SONATAnotes

How AI Partners Can Make Consultants More Successful

When humans domesticated dogs 15,000 years ago, the partnership made them the most effective hunters on the planet. Between dogs’ speed and sense of smell and humans’ ability to use tools and traps, they were far more effective together than either species was alone.

Fast forward to today, and humans have a new companion: artificial intelligence. And for consultants and knowledge workers, AI brings a set of capabilities every bit as variable as canines’.  A study by Boston Consulting group (BCG) found that consultants who used generative AI completed 12.2% more work and delivered tasks 25.1% faster, with a 40% boost in quality compared to their peers working without AI assistance.

While some see artificial intelligence as a competitor, and fear that clients will start asking ChatGPT for advice and assistance rather than hiring consultants, those who use it regularly know AI won’t replace the best consultants anytime soon.  And just like a prehistoric hunter and their dog, an expert with an AI agent should be able to bring in more business than they ever could alone.

Of course, no generic AI model is going to be a productive “hunting companion” out of the box (and that’s a good thing for humans).  If you want an AI agent that knows your methods and matches your style then – like a dog breeder – you’re going to have to develop one.

Teaching AI Your Kung Fu

One challenge with generative AI is that it can accomplish so much with minimal instruction that people mistakenly assume that any task it can’t complete effortlessly must be beyond its scope.

For instance, if you ask ChatGPT, “What are the best fundraising strategies for a small nonprofit?” it will likely offer generic advice—donations, grants, crowdfunding—similar to what you’d find on a blog. However, unlike a skilled consultant, AI won’t probe deeper with discovery questions that reveal the root causes of any funding challenges facing an organization. Worse, it may take user inputs at face value, even though experienced consultants know that clients often “don’t know what they don’t know.”

But this doesn’t represent a “failure” of AI – if anything it’s an issue with lack of clear direction.  Just as hunting dogs need to be taught how to work with humans (and consulting firms must spend time onboarding fresh graduates), consultants must invest time teaching AI agents their methodologies.

Here’s an example of an AI nonprofit fundraising advisor we developed, which asks discovery questions, analyzes responses for inconsistencies, and suggests strategies only after gathering relevant context. 

Achieving this took more than a simple two-sentence or two-paragraph prompt: behind the scenes, the agent operates on over 55,000 words of instructions and criteria for which fundraising strategies fit a given organization.  

For example:

Developing detailed guidance for AI may seem daunting, but this is actually good news. If an AI could replicate a human expert’s advice with just a few sentences of instruction, that would be bad news for humanity.   Rather, crafting effective AI agents for consulting work requires a collaborative effort between consultants and AI developers (“prompt engineers”) to distill the essence of your methods and perspective into a structure AI agents can follow. 

The Right Robot for the Job

The success of an AI agent depends not only on the quality of your instructions but also on the strengths and weaknesses of the underlying AI model. Just as humans bred dogs for specific purposes—basset hounds for scent-tracking, salukis for speed— consultants can build agents with different AI models for different tasks.

While the actual technical differences between AI models can get esoteric, here’s a simplified comparison, from a real-world application perspective:

  • ChatGPT (GPT-4o): Excels at following detailed instructions and making judgment calls but its actual writing can sometimes come across as overly formal and stiff.
  • Claude (Sonnet 3.5): Can write stories you might actually enjoy reading, but can be overly verbose when all you want is a quick summary, and frustratingly hesitant to make decisions.  
  • Gemini Pro 1.5: is great for code generation and writing about medical subject matter in particular, though less versatile for general problem-solving.
  • Llama 3.1 (405b) does a great job of capturing human conversational tone well but can struggle with complex reasoning.

Fortunately, building effective AI agents doesn’t mean sticking to one model. Tools like Perplexity integrate AI with search engines, providing up-to-date information and citations—something most language models struggle with. And my own company’s platform, Parrotbox.ai, goes a step further by enabling AI agents to switch between models mid-conversation and access external tools like calculators and slideshows.

For example, our “AI Teacher” uses different models and tools depending on the task: one configuration for explanations, another for role-play exercises, and a third for facilitating discussions – while presenting it all as a single, continuous conversation with the user:  

Stronger, Faster, Better: Adapting Your Approach for AI

So, in practical terms, how can AI agents multiply your capabilities as a consultant?  Let’s look at a few ways:

  • Scale: If you run a small consulting firm that serves large organizations, you probably aren’t reaching 100% of the people within client organizations who could benefit from your expertise.  Maybe the CHRO at a bank loves your communication skills workshops, but it’s physically impossible for you to deliver it to every local branch – even via Zoom.  Building an AI agent that can deliver workshops and provide coaching grounded in your methodology can expand your reach without having to hire and train up other consultants.
  • Availability: If you’ve been in consulting long enough you probably have clients who can’t make progress without you there to hold their hand. Perhaps you walked a small business owner through your marketing methodology during a paid session, only to have them call your direct line every other day for the next two weeks with questions.  Providing an AI agent that can not just answer those questions (like a garden variety chatbot) but actually walk people through your process can help clients feel supported without helping themselves to your time. 
  • Efficiency: There are probably specific tasks in your work that don’t require an expert, but are too nuanced or complex to hand off to some random human assistant. Designing an AI agent capable of summarizing documents like historical incident reports or feedback surveys according to your preferred framework can save time, reduce costs, and let you focus on the higher value aspects of your work that you actually enjoy.
  • Support: It can be lonely being the expert in the room, and sometimes you just need someone to be a sounding board for your ideas and provide some support when engagements get challenging.  However, your human friends and colleagues have their own problems and can only listen so long.  Creating an AI “digital twin” grounded in your methodology and style can give you an always-available brainstorming buddy, devil’s advocate, and sounding board. Sometimes, talking through a challenge with an impartial agent can help you spot gaps in your thinking and reach a decision faster than you would on your own.

Humans + AI: The Undisputed Champs

The future of consulting won’t be about humans versus AI. It will be about which consultants make the best use of AI. As NYU business professor Scott Galloway puts it: “AI won’t take your job, but someone who knows how to use AI will.”

As of this writing, only 5-7% of U.S. companies outside consulting have formally adopted generative AI into their work process, this will likely change rapidly.  Like the adoption of websites and desktop computers in the past, organizations will likely go from not understanding AI to demanding AI practically overnight. In my own field – learning and development – we’re seeing more and more clients insist that consultants use AI to create training courses, as they don’t want to pay for work done at a slower, unassisted human pace.

But the good news is that, no matter how capable AI becomes, if you give two consultants the same AI agent, one will produce better results than the other. Success will ultimately depend on who can A) build better AI agents, B) use AI agents more effectively, or C) both.

Fifteen thousand years ago, it must have taken courage to approach a pack of wolves and try to make friends. But once you learn to work with AI agents, living without them may became unimaginable. 

Hopefully you found this article helpful.  If you are interested in creating AI agents for training, coaching, or on-the-job performance support, please consider reaching out to Sonata Learning for a consultation.