4 Step Framework for Advancing Your AI Usage

In this article I walk through some of the common "Levels" we see users at when using LLMs. By investing in tools that makes it intuitive for users to level up and automate more complex and time-consuming workflows, e.g. an RFP response, you allow your team to focus on higher impact work. Teams that don't run the risk of falling behind.


Artificial Intelligence offers businesses the potential to reallocate resources from routine tasks to high-impact ones.

In this article I walk through some of the common ways we see users interact with LLMs. Usually, the users we speak to are scraping the surface of what they can do with LLMs. They might use them for simple tasks like rewriting or search because they're fairly simple to do - but the more complex tasks and workflows require mode advanced knowledge and tooling.

By investing in tools that makes it intuitive for users to automate the more complex and time-consuming workflows, e.g. an RFP response, you allow your team to focus on higher impact work. Otherwise, teams run the risk of falling behind.

Advancing Across These Levels Helps Address Issues Like:

  • Low quality outputs and hallucination from LLMs
  • Lagging user adoption of AI tools
  • Tool bloat

At DealPage, we help teams move from Level 1 to Level 4+ on this framework with a simple-to-use AI Agent that can complete a variety of time-consuming, low-impact Sales Engineering tasks.

Four Progressive Levels of AI Usage

To keep this more neutral and general, consider the really common example (and non-DealPage use case) of writing cold emails—a task that requires product knowledge, understanding of the recipient, and some contextual awareness.

Level 1: Basic Prompting

  • At this first level, AI interaction is straightforward. A user might ask the AI to generate cold emails for a specific persona
  • Benefits include speed and user-familiarity with the chat interface, making it suitable for quick, low-stakes tasks.
  • However, the main drawback is the lack of context, which often results in generic responses that require further manual refinement. Responses in Level 1 rely on a model’s training and imperfect recall, sometimes leading to hallucinations.

Level 2: Manual Prompting with Context

  • More advanced users enhance AI performance by manually adding relevant context to the prompts. This might involve including details like a recipient’s LinkedIn bio or specific product features.
  • This method improves the relevance and quality of AI-generated content by providing more detailed instructions, which helps in achieving more accurate outputs.
  • The trade-off is the increased manual effort required to input detailed context, making the process potentially tedious and repetitive.

Level 2.5: Context-Enhanced System Prompts

  • At this mid-point level, prompts are constructed with predefined system settings that incorporate user preferences and product information, which are stored and reused.
  • This reduces the burden of repeatedly entering detailed contexts and streamlines the interaction with the AI.
  • A challenge here is maintaining user awareness of the underlying settings, which can lead to confusion or unexpected results if users are not consistently reminded of the context being applied.

Level 3: Simple Retrieval-Augmented Generation (RAG)

  • AI tools automatically compile detailed prompts using external data sources, such as real-time data from LinkedIn, without user intervention beyond the initial setup.
  • The quality of AI interactions is significantly enhanced, combining the ease of basic prompts with the depth of context-driven responses.
  • Simple RAG tools are specialized for individual tasks - such as question-answering and email writing. This leads to potential tool-bloat for teams that want to automate a variety of tasks.
  • Note: Retrieval results are sometimes added to the chat history, but this implementation works too.

Level 4: Dynamic Data Agents

  • The most advanced level involves AI systems known as Agents that dynamically gather and integrate data from a variety of sources as needed, such as LinkedIn profiles, internal databases, and online news. Agents often involve a “planning” phase where models execute judgement on what tools to use to complete a given task.
  • These agents enable highly contextualized and relevant interactions across a wide range of business tasks, adapting prompts in real-time based on the latest available data.
  • While offering the highest potential for automation and accuracy, this level also presents significant challenges in terms of development, maintenance, and reliability due to the complexity of the systems involved.
  • Note: Plans and tool calling results are often added to the chat history, but this implementation works too.


At the end of the day, the teams with more capable AI users and AI tools will pull ahead due to more efficient time and capital allocation. We encourage teams to invest in AI platforms, not spot solutions, that connect important datasources to AI Agents which can then complete a variety of workflows.

DealPage helps sales teams connect their sales knowledge, technical knowledge, and CRM data to an AI Sales Engineer named Paige. Paige helps with a variety of workflows including RFP responses, security questionnaires, meeting follow ups, and account handoffs. All from natural language prompts and intuitive UIs that don't require tons of training.