Devising an AI Strategy for Your Business
Companies everywhere are scrambling to decide how AI is going to impact their business. Here's how to approach it in a systematic way.
Pieter van Noordennen |
Devising an AI Strategy for Your Business
Companies everywhere are scrambling to decide how AI is going to impact their business. Will it disrupt them? Will it empower them? Should we be using AI in our product? In our operations? What if our employees already are and we don’t know about?
The power of tools like LLMs (ChatGPT), image generators (Midjourney), and voice interfaces (ElevenLabs) is profound. If you aren’t talking about how this is going to change your business, you are already behind.
But devising an AI strategy doesn’t mean you need to go back and get a Ph.D in discrete mathematics or become a Prompt Engineer yourself. AI is a tool, and like any tool, it requires strategy, thoughtfulness, and a clear understanding of customer needs to be used effectively.
If, like Kamala Harris, you find yourself in the position of being your company’s “AI Czar”, here is a simple framework for approaching an AI strategy.
Be Clear on the Problem You Are Trying to Solve
Wait, Pieter, I thought we were talking about emergent technology and neural networks and all that. This sounds like regular old business sense. But the truth is, the AI space is emerging, changing, and improving at a rate we’ve never seen before, and this old business maxim will keep you focused on outcomes as you enter the ups and downs of weekly disruption in AI.
Write a clear problem statement that describes the challenge your organization hopes to solve with AI. Better yet, have ChatGPT or Bard help by feeding it a bunch of information about your company and context, and ask it to write the problem statement.
Generally, I see problems break into two categories: Product and Operations.
Product Use Cases for AI
You might be a product or strategy leader tasked with figuring out how to get AI into your product. If that’s the case, the same tenets that drive good product strategy will drive AI strategy. What are customers looking for that is difficult for them to get today? Focus your efforts there, and then look to things that AI does well, at scale, for where you can begin to experiment.
AI tools and techniques get deep quickly, and its easy for teams to get lost in a sea of chains, agents, prompts, and vector stores, only to pick their head up weeks later and find there are new and better ways to solve the problem. De-risk that scenario by having a crisp problem to solve and stay myopically focused on that.
For example, one thing AI is great for is automating manual processes in your product. If you have a lot of proprietary data in text format — say you generate business insights for clients based on a set of analysis tools developed by your in-house subject matter experts — than leveraging summarization capabilities of LLMs makes a ton of sense. You’ll save customers time and energy in reading and note-taking, and they will love (and pay you) for it.
To ChatBot or Not to ChatBot
One of the big product strategy questions you will have is around ‘interface’ — namely, how will users interact with your AI-based content. Generating text or image content that lives statically in a database or on a blog is fairly straightforward and low risk. However, if you plan for a Q&A style chatbot interface, you’ll have to deal with as-yet unsolved problems such as Prompt Injection and AI security, not to mention cost controls.
That said, AI tools have opened a whole new way of looking at interfaces. Midjourney is primarily a Discord bot, which means that the distribution model is innately community orientated and that it has a built-in product-led growth motion. Be creative in the interface you choose to showcase your AI, and make sure the UI best suits your intended audience and growth model.
Operational Use Cases for AI
Maybe you aren’t planning to use AI in your product (or you already are), but your teams are yammering to make the leap into AI-enabled tools. AI can provide tremendous productivity boosts to your organization, and not doing so could put you at risk of being disrupted. But working AI into the daily operations of your business is no easy thing.
Here are some things to keep in mind when making “build or buy” decisions about AI tools.
Identify the Champions and Take It Step by Step
While the easy thing is to just let everyone self-organize and try whatever systems they want, your progress will likely go too slowly to be effective. You’ll wind up with a licensing nightmare, cobbled together systems, and proof-of-concept demos that can’t scale.
Like with product, identify areas of “busy work” where AI can help, and roll out new implementations systematically. Let’s say your marketing team is spending a lot of time creating content for LinkedIn posts and Tweets. Rather than trying to fully automate posts with Zapier and AutoGPT, start by simply buying them a ChatGPT license and showing them some best practices for prompting. They’ll go faster, but you’ll have a quality assurance stopgap in place to ensure nothing goes haywire.
Pick projects where you have knowledgeable, capable champions who are willing to share what they learned and bring others along the way. This last part is critical. At the end of the day, humans have to use the AI systems you deploy, and disseminating learning in your organization will give you the competitive edge you desire.
Treat It Like Any Other IT Spend
Define budgets per team and buy licenses just like you would any other software. At Datadog, they bought OpenAI licenses not just for the engineering teams, but for marketing and sales as well. Start budget planning for AI spend now, and expect that new systems, tools, and services will become available in the near future.
Also, look at your existing operational stacks and ask if they are implementing high-quality AI in a meaningful way. If not, consider switching to a competitor who already has AI built-in. We updated our internal documentation system from a legacy offering to Notion, whose Notion AI tool is dead simple to use and was one of the earliest adopters in this space. First mover advantages matter in AI.
Be Aware of the Current Limitations
There are still areas where current AI tools have some gotchas.
The main categories here are:
- Accuracy: Hallucinations are real in LLM models like ChatGPT. For all the impressive demos and wow-factors of this awesome technology, the tools will answer questions confidently whether they are sure of the answer or not. Thus, make smart choices as to where you implement tools in your organization, and carefully assess the cost of a false positive or a false negative.
- Security: AI tools, and especially LLMs, open the doors to a whole new world of potential security issues, and, worst of all, many of those security issues simply don’t have solutions today. If you are opening up your chat to the open internet, expect Prompt Injection attacks and similar. Closely monitoring any product integrations and API calls is the first line of defense, but be sure to include a strong security policy in any AI strategy.
- Data Privacy: Plugging confidential data into ChatGPT or similar online tools may mean that data is used to train the model later. Your employees will do it, just like they did with StackOverflow, Quora, and other internet outlets. Thankfully, there are ways to keep your data private, and a whole new industry of startups tackling this problem.
Developing an AI strategy will be critical to staying competitive for organizations in the days, weeks, and months to come. By sticking to your core business principles, enabling champions in your organization, and being realistic about the capabilities and limitations of these tools will help you navigating this next revolution in tech.
Want to discuss your AI strategy? Connect with me on LinkedIn: linkedin.com/in/psvann.