There’s maybe an AI for that

More lessons from the Post Office

Before 1960, postal employees cancelled stamps strictly by hand. When commercial mail volumes exploded, the post office introduced automated cancelling. It was poorly received – replacing people and that sort of thing.

Recently the post office announced it’s changing the date-stamping part of its automation. It now makes more sense to the USPS to hand-stamp the few pieces that still need today’s date recorded on the mailpiece. Automation routines are not permanent – it only makes sense when the volume justifies the complexity.

The AI Gold Rush

You may have looked around your place of business and thought, ‘there must be waste in the routine and manual.’ Three years ago ChatGPT arrived with an influx of automating-promising AI tools. Because of the endlessness and variety of internet promises, it’s easy to not think strategically about what AI can realistically do for your organization. The internet can feel like flashing lights on a slot machine – not exactly a rational appeal so you can weigh your odds with gravity.

But if you and your staff identify areas that feel like they might be automated for a specific gain, that’s a good starting point to see what AI tools exist that might improve things. Approach AI with strategic goals in mind.

Where AI shines

Look for predictable patterns that recur regularly. Start with tasks like scheduling, ticketing, and converting data from lead generation systems to CRMs. The main idea is to move your people off routine tasks into higher-level projects that serve your organization’s goals.

Even though AIs can present solutions quickly, adding those solutions to your workflow may take time and modifications along the way. If someone in your office is vibe coding, you want to make sure of the soundness of the code and that permissions are properly handled before introducing new code into your systems.

What AI doesn’t do so well

I like checking the There’s an AI for That site that shows off many AI tools. But when you drill down, you see they’re often very specific. I’ve tried AI image upscaling tools – the cheap Chinese VanceAI and the higher-end alternatives. They are smoother than Photoshop’s bicubic guessing, which is not always desirable – it can look weird. And enlarging images with text? Almost nothing does it well. I’ve had blueprints that were unreadable, hoping an AI tool could ferret out meaning from tiny type. Nothing useful, nothing like the movies where they zoom into a license plate and suddenly it’s clear.

Chatbots are best thought-of like an obsequious Google search that organizes words by how likely they are to occur in order in its database. Ask a chatbot ‘how should I accomplish X?’ It will give a near-instantaneous, confident answer that – if you scrutinize the magically appearing words – often has elements that feel untrue or are redundant. Push back and the chatbot will likely say, ‘you’re right, I should have accounted for Y.’

The Customer Service solution?

Last year my company used an AI-supplemented chat agent on its website. People would ask questions based on our site and it would give answers that had nothing to do with our content. Capturing leads from a website feels like an ideal use-case for AI, but the options are quite specific – medical appointment-making, haircuts, IT companies.

I saw one supposedly for IT companies where the caller mentioned a possible breach. The AI took this in stride like the guy was asking to compare IT services and continued asking its programmed questions. It ended with a cheery, non-committal ‘this message will be channeled to the proper department. Goodbye!’ When people have a possible breach, they are not patient. They want actual understanding.

When things are copacetic, everyone wants the shiny new machine. When things take a serious turn for the worse, everyone wants actual understanding.

Questions Before Deploying AI

• How much time will be saved versus a person doing the task?
• What will the person be doing that is of higher strategic value?
• Does this task occur frequently? How many times per week?
• Are there variances within the pattern that if not understood might alienate a prospect or customer?
• Can you estimate the damage caused by a poor interaction versus the automation time-savings?

AI is a long way from being the understanding part of the sales process. And understanding people is the foundation of sales. Like the post office learned, automation should be sensitive to volume and predictability to justify the complexity. The trick is knowing the difference.

---