deals: pattern in scoring?
• Adjust thresholds and weightings

Example: Sophie notices 30% of her won deals had score <60. Analyzing, she realizes her “Tech Stack” attribute penalizes companies without mature CRM too much — when they’re exactly her ICP. She adjusts weighting.

4. Combine standard + custom attributes for maximum value

Principle: Custom attributes are more powerful when combined with classic enrichment.

Optimal enrichment stack:

Layer 1 — Standard enrichment (email, phone, LinkedIn, firmographics)

• Tools: Derrick Email Finder, LinkedIn Scraper, Phone Finder
• Time: A few seconds per lead

Layer 2 — Custom attributes (scoring, classification, signals)

• Tools: Ask Claude, Ask OpenAI
• Time: A few seconds per lead
• Based on Layer 1 data

Layer 3 — Action (CRM push, email sequences, routing)

• Tools: Zapier, HubSpot, Salesforce
• Time: Automatic

Concrete example — Mike, Growth Marketer:

1. Standard enrichment → Gets email + position + LinkedIn + company size
2. Custom attribute “Training Signal” → Analyzes enriched data to detect if company trains actively
3. Custom attribute “Fit Score” → Combines size + position + training signal to score
4. Action → If Fit Score >70 AND Training Signal = Yes → Send email sequence A / Otherwise → Sequence B

Result: 2.5x higher response rate thanks to enrichment + custom attributes combination.

5. Test multiple AI models for the same attribute

Principle: Claude and ChatGPT have different strengths.

General rule:

Recommended A/B test:

1. Generate same custom attribute with Claude AND ChatGPT on 50 leads
2. Compare result quality
3. Deploy with best-performing model

Example: Emily tests “Fit Score” with both models:

• Claude: 87% accuracy, better nuance detection
• ChatGPT: 82% accuracy, 2x faster

→ She chooses Claude for accuracy (score is critical for her prospecting).

6. Use custom attributes for anti-targeting

Principle: Eliminating bad leads is as important as scoring good ones.

Disqualification custom attributes:

Example: Sophie creates a “Competitor Employee” attribute that detects if prospect works at a competitor. These leads are automatically excluded from her campaigns.

7. Measure business impact, not just technical accuracy

Principle: A 95% accurate custom attribute that doesn’t improve conversions is useless.

Business metrics to track:

Measurement framework:

Step 1: Establish baseline (before custom attributes)

• Measure current metrics over 1 month

Step 2: Deploy custom attributes

• On 50% of leads (A/B test)

Step 3: Measure impact after 1 month

• Compare test group vs control group

Step 4: Calculate ROI

• Time savings + Conversion improvement = Generated value
• Enrichment cost = Investment
• ROI = (Value – Cost) / Cost

Minimum threshold: If ROI <300%, question the attribute’s usefulness.

Key takeaways

• Custom attributes let you create personalized enrichment fields that standard tools don’t provide: custom scoring, tailored classification, detected buying signals, and much more.
• Generative AI transforms attribute creation: instead of coding rigid rules, you simply describe what you want in natural language and Claude or ChatGPT generates the data.
• Start simple with 2-3 priority attributes, validate on a sample of 50-100 leads before deploying, and iterate based on your sales teams’ feedback.
• The most profitable use cases are personalized scoring based on your unique ICP, real-time buying signal detection, and smart summaries that divide manual research time by 5.
• Derrick integrates Claude and ChatGPT directly into Google Sheets to generate custom attributes without leaving your spreadsheet, with credits that roll over month to month.
• Measure the real business impact of your custom attributes: conversion rate, time saved, and ROI rather than just technical accuracy of generated results.

Conclusion: Take action today

Custom attributes are no longer a luxury reserved for companies with data teams. With generative AI integrated into tools like Derrick, any sales or marketing team can create their own personalized enrichments in minutes.

Where to start?

1. Identify your #1 need: What info are you missing most to qualify leads?
2. Create your first attribute: Start with simple scoring or custom classification
3. Test on 50 leads: Validate that results are consistent
4. Deploy and measure: Track impact on your conversion rate

Generate Your Custom Attributes in 5 Minutes

Derrick integrates Claude and ChatGPT into Google Sheets. Create your first personalized attributes right now.

Try for Free →

FAQ

What’s the difference between a custom attribute and a classic calculated field?

A classic calculated field applies a rigid mathematical formula (e.g., IF position = "VP" THEN score = 10). A custom attribute uses AI to interpret context and reason in a nuanced way, even on unstructured data.

How much does generating a custom attribute cost?

With Derrick, each Ask Claude or Ask OpenAI call costs 1 credit (approximately $0.002 to $0.02 depending on plan). You can generate 100-500 custom attributes per dollar. Unused credits roll over to the next month.

Can I create custom attributes without technical skills?

Yes. With tools like Derrick, you simply describe what you want in English in a prompt, and AI generates the attribute. No code required.

How do I ensure my custom attributes comply with GDPR?

AI-generated custom attributes analyze data you already legally possess. Ensure your base enrichment is GDPR-compliant, and derived attributes will be too. Never ask AI to generate sensitive data it doesn’t have.

What’s the accuracy of AI-generated custom attributes?

With a well-designed prompt, expect 80-90% accuracy on scoring and classification tasks. For more complex tasks, 70-85%. Always validate on a sample before deployment.

Can I use custom attributes in my CRM?

Yes. Custom attributes generated in Google Sheets with Derrick can be automatically pushed to HubSpot, Salesforce, Pipedrive via Zapier, Make, or native integrations.