The Right and Wrong Ways to Integrate AI in Business

AI integration agencies continue to pop up left and right. While there's a legitimate business opportunity in AI technology, we can't ignore the hype - or the risks.

Unfortunately many such agencies integrate AI in a dangerous way that can be disastrous to the business.

While AI offers tangible benefits, and business that adapt it in their workflow will see the competitive advantage, we cannot overlook the dangers.

Let's talk about where the "dragons" lurk and how to avoid them.

Risk #1: Unsupervised "agent"

The most common AI integration I see today involves agencies adding a custom-trained chatbot to their client's website. This seemingly harmless practice can backfire when chatbots provide customers with incorrect information, make things up, or promote legally binding offers that could harm the business.

Then we have the subset of jokers who love getting website chatbots to write poetry, Python code, and make off-hand commentary on current events. These interactions quickly end up on Reddit, damaging the business's reputation - making the company a public laughing stock.

Many AI-Integration-Agency (AAA) step it up the notch: they give AI customer support access to user's own data or worse, entire production database.

Anyone with common sense understands how dangerous this is, yet few business owners are aware of such integration.

Recently, I spoke with a support bot for the ride-sharing app Bolt. When I complained about my past order, the chatbot automatically credited my account. No human participated in that decision - the chatbot had direct access to my account or authority to issue credits.

What else did it have access to?

A poorly integrated customer support AI agent can:

• Leak sensitive data
• Expose internal business logic
• Violate AI regulations in your jurisdiction

Risk #2: The Content Slop Machine

So AI writes some articles for a company blog - what's the big deal?

It's not the articles themselves that cause problems, but how these shallow content pieces affect the search engine rankings (SEO helps websites appear in Google search results).

Many business owners I talk to simply don't have time to create content. They want to click a button and publish engaging blog posts (Don't we all?)

I fell for this hype myself and quickly realized that AI content doesn't provide the insights visitors (potential clients) seek.

But what's the real cost?

Lost opportunities - even if the article ranks well, readers come and leave. Readers that will never engage with your company or become a customer.

We could debate creativity and originality all day, but most business owners don't have advanced AI prompting skills and struggle to get AI to write what they want.

Many are discovering ChatGPT for the first time, wrestling with prompts instead of writing. This creates a false sense of productivity for them.

It is a skill issue.

Congratulations - you now have "100 AI articles" nobody wants to read. Instead of spending hours clowning with prompts, you could have written 2-3 excellent articles.

AI-generated content struggles with:

• Brand voice consistency
• Accuracy (it often makes things up)
• Depth of expertise

Readers and potential customers want human elements:

• Experience-based insights
• Industry-specific expertise
• Emotional intelligence and context

Risk #3: Math Doesn't Math

Less common AI integration request is accounting and reporting.

The math.

The numbers.

To most people that understand AI, this just sounds ridiculous, but not to a non-technical manager or busy CEO.

First of all, the models we currently use are called Large Language Models (LLMs)…

Yes, "language", not "math". Using them to add, subtract, multiply and divide numbers is just not the use case people think it is.

LLM predicts the most likely next word (or number) based on the data set it was trained on.

Model has been trained and "saw" 2+2=4 a billion times - it gets very basic math correct. But it does not mean it will recognize the numbers and math formulas in your Excel sheet of broader complexity.

To summarize:

• accuracy is going to be crap
• in complex accounting it will use wrong formulas
• it can get you into legal trouble

Do not.

Where AI can actually help?

Let's revisit the same three needs: Customer Support, Content Creation and Reporting & Analytics

Internal Support Assistant

Picture this scenario:

Your customer support rep receives an email about a broken feature or complaint. An internal AI assistant (note: internal, not customer-facing) analyzes the email, finds relevant documentation, checks the customer's CRM data, and drafts a response for the human agent to review and send.

Another example:

Your sales rep answers a phone from a potential customer. The AI transcribes the conversation and identifies customer needs. As salesperson looks at his screen, the AI Assistant has already typed questions he should ask the lead on the phone. As potential lead raises concerns and consideration, AI finds a Use Case or Case Studies for this customer.

Once lead mention his company, the AI already accessed key datapoint about the company, any publicly available information or any important facts that would help the sales person on the phone.

They already have the relevant information at the point of conversation.

The case:

• Internal customer support assistance
• Analysis of customers' complaint
• Sales assistant
• Lead research

Writer's Assistant

Remember the content problems we discussed? Here's how AI should help by providing the following:

• Suggesting article topics
• Creating outlines
• Generating relevant questions for authors to answer with their expertise
• Checking for grammatical errors, awkward phrasing, and bias
• Polishing specific segments

Furthermore, a web based AI tools can research topics and provide references (for example, Perplexity, web browsing with ChatGPT, DeepSeek).

Finally, an article can be translated with AI or video voice dubbed with tools like ElevenLabs and made available to wider audience.

Additionally in (social) media creation field, AI can generate subtitles for videos, transform existing articles into shorter email, create enticing hooks, descriptions and titles.

Executive Assistant

Managers and C-level executives love reports. They love bird's eye view at what is happening in the lower ranks of the company.

At Enterprise and Large Business level, those reports are generally created by independent third party consulting companies (Deloitte, McKinsey, etc.).

Some of those reports can be automated and integrated within internal system. Some of the use cases:

• Automating some reports within existing CRM/ERP systems
• Analyzing customer feedback across platforms (find gaps)
• Identifying patterns in support conversations (assist in training)
• Tracking social media sentiment (for brand PR)

Beyond internal analytics, AI can ingest existing reports and analytics and provide industry trends. It can look at competition, their content strategy, SEO rankings, media topics and provide better understanding on gaps or coverages competing companies have.

Let's Not be Clowns in AI world

If you're integrating AI in your business, do not let it face your customer.

Not just yet.

Maybe in few years, but for now, it's a "backend" tool.

Always human-in-the-loop your AI, the finally decision should be made by human.

No matter how great AI gets at reasoning, it still lacks common sense.

Stay safe there and keep your customers' data safe.

The Tools

My goal is to provide business owners with tools and instruments they can integrate without much technical knowledge or expertise.

None of those tools are affiliates or sponsored links.

They are tools I have personally RESEARCHED, USED, EXPERIMENTED WITH or INTEGRATED. Some are straightforward and some you might already know and recognize.

I am being wide and broad here.

Customer Support:

• Forethought AI
• Front AI
• Otter AI
• ElevenLabs
• Intercom with AI (internally)
• Cognigy

Content Creation & Research

• DeepSeek,
• Perplexity,
• ChatGPT
• Runway by Synthesia
• ElevenLabs
• Jasper, Copy AI, Typeface AI
• Grammarly
• Clearscope
• Frase IO
• Notion AI (I am not the biggest fan, but it is what it is)

Data Analysis, Research & Reporting

• Elicit
• Research Rabbit
• Semantic Scholar
• Quantilope
• Consensus
• Tableau (heavy hitter)
• Databricks (entering Enterprise category here)
• Power BI with Azure AI