Episode XXIX: The Prompt Engineering (How to Make a Toddler Act & Talk Nice :)
Authors: William Caban , Fatih E NAR, Anil Sonmez
Let’s be honest. You’ve probably tried throwing some basic questions at ChatGPT/Claude/Groq about your 5G rollout strategy and gotten back responses that sound like they were written by someone who learned about telecom from watching YouTube commercials (like a toddler). The good news? It’s not the toddler’s fault — it’s your homeschooling failure, young parent!
In telco, precision matters (because nobody wants their network going down during a high-traffic/emergency), effective prompt engineering isn’t just a nice-to-have -> it’s existential.
While model training & fine-tuning can be an option, it could be a waste of time & money if you have not identified and understood the model’s limitations. Finding these limits is done by developing , well engineered & purposeful prompts with optimization cycles. These cycles help develop the right prompts associated with a specific foundation model that can deliver the desired outcomes for your use-case. These iterative optimization processes will lead to value in a shorter timeframe. Whether you’re analyzing network performance, creating content strategies, or troubleshooting customer issues. It is only after evaluating the results of this optimization cycle that we can quantify the limits of the model (if any) for the targeted use-case.
The optimization cycle consists of:
- Role assignment that distills domain expertise.
- Chain of thought which ensures transparent reasoning.
- Few-shot examples to guide the model’s behavior.
- Smart constraints maintain domain value boundaries.
- Benchmarks measure effectiveness.
Role Assignment: Dress Your AI for Success
The Problem
Generic AI responses are like wearing flip-flops to a wedding -> technically functional, but completely inappropriate for the context (except maybe Vegas weddings).
The Solution: Be Specific About Who You Want AI to Be
Bad Prompt:
Analyze our network performance data.Good Prompt:
You are a senior network operations engineer with 10+ years of experience in LTE and 5G networks.
You specialize in performance optimization and have been working with major carriers like Verizon and AT&T.
Analyze the attached network performance data focusing on latency issues, throughput bottlenecks, and capacity utilization patterns.
Provide actionable recommendations that a NOC team could implement within the next 30 days with minimal to no service outage.Industry-Specific Role Examples
For Content Strategy:
You are a streaming media content strategist who has worked with Netflix, Disney+, and HBO Max.
You understand viewer engagement metrics, content acquisition costs, and regional content preferences.For Customer Service:
You are a tier-2 technical support specialist for a major telecommunications provider.
You have experience in troubleshooting residential broadband, mobile connectivity, and VoIP services.
You communicate complex technical concepts in simple terms that customers can understand.Pro Tip: Don’t just assign roles -> give your AI a backstory as well. The more context you provide about their “experience, focussed knowhow area” and specialization, the more targeted and beneficial their responses will be.
Chain of Thought: Teaching AI to Show Its Work
The Magic of “Think Step by Step”
In telco, decisions often involve complex technical considerations and trade-offs. Chain of thought prompting forces AI to break down its reasoning process, making it easier to spot errors and understand the logic behind recommendations (like explaining a 4yo why they shall not talk while eating their food, don’t just say cos you are making a mess).
Example: Network Capacity Planning
Without Chain of Thought:
How many cell towers do we need for downtown coverage?Result: 42 :) Or another magical number with no justification or little explanation which does not make any sense for an expert eye.
With Chain of Thought:
You are a radio frequency engineer planning cell tower deployment for downtown coverage.
Think through this step by step:
1. First, analyze the coverage area dimensions and terrain.
2. Then, calculate the expected user density and traffic patterns.
3. Next, determine the coverage radius per tower based on frequency band and power levels.
4. Consider interference patterns and overlap requirements.
5. Factor in capacity requirements based on expected data usage.
6. Finally, provide your tower count recommendation with justification for each step.
Coverage area: 2.5 square miles, mostly high-rise buildings
Expected peak users: 50,000 concurrent.
Frequency: 3.5 GHz (mid-band 5G).
Target: 100 Mbps average download speed per user.Chain of Thought for Content Analysis
Prompt:
You are a streaming analytics expert. Analyze this show's performance data using the following process:
1. Examine the viewership trend over the first 30 days.
2. Compare completion rates by episode.
3. Identify demographic patterns in the audience.
4. Assess social media engagement correlation.
5. Predict renewal likelihood based on these factors.
6. Recommend specific actions for season 2 (if applicable).
Show your reasoning for each step and explain how you arrived at your conclusions.Few-Shot Examples: Training AI with Real-World Scenarios
The Power of Pattern Recognition
Few-shot prompting is like giving AI a few examples of exactly what you want before asking it to do the task. In telco, this is incredibly powerful because industry-specific outputs have particular formats and requirements.
Example: Network Incident Reports
You are a network operations specialist writing incident reports. Here are examples of the format and tone:
EXAMPLE 1:
Incident: #NOC-2024–0156
Time: 14:23 EST - 14:47 EST
Impact: Service degradation affecting 12,000 customers in Seattle metro area
Root Cause: Fiber cut on primary trunk between SEA-01 and SEA-03 nodes
Resolution: Traffic automatically rerouted via backup path, primary path restored
Action Items: Coordinate with construction company to prevent future cuts, review path diversity
EXAMPLE 2:
Incident: #NOC-2024–0143
Time: 09:15 PST - 11:22 PST
Impact: Complete service outage affecting 8,500 customers in Phoenix suburbs
Root Cause: Power failure at PHX-02 cell site, backup generators failed to activate
Resolution: On-site technician manually started backup power, primary power restored by utility
Action Items: Replace faulty generator transfer switch, audit all backup power systems
Now write an incident report for: [your incident details here]Example: Content Performance Analysis
You are a content analyst creating performance summaries. Follow these examples:
EXAMPLE 1:
Title: "Slow Horses" S4 Performance Analysis
Launch Week Views: 87M hours (↑23% vs S3)
Completion Rate: 73% (↑8% vs S3)
Key Demographics: 65% ages 18–34, 58% international
Social Engagement: 2.3M mentions, 89% positive sentiment
Recommendation: Greenlight S5, increase international marketing budget
EXAMPLE 2:
Title: "The Crown" S6 Performance Analysis
Launch Week Views: 34M hours (↓15% vs S5)
Completion Rate: 81% (↑3% vs S5)
Key Demographics: 72% ages 35–65, 45% UK/Commonwealth
Social Engagement: 890K mentions, 76% positive sentiment
Recommendation: Natural series conclusion, develop royal documentary spin-offs
Now analyze: [your content data here]Constraint Settings: Keeping AI in Its Lane
Why Constraints Matter in Telecom & Media
Without proper constraints, AI might suggest solutions that violate regulations, exceed budgets, or ignore technical limitations. Smart constraints keep responses practical and implementable.
Technical Constraints Example:
TECHNICAL LIMITS:
- Maximum tower height: 150 feet due to FAA regulations
- Available spectrum: 3.7–3.98 GHz (280 MHz total)
- Power limits: 1600W EIRP maximum per sector
- Backhaul: Fiber available to 70% of planned sites, microwave backup required
BUSINESS CONSTRAINTS:
- Budget: $50M total capex for this phase
- Timeline: 18 months from approval to service launch
- Coverage target: 85% of metropolitan area population
- Capacity target: 500 Mbps average user speeds during peak hours
REGULATORY REQUIREMENTS:
- Environmental review required for sites near protected areas
- Historic preservation review for sites in downtown core
- Minimum 6-month lead time for zoning approvals
Provide your network design recommendations within ALL of these constraints.You are designing a 5G network expansion plan. Work within these constraints:
Content Budget Constraints:
You are a content acquisition strategist working within these constraints:
BUDGET LIMITS:
- Total content budget: $2.1B annually
- Maximum per-title: $25M for originals, $5M for licensed content
- International content: Must represent 40% of total originals budget
- Documentary allocation: Minimum 15% of originals budget
STRATEGIC REQUIREMENTS:
- Must deliver 150 original titles annually
- 60% must target 18–34 demographic
- All originals must be available globally (except where prohibited)
- Minimum 2-year exclusive licensing terms for acquired content
PLATFORM CONSTRAINTS:
- Maximum 4K/HDR for premium tier only
- Multi-language dubbing for top 8 markets required
- Closed captioning in 12 languages minimum
Create a content acquisition strategy that works within ALL these parameters.Different Patterns for Full Telco House
The “Stakeholder Perspective” Pattern
Different departments care about different metrics. This pattern helps AI tailor responses to specific audiences.
Analyze our fiber deployment progress from three perspectives:
ENGINEERING PERSPECTIVE (focus on technical metrics):
- Network performance and capacity utilization
- Construction challenges and solutions
- Equipment performance and reliability
FINANCIAL PERSPECTIVE (focus on business metrics):
- ROI and payback periods by market
- Capital efficiency and cost per mile
- Revenue impact and subscriber growth
EXECUTIVE PERSPECTIVE (focus on strategic metrics):
- Market share implications
- Competitive positioning
- Regulatory compliance status
Present findings in separate sections tailored to each audience.The “Scenario Planning” Pattern
Perfect for the uncertainty inherent in telecom and media markets.
You are a strategic planning analyst. Create three scenarios for our streaming service growth:
OPTIMISTIC SCENARIO (assume everything goes right):
- 25% annual subscriber growth
- Content costs remain flat
- No major competitive launches
- Successful international expansion
REALISTIC SCENARIO (assume typical market conditions):
- 12% annual subscriber growth
- 8% annual content cost inflation
- 2–3 new major competitor launches
- Mixed international results
PESSIMISTIC SCENARIO (assume significant challenges):
- 5% annual subscriber growth
- 15% content cost inflation due to bidding wars
- Major tech company enters with massive content budget
- Regulatory restrictions in key international markets
For each scenario, provide:
- Financial projections (revenue, costs, profitability)
- Strategic recommendations
- Key risk mitigation strategiesThe “Competitive Intelligence” Pattern
You are a competitive analyst specializing in telecommunications. Analyze [competitor] using this framework:
NETWORK STRATEGY:
- Technology choices (5G bands, equipment vendors)
- Coverage priorities (urban vs rural focus)
- Infrastructure investments and timeline
PRICING STRATEGY:
- Plan structures and pricing positioning
- Promotional strategies and customer acquisition costs
- Value proposition differentiation
CUSTOMER EXPERIENCE:
- Service quality metrics vs industry benchmarks
- Customer satisfaction scores and pain points
- Digital experience and self-service capabilities
Based on this analysis, identify:
1. Three strategic vulnerabilities we could exploit
2. Three strengths we need to counter or match
3. Recommended competitive response tacticsCommon Pitfalls and How to Avoid Them
Pitfall #1: The Jargon Trap
Problem: Using too much industry jargon confuses AI.
Solution: Define terms or ask AI to explain them back.
Bad:
Optimize our CAPEX allocation for FTTH deployment considering PON architecture constraints.Better:
You are a fiber network planning engineer.
Optimize our capital expenditure allocation for fiber-to-the-home deployment.
Consider these factors:
FTTH = Fiber to the Home (bringing fiber directly to residential customers)
PON = Passive Optical Network (the splitting technology we use)
CAPEX = Capital expenditure budget
Focus on maximizing homes passed per dollar spent while maintaining service quality standards.Pitfall #2: The “Do Everything” Prompt
Problem: Asking AI to solve world hunger and achieve world peace in one prompt .
Solution: Break complex tasks into sequential prompts.
Bad:
Create a complete 5G strategy including network design, spectrum planning, device ecosystem, pricing, marketing, and competitive analysis.Better:
Part 1: You are a radio frequency engineer.
Design the technical foundation for our 5G network rollout, focusing on spectrum allocation and tower placement strategy.
[Get response, then continue with Part 2]
Part 2: You are a product marketing manager. Using the technical strategy from Part 1, develop pricing and go-to-market strategies for our 5G services.Pitfall #3: Ignoring Output Format
Problem: Getting walls of text when you need structured data .
Solution: Specify exactly how you want the output formatted.
Specify Format:
Present your network optimization recommendations in this format:
## EXECUTIVE SUMMARY (3 bullet points max)
## TECHNICAL RECOMMENDATIONS
| Priority | Action | Timeline | Cost Est | Impact |
| - - - - - | - - - - | - - - - - | - - - - - | - - - - |
| High | [action] | [time] | [cost] | [impact] |
## IMPLEMENTATION ROADMAP
**Phase 1 (0–3 months):** [details]
**Phase 2 (3–6 months):** [details]
**Phase 3 (6–12 months):** [details]
## RISK MITIGATION
[Key risks and mitigation strategies]Industry-Specific Prompt Libraries
Network Operations Prompts
Performance Analysis:
You are a senior network performance engineer.
Analyze the attached performance data and:
1. Identify the top 3 performance bottlenecks
2. Quantify the impact on customer experience
3. Recommend specific optimization actions
4. Estimate implementation time and resources needed
5. Calculate expected improvement metrics
Focus on actionable insights that our NOC team can implement.Capacity Planning:
You are a network capacity planner.
Project our bandwidth needs for the next 18 months considering:
- Historical traffic growth patterns (attach data)
- Upcoming service launches (5G, IoT, enterprise solutions)
- Seasonal traffic variations
- Competitive market dynamics
Provide month-by-month capacity requirements and investment recommendations.Content Strategy Prompts
Content Performance Analysis:
You are a streaming analytics expert.
Evaluate this content’s performance using industry-standard metrics:
ENGAGEMENT METRICS: View-through rates, completion rates, re-watch behavior
AUDIENCE METRICS: Demographic breakdown, geographic distribution, device usage
ACQUISITION METRICS: How this content drives new subscriber sign-ups
RETENTION METRICS: How this content affects churn rates
Provide specific recommendations for content programming and marketing strategies.Content Acquisition Strategy:
You are a content acquisition specialist.
Evaluate this licensing opportunity:
CONTENT: [Title/Description]
ASKING PRICE: [Amount]
TERM: [Length]
TERRITORY: [Geographic scope]
Analyze:
1. Comparable content performance on our platform
2. Expected viewership and engagement
3. ROI projection over the license term
4. Strategic value (genre diversity, audience expansion, etc.)
5. Negotiation recommendations
Provide a clear go/no-go recommendation with supporting rationale.Customer Service Prompts
Technical Troubleshooting:
You are a tier-2 technical support specialist.
A customer reports: [issue description]
Walk through your troubleshooting process:
1. Information gathering questions to ask
2. Diagnostic steps to perform remotely
3. Most likely root causes based on symptoms
4. Resolution steps in order of complexity
5. When to escalate to tier-3 or dispatch technician
6. Follow-up actions to prevent recurrence
Keep explanations customer-friendly while being technically accurate.Measuring Prompt Effectiveness
Key Metrics for Telecom & Media Prompts
Accuracy Metrics:
- Technical accuracy of recommendations
- Compliance with industry regulations
- Alignment with business constraints
Efficiency Metrics:
- Time saved vs. manual analysis
- Reduction in back-and-forth clarifications
- Speed of implementation
Quality Metrics:
- Actionability of recommendations
- Stakeholder satisfaction with outputs
- Success rate of implemented suggestions
A/B Testing Your Prompts
Version A (Basic):
Analyze our customer churn data and recommend retention strategies.Version B (Optimized):
You are a customer lifecycle analyst specializing in telecommunications.
Analyze the attached churn data focusing on:
1. Churn patterns by customer segment (postpaid/prepaid, consumer/business)
2. Key churn indicators and timing patterns
3. Revenue impact analysis
4. Recommended retention tactics with expected effectiveness
5. Implementation roadmap and resource requirements
Present findings in executive summary format with supporting data tables.
Measure: Response quality, time to insight, implementation success rate.Balancing System vs User Prompts
In enterprise telco environments, not every user has the expertise or time to craft optimal prompts. Network engineers focus on RF optimization, not prompt engineering. Content analysts prioritize viewership metrics, not AI instruction design. Business executives need quick insights, not prompt crafting sessions. This creates a fundamental challenge: How do you democratize AI effectiveness across skill levels while maintaining output quality?
System Prompts as Intelligence Amplifiers
Smart organizations solve this through strategic system prompt design that acts as platform-level guardrails. System prompts can compensate for basic prompt deficiencies by:
- Auto-applying industry constraints and regulatory compliance
- Enforcing output formats that match business requirements
- Injecting domain expertise when user prompts lack technical depth
- Preventing dangerous recommendations through built-in safety checks
Example — Basic User Prompt:
User: “How to improve network performance?”System Prompt Compensation:
System: You are a senior network optimization engineer.
When users ask about performance improvements, always:
1. Request specific metrics (latency, throughput, packet loss)
2. Consider regulatory compliance (FCC, local zoning)
3. Include budget impact analysis
4. Provide implementation timelines
5. Highlight customer impact risks
Result: Even a vague user prompt generates comprehensive, actionable output.Efficiency Through Intelligent Defaults
System prompts reduce cognitive load by embedding institutional knowledge:
Traditional Approach: Each user must learn and remember:
- Industry-specific constraints.
- Company policies and budgets.
- Regulatory requirements.
- Output formatting standards.
- Escalation procedures.
System Prompt Approach: Platform handles complexity automatically:
- Auto-applies company-specific constraints.
- Ensures consistent output quality across users.
- Reduces training overhead for new team members.
- Maintains compliance without user expertise.
Well-designed system prompts can multiply user effectiveness:
- Novice Users: Get expert-level guidance automatically .
- Experienced Users: Focus energy on domain-specific details rather than prompt mechanics.
- Executives: Receive consistent, decision-ready outputs regardless of who’s asking.
Implementation Strategy
Layer 1 (System): Organizational standards and guardrails
- Compliance requirements
- Budget constraints
- Risk thresholds
- Output formats
- Escalation triggersLayer 2 (User): Task-specific context and requirements
- Specific data sets
- Immediate objectives
- Timeline constraints
- Stakeholder requirementsThis approach transforms AI from a tool requiring expertise to a platform delivering expertise. Please visit our Telco-SME repo under Telco AI Experiments for more examples for different intent (cloud expert, network expert, customer support etc) driven system prompts.
Closure: Clueless Toddler -> Experienced Professional
Effective prompt engineering in Telco isn’t about being clever with words, it’s about being strategic with context and being smart with the approach. The best prompts combine deep industry knowledge with clear communication guidelines and principles.
Remember the key principles:
- Role Assignment: Make AI an expert, not a generalist.
- Chain of Thought: Force AI to show its work.
- Few-Shot Examples: Train with real scenarios.
- Smart Constraints: Keep solutions practical.
- Benchmarks: Get results you can use to improve outcomes.
The difference between a $MM AI implementation that gathers dust and one that transforms your business often comes down to the quality of your understanding on your AI platform & capabilities together with well benchmarked models to run on. Master these techniques, and you’ll turn AI from an expensive experiment into your competitive advantage.
While identical prompts will behave differently across different models, following these principles will provide a structured approach to identifying an optimized prompt that works with a different model for your use case.
Please avoid jumping directly into fine-tuning models. Don’t waste time and money if you haven’t first discovered the limits of purposefully built, optimized prompts for your use-case that can truly work with your model. Smart organizations use expertly crafted prompts to achieve the maximum performance of their AI applications in weeks, not years. With modern models, this might be enough for achieving a production-ready AI application. Only when evaluations demonstrate that a limitation exists (e.g., a required performance accuracy of >90% for a specific task), then it is time to consider alternatives like fine-tuning a small/medium model for that particular task. The future belongs to those who can extract maximum value from cutting-edge open source foundational models through smart prompt engineering.
Now stop reading and start prompting :). Your network won’t optimize itself, and neither will your operations strategy. But with the right prompts, AI could help you do both better than you ever thought possible.
