TrueML Talks #28 - 営業アウトリーチのためのGenAIとLLM @ OneShot

We are back with another episode of  True ML Talks. In this, we again dive deep into GenAI and LLMs for Sales Outreach at OneShot and we are speaking with Peda Venki Pola

Venki is the founder and CTO at OneShot and before this, he worked at Salesforce as a Software Architect. OneShot helps B2B SaaS companies to generate the top of the funnel for their business.

📌

Our conversations with Venki will cover below aspects:
- AI-Powered Sales Outreach
- Unique pricing model at OneShot
- Tech Stack at OneShot
- How OneShot Manages LLMs with Prompts
- Full-Stack Goes AI: How LLMs are Redefining Development
- Futureproof Your Business: Leveraging LLMs for Success

Watch the full episode below:

AI-Powered Sales Outreach

OneShot's AI-Powerered Sales Outreach starts with your Ideal Customer Profile (ICP). They don't just take your word for it; they analyze your existing customer data to identify patterns and suggest the perfect target audience.

Once the ICP is defined, OneShot's AI goes into overdrive. They leverage their models and open-source tools like Hugging Face's Hub to scour the web, scraping data from LinkedIn profiles, company websites, and even financial reports. This treasure trove of information is then summarized and analyzed by powerful LLMs like GPT-4 and Claude, revealing insights about companies and individuals that traditional sales tools simply can't match.

But data is just the foundation. OneShot uses its understanding of your business and the insights gleaned from prospect research to generate personalized outreach messages that resonate. No more generic greetings and impersonal pitches – each message is tailored to the specific needs and interests of the recipient.

OneShot doesn't stop at email. They can automate outreach across multiple channels, including LinkedIn messages and even call scripts, ensuring your message reaches your prospects wherever they are.

OneShot's AI is constantly learning and evolving. They track the performance of their outreach campaigns, using reinforcement learning to identify what works best and adapt their strategies accordingly. This ensures that your outreach efforts are always optimized for maximum impact.

Two Sides of the Coin

AI-powered outreach has two dimensions:

• Sender: Businesses using AI to send messages face concerns about transparency and approval. OneShot offers options like co-pilot and autopilot modes, allowing users to review and personalize AI-generated content before sending.
• Receiver: While some may find AI-generated messages impersonal, many appreciate the research and personalization they offer. It's a shift from generic greetings to targeted content that sparks genuine interest.

Humanizing the Bots

Generic AI solutions won't work long-term. Businesses need to:

• Customize Language: Adapt the AI's tone and style to your brand voice and audience. Imagine tailoring messages for CTOs versus individual contributors.
• Leverage Business Data: Train AI models on your specific data to generate messages that resonate with your ideal customers.

A Future of Collaborative AI

AI isn't replacing humans, it's augmenting them. Imagine:

• AI handling the legwork: Researching prospects, crafting initial messages, and automating repetitive tasks.
• Humans adding the personal touch: Reviewing and refining AI output, building relationships, and closing deals

‍

How to Transform 3 Common Intent Signals Into Opportunities with AI

Revealing intent is one thing. Actioning it is another. In this post we’ll explain how to use take three common intent signals and use them to create hyper relevant and personalized outreach.

Gautam Rishi2023-12-08

‍

Unique pricing model @ OneShot

People, Not Sales: A Different Approach to Pricing

Unlike traditional sales software, OneShot doesn't charge based on closed deals. They focus on what they can directly control: the number of prospects you reach. Why this approach?

• Focus on the Journey: Reaching the right people is crucial for sales success, even if not every interaction leads to a deal.
• Transparency and Fairness: Businesses only pay for outreach efforts, not for unpredictable outcomes like closing rates.

Beyond Just Numbers: Tailoring Value

It's not just about quantity; OneShot considers the quality of outreach as well. Their pricing reflects:

• Research Depth: Do you need basic info or an in-depth analysis of each prospect?
• AI-Powered Messages: How many personalized messages will you send per prospect?
• Multi-Channel Reach: Email, LinkedIn, calls – choose the channels that fit your strategy.

A Model for Mutual Success

This pricing structure benefits both OneShot and its customers:

• OneShot: They're incentivized to deliver high-quality leads and engagement.
• Customers: They only pay for the outreach they need and see the value in each potential connection.

Tech Stack @ OneShot

From Langchain to Building Their Own

OneShot started with Langchain, a popular AI/ML toolkit. While it offered a good foundation, it lacked the flexibility needed for OneShot's specific needs. So, they're transitioning back to a custom-built solution that allows them to create things like a "gateway" for connecting to multiple AI models.

The Quest for Perfect Embeddings

OneShot searches through a massive database of 40 million companies to find the perfect fit for each business. To achieve this, they use various tools:

• GPT Embeddings: These capture the meaning of text and help find relevant companies based on keywords and descriptions.
• Pinecone & Other Vector Databases: These store the embeddings efficiently and enable fast searches.
• ChatGPT Models: These analyze the user's query and identify the most relevant knowledge articles.
• Switching LLMs on the Fly: Depending on the task, OneShot can use different models like GPT-4 or Claude.

Leveraging Open Source

OneShot doesn't go it alone. They leverage open-source tools like yours for fine-tuning models and hosting solutions, ensuring efficiency and access to the latest advancements.

Challenges and Lessons Learned

Building an AI platform isn't without its hurdles. OneShot faces challenges like:

• Balancing Flexibility with User-Friendliness: Catering to both tech-savvy and non-technical users requires careful design choices.
• Keeping Up with the Evolving LLM Landscape: New models and tools emerge constantly, requiring adaptation and exploration.
• Monitoring and Maintaining the AI Engine: Ensuring smooth performance and reliability is crucial for user trust.

The Future of OneShot:

OneShot currently relies on its API to connect to AI models, but they're exploring new horizons. As they scale, they're considering integrating with platforms like Azure OpenAI to access a wider range of models.

However, OneShot is constantly innovating. They're exploring possibilities like:

• Bringing Your Own LLM: Allowing users to choose their preferred models for even more customization.
• MLOps Integration: Leveraging platforms for robust monitoring and management of their AI infrastructure.

By embracing flexibility, tackling challenges, and staying ahead of the curve, OneShot is building a powerful and accessible AI platform that empowers businesses to achieve success.

How OneShot Manages LLMs with Prompts

Think of prompts as instructions or questions that help the LLM understand what information you're looking for.

OneShot doesn't use a one-size-fits-all approach. They provide different prompts for different uses:

• Lead Qualification: 連絡すべき相手はこれで合っていますか？プロンプトはAIがデータを分析し、明確な「はい」または「いいえ」の回答を出すのに役立ちます。
• 情報抽出： 見込み客のビジネスに関する主要なポイントを要約します。プロンプトはAIが関連情報を見つけ、簡潔に提示するよう導きます。
• 営業コンテンツ作成： パーソナライズされたメール、LinkedInメッセージ、コールスクリプトを作成します。プロンプトはAIが特定の見込み客と希望するトーンに合わせてコンテンツを調整するのに役立ちます。

OneShotは、ユーザーがプロンプトを自由に制御できるようにします。プロンプトを実行するモデルを選択し、「クリエイティブ」や「決定論的」といった設定を調整して、結果を微調整できます。

フルスタックがAI化：LLMがいかに開発を再定義しているか

「フルスタック開発者」がフロントエンドとバックエンドの両方をこなすことを意味していた時代を覚えていますか？しかし、今やAIがその一員となり、状況は一変しました！LLMは開発のあり方を完全に変えつつあります。

1. AIが新たな「フルスタック」： もはやコードだけではありません。開発者は今、AI機能、埋め込み、そして MLOpsプラットフォーム。パイプライン全体にAIが組み込まれています！

2. AIのコモディティ化： ユーザーフレンドリーなプラットフォームのおかげで、AIモデルの構築と利用がこれまでになく簡単になりました。開発者は今、微調整とカスタマイズに集中できます。

3. チャットボットが新たなコパイロットに： 無限のGoogle検索の時代は終わりました。エンジニアは今、ChatGPTのようなAIアシスタントを活用して、より良いコードを書き、生産性を向上させることができます。

‍

‍How AI SDRs Will Impact Sales Development in 2024

Will 2024 be the year of the AI SDR? While the sales development role may seem like the first in line to be automated, the reality is a little more nuanced. Read on to learn what the rise of AI in sales means for SDRs.

Dan Moran - 2023-12-05

‍

ビジネスを未来志向に：成功のためのLLM活用術

AIの世界は急速に進化しており、大規模言語モデル（LLM）はその変化の最前線にいます。しかし、これほど多くのイノベーションが起こっている中で、どこから始め、ビジネスを未来に向けてどのように準備すればよいかを知るのは難しいかもしれません。

LLMはもはや未来の概念ではありません。現実に存在し、利用可能であり、さまざまな事業部門で真の価値を提供しています。営業やマーケティングにおける従業員の生産性向上であれ、顧客向けのパーソナライズされた体験の創出であれ、LLMをワークフローに統合することで、大きな競争優位性をもたらすことができます。

しかし、単に一般的なLLMソリューションを導入するだけでは、その真の可能性を解き放つには、 カスタマイズ する必要があります。これには、モデルのファインチューニング、人間の専門知識の組み込み、継続的な改善のための効率的なMLOpsプラットフォームの活用が含まれます。LLMはツールであり、それを適用するあなたの専門知識こそが、他との差別化を図るものです。

読者への実践的なアドバイス：

• 小さく始めて実験する： 一度にすべてをやろうとしないでください。LLMが明確な価値を提供できる特定のビジネス課題を選び、さまざまなモデルやアプリケーションで実験を始めてください。
• カスタマイズに注力する： 覚えておいてください、万能なソリューションはめったに機能しません。独自のニーズを理解するために時間を投資し、特定の状況に合わせてLLMソリューションを調整してください。
• 継続的な学習の文化を築く： LLMの状況は常に進化しています。最新情報を入手し、新しい開発を探求し、このダイナミックな分野でチームが生涯学習を受け入れるよう奨励してください。

True ML Talksシリーズの以前のブログを読む：

‍

‍GenAI and LLMOps for Customer Success @ Level AI

Learn how Level AI leverage GenAI and LLMOps to solve Customer Success and Quality Assurance from Abhimanyu, Staff AI Research Engineer @ Level AI

TrueFoundry Blog TrueFoundry

‍

TrueMLの YouTubeシリーズ を視聴し、TrueMLの ブログシリーズを読み続けてください。

TrueFoundry は、Kubernetes上で動作するMLデプロイメントPaaSであり、開発者のワークフローを加速させるとともに、モデルのテストとデプロイにおいて完全な柔軟性を提供し、インフラチームには完全なセキュリティと制御を保証します。当社のプラットフォームを通じて、機械学習チームは デプロイおよび監視 モデルを15分で、100%の信頼性、スケーラビリティ、そして数秒でのロールバック機能を備えてデプロイ・監視できます。これにより、コストを削減し、モデルをより迅速に本番環境にリリースできるようになり、真のビジネス価値の実現を可能にします。