As a business owner who has to set the course of your business for success, you need a valuable skill— you need to be a good business problem identifier.
Afterwards you can interface your business with AI technology to solve the problem you have identified.
Take these 15 steps:
- Ask: “What is the thing I am trying to solve? What is my business suffering from?” Then ask “Why?” (Every great innovation started with that question.) Keep on asking why till you get the clearest picture of the problem.
- Find a high performing LLM.
- Now formulate the problem as a prompt for your model, in clear, precise, logical terms, with clear, measurable parameters and outcomes. (Good prompt writing/engineering is becoming a core job skill.)
- Ask yourself: “Which core data would I need for this specific use case?”
- Then get that data ready for the agent. Be tactical about what data you need. (If you feed fragmented customer and product data into the model, it cannot optimise.)
- Choose the agent that has been trained for sales, for example, then fine tune it off of your specific company’s info.
- Then add more context to it— your knowledge on the info (or your team’s)
- Then fine tune it off of your data for your individual business needs.
- Build an evaluation/benchmark against that, to be able to gauge how it is performing.
- Tailoring the LLM to your context: the plan is to build a foundation model— at least as far as pre-training and post-training are concerned— for a specific industry on proprietary data sets.
- What does post-training look like? It is supersized fine tuning of a pre-trained generalist, off the shelf model, plus post-training off internal benchmarks in order to pass evaluation.
- You can sanction an AI agent for imitation learning, by shadowing skilled people on the job.
- Then: continuous monitoring, evaluating, updating, double checking to make sure it stays in line with your goals. (Beware: No rubber stamping as it is able to lie to you.)
- Validate that the output of the project matches your business's requirements as in step 1.
- You will always need human feedback on almost every different agent you want to roll out.
Shadowing: Onboarding AI shadow workers on the job during training
When implemented intentionally by the company, official AI agents shadowing workers can be a powerful tool, as this enables the AI to understand how to handle complex, multi-step tasks by watching experienced employees. They often use non-invasive monitoring tools to observe, record and analyse workflows and decision-making patterns. Workers validate the AI's actions and provide feedback that helps the agent to refine its skills.
In conclusion; a word to the wise.
- Create clear guidelines across the company for safe, approved AI usage.
- Offer sanctioned, enterprise-grade AI tools that meet employee needs.
- Data leakages happen when employees paste proprietary code, financial forecasts, or customer data into public, free AI tools, potentially exposing company/trade secrets.