Building AI processes that actually improve your company

Do you want to understand how AI can help your company improve? This worksheet may be for you.

It provides a structured way to identify one or two company processes you could improve with AI. If you can identify these processes, the worksheet will also give you some tips on where to begin.

Step 1: Write down some existing process candidates

Everyone is talking about AI these days, and it is clearly a powerful technology. Computers can now perform tasks once considered uniquely human such as writing text, but not all of those tasks should be replaced by AI.

To narrow the list, use your intuition and write down tasks that you think could benefit from AI. Then consider whether each one is part of an already optimized process.

To be clear: A process is a list of tasks that should have a clear purpose and end (a definiton of when it is done). Every business has some of those.

An optimized process is clearly documented and has been reviewed for inefficiencies.

Task Team or people involved Is the task part of an optimized process?

Respond to customer feedback on Google Customer Reviews Customer success team yes

Scanning all documents on client intake Client prep team no

Alignment on work assignments Technicians / customers / customer success yes

Outreach to customer prospects Marketing yes

Step 2: Optimize the processes

If you have any tasks that are part of unoptimized processes, now is the time to optimize the process first.

Because if a task belongs to an unoptimized process, it might not even be necessary at all.

In the example above, it might be unnecessary to scan all documents on intake. If so, there is no need to invest in automating that task with AI.

This may eliminate some candidates already.

Step 3: Assess the human connection involved in the tasks

Generative AI can produce output very similar to what a human can create, but AI systems are not very good at creating emotional connections. Businesses depend on these connections. So we need to pay attention to that.

Humans will always matter, and reports claiming that all of our jobs are being replaced by AI are clearly an overstatement. That is encouranging, but it also means we can fall into the trap of replacing tasks with AI where humans currently do better.

The following questions can help you avoid that trap:

Interview the relevant teams and people and ask three questions for each task:

Do you experience joy doing the task?
Have you experienced discomfort doing the task, such as annoyance or anger? That often happens when the task involves challenging interactions.
Have you learned something important when a task became uncomfortable?

These questions can help you narrow the list of possible tasks even further.

Do not remove human creativity

If a task creates joy, it is likely a place where your employees are creative and doing their best work. That is a sign you have hired the right people and created an environment where they can flourish. Honestly: congratulations!

In this case, you probably do not want to automate away that creativity.

Think of an employee who responds to customer feedback and enjoys the work. Yes, you could replace it with an LLM, but customers will notice. They may not be happy, and you might reduce costs while also reducing quality.

Supercharge human creativity

On the other hand, AI can extend human creativity.

To stay with the example of responding to customer feedback:

What if the customer success team had enough resources to handle Google reviews but not other platforms such as Yelp? AI automation could gather information from several platforms, pre-sort it, and assign the person the most urgent or important responses.

Be careful removing meaningful interactions or learning experiences

If your employees experience discomfort such as annoyance, anger, boredom, or overwhelm, the task is more likely a candidate for automation.

But beware of outsourcing a task to a computer if that removes meaningful interactions that previously led to important learning experiences.

Let’s assume you have a company that builds custom cabinets for customers. At some point, the technician needs to turn the customer’s instructions into a plan for building the cabinet. This requires careful alignment on the requirements.

Over the years, the technicians have learned about common confusions and mistakes that they should correct or double-check with customers. This interaction creates a delay and can be annoying for both sides, but it also provides valuable lessons and builds trust over time.

This is an example where fully replacing customer interaction with an AI bot might decrease quality and trust over time. A better approach might be to add an earlier feedback step that surfaces past lessons before the technician starts the work, which can streamline the process and improve satisfaction.

Step 4: Think of new opportunities

After going through the first three steps, you might have an empty list of potential tasks.

But you may now have a better understanding of what AI can and cannot do. With that understanding, you can start thinking about tasks you may have avoided in the past because they were too expensive or too difficult to tackle. You may have avoided them because they required expensive software, internal education, or hiring new specialists.

Tasks with unclear long-term benefits

Many business owners and their team members have ideas for making the business better, but they often avoid these options because hiring an expert, researching, purchasing, and training staff on new software is expensive, especially when the long-term benefits are uncertain.

With AI, the cost to get started drops significantly. And with that also the cost to estimate the long-term benefits.

Large language models, such as ChatGPT or Google Gemini, are useful for sketching out plans and reducing initial research costs.

To try a new idea, you often do not need to hire an expert to get started.

I started creating small comic strips with ChatGPT. I am not a graphic designer by any means, but I like explaining ideas with humor, and I experimented with it on my website. I am not particularly excited about the art, because it is often too generic or literal and sometimes does not support the comic’s punch line as well as I wish it would, but it is a start. And if I ever hire a professional, I will have a better understanding of what to look for.

More people can program now, with some limitations. Some tasks that used to require specialists with deep knowledge of certain programming languages, no longer do. For certain tasks, you can prompt LLMs to create a script or even an entire app that builds custom reports from databases. If the task is simple enough, that may be all you need. This is a bigger mindset shift, but not a far-fetched one for some companies.

Step 5: Prompting the right questions and finding the right tools

Lastly, if you have identified one or two tasks that can benefit from AI, make sure you understand the desired outcome. What do you want to get out of it?

Task Desired Outcome

Respond to customer feedback The team can respond to feedback across several platforms easily, and high-severity feedback is prioritized to reduce response time.

Alignment on work assignments A system gives customers real-time feedback about common problems and allows technicians to update the list of known problems.

Gain insights from online user behavior A custom dashboard that combines several databases and is easy to extend.

With this list, you can prompt any LLM for a plan. I often get asked which tool is best or what a good prompt looks like. The truth is that both are task-dependent and will change over time.

A good prompt starts by asking for a plan. If you know the desired outcome, tell an LLM such as ChatGPT to “Make a plan to achieve [insert desired outcome]. What context do you need to know to make good decisions? What risks and costs do I need to consider when executing the plan?”

If you are unsure which LLM is best suited for the task at hand, try several of them and see which one gives the most convincing output.

If you want support

Often this worksheet is enough to get started.

But human connections are valuable. They can help you filter and find potential tasks, and they can help you get to one or two outcomes worth pursuing. For this kind of service, I charge an hourly rate.

Once you have found the process, I can also help execute the plan. I offer the following options:

Retainer: You or the team do the work, and I am available to mentor and assist if you need starting points or get stuck.
Full service: I execute the plan for a fixed fee. This works well if you want proof of what is possible with AI.
Full service plus mentoring: I execute the plan and hand over the project so the team can modify the product or build a similar one by themselves.