From Parliament to AI: An Unexpected Journey

An interview with Joseph Lewis, founder Elect AI, who's helping businesses, charities, and job seekers transform their workflows with practical AI training that delivers immediate results. 

Joseph Lewis, founder Elect AI.

Joseph Lewis didn't set out to become an AI consultant. After working in UK Parliament and international public affairs, he discovered that AI could revolutionise how people work-saving every trained person at least an hour daily. Now, through Elect AI, he's trained hundreds of people with plans to reach thousands more, delivering measurable outcomes like reducing 3-week processes to 3 days and enabling de facto 4-day work weeks. In this conversation, Joe shares practical insights on AI tools, job search strategies, security considerations, and why most organisations are getting AI training completely wrong. 

Q: You studied political science and worked in Parliament. How did you end up founding an AI training company? 

I've always been interested in politics and communications. I studied political science at Leeds University and worked in Parliament, doing communications and public affairs work. After that, I went to Dubai to work for a consultancy on projects like helping Saudi Arabia modernise their economy. 

While I was in Dubai, I met someone who had started her own AI consultancy. I did some freelance work helping with training programmes, and I thought, "This would be really useful for people in Parliament." She was doing big transformation consulting, but I saw an opportunity to focus specifically on practical training for politicians. 

I started Elect AI in early 2025, initially targeting politicians. But through word-of-mouth, I quickly got interest from businesses and charities. Now, most of my clients are outside politics-businesses, charities, and organisations that want their teams to actually use AI effectively, not just tick a training box. 

What Makes Elect AI Different: Process Transformation Over Tool Training 

Q: There are countless AI consultants out there. What makes your approach different? 

The fundamental difference is that we focus on process transformation, not just tool training. Most organisations make a critical mistake: they buy generic video courses, tell employees to watch them, tick the box, and move on. Nothing actually changes. 

What I tell people is that AI is going to reinvent what being competent looks like. You can't just fiddle around with ChatGPT and expect results. You need to actually stop, pause, and engage with how AI can fundamentally change your workflows. That means understanding: 

• Different AI tools are better at different things 

• You need to train the AI to understand how you work and communicate 

• Security settings must be properly configured 

• You have to change your processes, not just add AI to existing ones 

For example, instead of writing a correspondence letter and then checking it with AI, you should train the AI to understand what you know, have it write the response, and then you check it. It's a complete mindset shift. 

Q: What kind of results are your clients seeing? 

Every single person I've trained saves at least an hour every single day. But we're seeing much more dramatic results: 

• Process Improvement: Reduced a process that took 3 weeks end-to-end down to 3 days 

• Annual reports: People who were putting them off for months now complete complex, long-form reports in 1-2 hours 

• Daily time savings: Most people save 1-4 hours per day consistently 

• 4-day work weeks: One person who works in Parliament now operates on a de facto 4-day work week because he's front-loaded all his work using AI 

The key is that these results happen immediately. Following a 90-minute session or half-day seminar, people know exactly how to benefit and speed things up right away. 

The Tool Philosophy: Why Joe Doesn't Use ChatGPT 

Q: You mentioned not using ChatGPT. What AI tools do you recommend and why? 

I don't use ChatGPT at all, actually. Here's why: being comfortable with AI isn't about fiddling around with one tool. It's understanding that different AI tools work better for different tasks. Some of my recommendations: 

Claude: This is my go-to for professional use. It's fundamentally the smartest AI in my opinion, and it was designed specifically for professional contexts. For business communications, complex problem-solving, and professional writing, Claude outperforms ChatGPT. 

Perplexity: Fantastic for research and has a hidden gem-a built-in language tutor feature. This is amazing for people changing jobs across countries who need to learn a new language. Most people don't even know these features exist. 

ChatGPT: While it's the most widely used, it has significant limitations. It produces formulaic outputs using words like "seamlessly," "delve," "rooted," and "embark"-words people don't actually use. This creates professional embarrassment. Somebody I know received 15 applications for a vacancy. Every single one had clearly been written with ChatGPT with no training at all. 

There are also specialised tools for different functions-note-taking, project management, budget management. Understanding which tool to use for what purpose is crucial. 

Security and Privacy: Demystifying the Concerns 

Q: Security is a major concern, especially in politics and business. How do you address this? 

Security concerns often stem from misunderstandings. In my training, I cover: 

Safety features to enable: There are specific settings you need to configure to ensure your information doesn't go anywhere it shouldn't. 

Demystifying misconceptions: For example, there was an incident where ChatGPT content ended up being regurgitated to other people. What actually happened was that people had taken their chats and published them on websites. ChatGPT scraped those public websites-not the private chats themselves. If you put your bank details into ChatGPT, it won't give them to someone else who asks. 

Proper configurations: With the right safety protocols in place, you can use AI securely. The key is knowing what to do and actually doing it. 

Once people understand these points, they're able to use AI confidently and benefit immediately while maintaining security. 

AI for Job Search and Career Transitions: Quality Over Quantity 

Q: For someone in career transition or job searching, how can AI help without making them sound like a robot? 

Large language models are inspired by the human brain, so fundamentally, anything you can think to do, they can help with. The applications for job search include: 

• CVs and cover letters: But with a critical warning-you must train the AI to speak in your voice, not just copy-paste generic outputs 

• Interview preparation: Practise responses, research companies, prepare questions 

• Language learning: Using Perplexity's tutor feature if you're moving countries 

• Application tracking: Tools like Monday.com, Sanebox, Spike, and Gemini integration help organise multiple applications and follow-ups 

The golden rule is quality over quantity. Yes, people today might need to apply to tens or even hundreds of jobs. But you're better off doing one or two really good, customised applications per day with AI than 20 generic ones. 

Q: What are the biggest mistakes people make when using AI for job applications? 

Not training the AI to understand their voice and communication style. This is the difference between sounding authentic and sounding like every other ChatGPT-generated application. 

There's a really simple trick you can use to avoid the formulaic phrasing-words like "seamlessly" and "delve" that instantly flag your application as AI-generated. The trick is part of what I teach in training sessions, but the principle is: make the AI learn from your actual writing, your tone, your experiences. 

When that organisation received those 15 applications, they could immediately tell none of the candidates had done this work. It was all generic, un-customised ChatGPT output. Those applications went straight to the rejection pile. 

AI in Recruitment: A Double-Edged Sword 

Q: On the flip side, can employers use AI to screen applications faster? 

Yes, you can attach CVs and cover letters to Claude or ChatGPT and ask which candidate is best, or to rank them. It's worth doing for an overview and to save time. 

But here's the critical caveat: training data bias means you cannot rely on AI alone. The AI might be skewed towards certain profiles or characteristics because of its training data. It will miss important context that a human would pick up on. 

Use AI as a time-saving aid to get an initial overview, but human oversight is essential for fair decision-making. You need to know the pitfalls and what to be aware of. 

Understanding AI's Limitations and Biases 

Q: What are the most important limitations people need to understand about AI? 

Lack of emotional intelligence: AI doesn't have reliable emotional intelligence. For example, ChatGPT was accused earlier this year of reinforcing toxic behaviour because it was optimised to be agreeable. People would describe abusive scenarios and ask if they were in the right, and ChatGPT would say, "Yes, that's understandable." Obviously not okay. 

Struggles with humour and creativity: There was a brilliant PR response when the Astronomer CEO was caught on camera at a Coldplay gig with the head of HR. They made a video with Gwyneth Paltrow (Chris Martin's ex-wife) to make a joke out of it. I tested this scenario with various AI tools-none could have conceived that solution. They suggested things like funny press releases, but nothing as creative or effective as what actually worked. 

Training data bias: AI inherits biases from its training data. This is especially critical when screening job applications or making hiring decisions. You cannot eliminate human judgement. 

The key principle: AI should augment human decision-making, not replace it. 

Building a Business Through Relationship and AI-Enabled Outreach 

Q: How do you find clients for Elect AI? 

I use a mix of approaches: 

• Cold email: Yes, with some AI assistance for efficiency 

• LinkedIn outreach: I've been experimenting with AI-enabled LinkedIn outreach for about a month 

• In-person networking: This is actually the most effective. I try to go to events and conferences to meet new people 

A friend who runs a campaign tech company has got all his clients over three years just from going to conferences and meeting people. I can completely see how that works. You just never know who you'll meet. 

My biggest client came from a friend tagging me on LinkedIn-someone I went to a Slipknot gig with who had been a director on a campaign I worked on. Very random, but that's how business development often works. 

Real-World Applications: From News Briefs to Legislative Amendments 

Q: Looking back at your work in Parliament, how would AI have changed those workflows? 

Dramatically. For example, when I did night shifts compiling news briefs-which took hours, from 6 PM to about 3 AM-I could now complete the entire thing in about half an hour using deep research tools. That's about 20% of the time. 

For legislative work, the difference is even more striking. During my time in Parliament, we spent about 2-3 months working out an amendment to a particular bill to achieve an affordable housing goal. After I left Parliament, I tested the same scenario with an AI tool that had been trained on law. I asked what would need to be changed in UK legislation to achieve this goal. It worked out exactly what we ended up doing after three months of discussion. It could have just been done instantly. 

These examples show how AI can handle complex research and analysis if you know how to use it properly. 

The Future of Elect AI 

Q: What's next for Elect AI? 

We'll continue finding new businesses to work with and evolving our offering based on what people need. Everything is already tailored to client needs. 

There might be more development and technical work in the future, but right now we're focused on helping people make the most out of available AI without needing to invest in ridiculously expensive custom models. If you don't know how to use AI yourself and don't train your employees properly, you're wasting your money. Whereas if you invest in training people on available AI-how to do it safely and effectively-you get noticeable, immediate results. 

We've trained 125 people so far, and that's going up to a couple of thousand over the next month or so. We're just pushing on and seeing where it goes. 

Key Takeaways: Actionable Insights for Using AI Effectively 

• Process transformation beats tool training: Don't just learn a tool-redesign your workflows around AI capabilities. Organisations must pause and properly engage, not just tick training boxes with generic video courses. 

  
  

• Different tools for different tasks: Claude excels at professional writing, Perplexity is best for research and language learning, and ChatGPT often produces formulaic outputs. Understanding tool selection is crucial for productivity. 

  
  

• Train AI to speak in your voice: The biggest mistake in job applications is using generic AI outputs. Take time to train the AI on your communication style to avoid embarrassing phrases like "seamlessly" and "delve" that flag AI-generated content.

• Quality over quantity in job search: Better to do 1-2 highly customised applications daily with AI assistance than 20 generic ones. Use tools like Monday.com, Sanebox, and Spike to track applications systematically. 

  
  

• Human oversight is non-negotiable: AI lacks emotional intelligence, struggles with humour and creativity, and inherits training data bias. Use it to augment decision-making, not replace human judgement-especially in hiring and sensitive contexts. 

  
  

• Security through proper configuration: Most security fears stem from misunderstandings. Enable safety features, configure settings correctly, and understand what AI companies actually do with your data before dismissing AI as unsafe. 

• Expect immediate ROI from proper training: Every person trained should save at least 1 hour daily immediately following training. Real results include reducing 3-week processes to 3 days and completing annual reports in 1-2 hours instead of months.