2023.09-Tartan-kyloe-icon
AI campaign (1200 x 500 px) (3)

  Home

Playbook: how to ensure success with AI

Not sure where to start with AI?
Follow these 4 steps so you can successfully adopt AI into your business.

Step one - getting started 

Playbook - step 1
Getting started with AI The basics

😎 Pick who gets it - Group your consultants into those who would benefit from it, who like change and are early adopters of technology and likely more junior. No point trying to teach an old dog new tricks, they will be replaced by people who do use it.

🗣️ Confirm the tone of voice and any regularity requirements.

📈 Review your data that will be used to feed the values in the prompts, this takes some time and is like a whole data map.

🤖 Find an AI savvy person who gets it and works out the prompts you need. Or get a partner like Kyloe to write them with you for the biggest use cases, that at the same time give a big increase in quality.

⚙️ Test on closed jobs in your database and compare the output with what you are looking for, small changes to the priming of prompts will make big differences, repeat for all.

✍️ Give it to all of your consultants and get feedback, repeat for 6 weeks.

✅ You have adopted -  now keep on top of changes.

🙌 Over all - about 120 hours of commitment, will repay in 3 months.

5 ways to shape early adoption of AI

We loved this report by McKinsey on what you should do as an early adopter of GenAI.

Our top 5 takeaways:

1. Eliminate the noise, and focus on the signal. 

Be honest about what pilots have worked. Cut down on experiments. Direct your efforts toward solving important business problems.

2. It’s about how the pieces fit together, not the pieces themselves.

Too much time is spent assessing individual components of a gen AI engine. Much more consequential is figuring out how they work together securely.

3. Get a handle on costs before they sink you.

Models account for only about 15 percent of the overall cost of gen AI applications. Understand where the costs lurk, and apply the right tools and capabilities to rein them in.

4. Tame the proliferation of tools and tech. 

The proliferation of infrastructures, LLMs, and tools has made scaled rollouts unfeasible. Narrow down to those capabilities that best serve the business, and take advantage of available cloud services (while preserving your flexibility).

5. Go for the right data, not the perfect data. 

Targeting which data matters most and investing in its management over time has a big impact on how quickly you can scale.

Step two - choosing the right AI model 

Playbook - step 2
Choosing the right AI model for recruitment:

ChatGPT-3.5, 4 Turbo, 4o, or Claude?

Untitled design (87)

An LLM is simply the tech that underpins the AI tool you input your prompt into, such as ChatGPT. It’s what generates the output in response to your prompt. Each LLM gives different outputs to the same prompt, so to ensure quality and efficiency, it’s important to understand the pros and cons of each.

But without trying them all, how can you know which is best? We’ve done the hard work for you. Keep reading to explore the capabilities of OpenAI's ChatGPT-4o mini, ChatGPT-4 Turbo, ChatGPT-4o, and Anthropic's Claude, and get our take on their best use cases in recruitment.
 
 
ChatGPT-3.5: Efficient for routine tasks

  • Ideal use cases: Generating job descriptions, summarizing candidate profiles and creating basic interview questions.
  • Strengths: Fast processing (3-6 seconds), suitable for text-based tasks with a context window of up to 3,000 words.
  • Limitations: Not ideal for complex queries or tasks requiring extended context retention.
  • Usage: 80% of Kyloe AI Assist prompts are handled by ChatGPT-3.5 due to its speed and efficiency.

ChatGPT-4 Turbo: Advanced analysis and tailored engagement

  • Ideal use cases: Analyzing multiple resumes against job criteria, conducting detailed candidate assessments, and generating tailored interview questions.
  • Strengths: Enhanced reasoning, longer context retention (up to 25,000 words), support text and image inputs.
  • Limitations: Longer response times (8-15 seconds) compared to ChatGPT-3.5.
  • Usage: 20% of prompts in Kyloe AI Assist are allocated to ChatGPT-4Turbo for more complex and detailed tasks.

ChatGPT-4o: Comprehensive recruitment support

  • Ideal use cases: Video interviews, real-time candidate evaluations, multimedia job postings.
  • Strengths: Fast processing, supports text, image, audio, and video inputs, improved accuracy.
  • Limitations: It is still in testing mode for some platforms but is expected to become a standard model soon.
  • Usage: As of 10th July 2024, ChatGPT-4o is integrated into Kyloe AI Assist - contact us for more details.


 

Claude: Enhanced context and ethical alignment while producing better communication outputs


  • Ideal use cases: High-level reasoning tasks, ethical decision-making in recruitment and detailed candidate profiling.
  • Strengths: Claude is the LLM that produces the best results when generating communication-based content (such as emails, interview questions, etc…). It has a stronger contextual understanding, and it’s designed with safety and ethical considerations in mind.
  • Limitations: Response times and accessibility might vary depending on deployment settings (specific configurations and conditions under which an AI model like Claude is implemented and operated).
  • Potential usage: Claude can be instrumental in ensuring fair and unbiased candidate evaluations, and adhering to ethical guidelines in AI-driven recruitment processes.
Claude 3.0 Haiku
"Haiku" is optimized for producing concise and impactful responses, similar to the structure and essence of a haiku poem. This makes it ideal for applications requiring succinct and meaningful outputs, such as summarization and brief content creation.
 
Claude 3.5 Sonnet
This model is designed to excel in generating creative and engaging text, particularly in artistic and literary contexts. The "Sonnet" variant emphasizes structured and poetic outputs, making it well-suited for tasks that require a balance of creativity and coherence.
 
So, what’s the conclusion on the best LLM?

There’s no single ‘best solution’ - it depends on the task at hand. And that’s why as part of Kyloe AI Assist, we’ve integrated multiple LLMs so that you can leverage the strengths of ChatGPT models and Claude to deliver powerful, efficient, and ethically aligned AI capabilities. Our customers love this for a few reasons:

  • Easily access multiple LLMs from within Bullhorn, with no need for copying and pasting between the LLM and Bullhorn.
  • We take care of the LLM accounts – you get the flexibility of different LLMs while avoiding the annoyance of setting up a different account for each LLM.
  • You can base your prompts on your Bullhorn data (such as name, skills, and years of experience), to power relevant results.

So whether you’re automating routine tasks, conducting in-depth candidate analyses, or providing multimedia support for interviews, Kyloe AI Assist will ensure you stay ahead in the recruitment game. Check our interactive demo here.

Step three - prompt writing 

Playbook - step 3
3 different types of prompts for using in recruitment
 
✅Single shot prompts  
✨Multi shot prompts
🌳Tree of thoughts

 

Put simply, the more words (tokens) you use, the better the output than if you used a few words in a single prompt. Lets start with a simple use case:

 

✅ Prompt 1: Single shot prompt

"Write me a job advert for this job description [Job description added here]"

This will give you a good answer but it will be very generic and sound like everyone else who uses that model and who asks the same simple question. It doesn’t follow any prompt framework, like the one we use of role, task, format (RTF).

 

✨ Prompt 2: Multi shot prompt  

"Acting as a recruiter, in a professional style, write me a job advert, for this job description, in 500 words, in a non technical manner [Job description added here]."

This will give you a better job advert, however, won't generate much that is specific to you.

 

🌳Prompt 3: Tree of thoughts

Fast forward to a very good method. This is where the 'tree of thoughts' comes into play. Let's think about how you would train a new consultant who was writing their first advert? What would you ask them to think of? It might go like this;

  1. Pick out the essential elements of a job description that you want to highlight.
  2. Use the company tone and style that maintains your companies voice - for example, be more formal than casual.
  3. Follow the guidelines you have been trained on, start with the company, then the role, then salary and benefits and finish with a strong call to action.
  4. What has worked in the past? Use an example.

Now lets think about how to write that down for someone to follow:

"Create a compelling job advert based on the provided job description by following these steps:

  1. Analyse the job description:
  •    Identify the key responsibilities and requirements of the role
  •    Determine the essential skills and qualifications needed
  •    Highlight any unique or attractive aspects of the position or company
  1. Draft an engaging introduction:
  • Begin with a strong hook that grabs the reader's attention
  • Briefly introduce the company and the role
  • Incorporate the company's mission, values, or unique selling points
  1. Outline the primary responsibilities:
  • Clearly and concisely list the main duties and tasks of the role
  • Use action-oriented language to make the responsibilities sound engaging
  • Prioritize the most important and appealing aspects of the job
  1. Specify necessary qualifications:
  • List the required education, experience, and technical skills
  • Include any preferred or desired qualifications that could set candidates apart
  • Mention relevant soft skills and personal attributes that align with the company culture
  1. Highlight benefits and perks:
  • Describe any unique or attractive benefits the company offers
  • Mention opportunities for growth, learning, and development within the role or company
  • Emphasize the company's commitment to work-life balance, if applicable
  1. Close with a compelling call to action:
  • Encourage interested and qualified candidates to apply
  • Provide clear instructions on how to apply and what materials to submit
  • Include a positive, welcoming statement to reinforce the company's inclusive culture
  1. Review and refine:
  • Read through the job advert to ensure it flows well and is free of errors
  • Make any necessary adjustments to improve clarity, conciseness, or persuasiveness
  • Ensure the tone and style align with the company's brand and target audience
  1. Add the Job description and other details here"

The main difference between a tree of thoughts prompt and a single-shot prompt is:  that the tree of thoughts approach breaks down the task into a series of smaller, interconnected steps. This allows the AI model to focus on each aspect of the job advert individually, ensuring that all essential components are addressed and well-crafted.

While a single-shot prompt can still generate a good job advert, the tree of thoughts approach encourages a more structured and thorough process, promoting better quality output and reducing the likelihood of missing important elements. By guiding the AI through each step, the tree of thoughts prompt helps create a more comprehensive and effective job advert, tailored to the specific requirements of the role and the company's needs.

10 levels of prompting - beginner to award winning

This is a really great video by Patrick Storm for getting started with prompt writing, level 1 to 10. 

Check it out here

Step four - choosing how to make AI work for you

Playbook - step 4
3 options when adopting AI into your business
To get the best out of AI in your business, one of three things has to happen:

💪You gain prompting skills internally to make the magic happen (see step three for help) 

🤝You find a partner who can work with you to get those prompts working with your data, to give you what you need

👏You find a product that does it for you.

 Each option has its own pros and cons, here's what we think: 

 💪Gaining prompting skills internally:

Pros:

  • Most cost-effective in the long run
  • Tailored to your specific business needs
  • Builds in-house expertise
  • Allows for quick adjustments and experimentation

Cons:

  • Time-consuming to learn and master

  • May require ongoing training as AI evolves

  • Potential for inconsistent results across team members
  • Relies on employee retention to maintain skills

🤝 Finding a partner to work with:

Pros:

  • Access to specialized expertise
  • Faster implementation
  • Can provide broader AI strategy and insights
  • Scalable based on your needs

Cons:

  • Potentially higher costs
  • Dependency on external party
  • May require time to find the right partner
  • Possible communication challenges or misalignment of goals

👏 Finding a product that does it for you:

Pros:

  • Often easier to implement and use
  • Typically more stable and reliable
  • Regular updates and improvements
  • Usually comes with customer support

Cons:

  • May be less flexible or customizable
  • Potential for higher ongoing costs
  • Might not perfectly fit all your specific needs
  • Risk of vendor lock-in
  • The product cannot keep up with changes


Test your AI knowledge
with our quiz! 

How good is your AI knowledge? Take our quick AI quiz to find out if you're an AI beginner, AI savvy or an AI expert! 

AI QUIZ header  (2)


Try Kyloe AI assist
free for 2 weeks

Get your 2-week free trial and experience our new AI solution for yourself. Dramatically improve speed, efficiency and quality with our reliable Bullhorn GenAI solution.

Untitled design (12)