AI has become the most effective marketing buzzword in recent memory. Every agency, consultant, and freelancer now claims to "integrate AI" into their work. Some of them genuinely do. Most of them use ChatGPT for content and call it a day. If you're a business owner looking to actually automate parts of your operation with AI -- not just have someone say the word in your strategy documents -- here's how to evaluate whether someone actually knows what they're doing. ## The Red Flags ### "We'll automate everything" No. Run. Automating everything is not a realistic AI strategy. It doesn't exist. AI automation works well for specific, bounded tasks -- processing documents, responding to common queries, generating first-draft content, classifying and routing enquiries. It does not run your business. Anyone who tells you they'll automate your entire operation with AI is either lying or doesn't understand what they're selling. Realistic AI projects have specific scopes, specific outcomes, and specific limitations. ### No mention of implementation complexity Implementing AI into an existing business is not simple. It requires process mapping, data preparation, integration with existing systems, staff training, and ongoing monitoring. If an AI consultancy describes their implementation as "straightforward" or "we'll just connect it to your existing setup," they haven't done the work to understand what they're dealing with. Every genuine AI implementation we've seen has required more time, more data preparation, and more iteration than the initial proposal suggested. The ones who are honest about this upfront are the ones worth working with. ### No ROI framework If someone can't tell you how the AI will generate measurable value -- in time saved, cost reduced, revenue increased, or errors eliminated -- they haven't thought it through properly. AI for the sake of AI is a waste of money. AI that doesn't connect to a commercial outcome is a science project. You need a business, not a science project. ### Vague descriptions of the AI stack Ask them what AI tools and models they use. If the answer is just "AI" or "machine learning" without specifics, they're probably not doing anything technically sophisticated. Legitimate AI consultancies will be able to explain their stack -- the models they use, why they use them, and what they're capable of. ## The Questions to Ask ### 1. Can you explain your AI stack? Not in marketing language. In specific terms. What large language models do you use and why? How do you handle data privacy? Where does the AI run -- third-party API or local deployment? What happens to my data? A consultancy that can't answer these questions at a technical level doesn't have much technical depth. ### 2. Can you show me a case study with real numbers? Not a testimonial. Not a press release. A specific engagement, with a specific client, where specific metrics improved. Time saved. Cost reduced. Enquiries processed. Something measurable. If they can't show you anything concrete, they probably don't have anything to show. ### 3. How do you handle the edge cases? AI makes mistakes. It will occasionally produce incorrect outputs, misunderstand queries, or behave unpredictably. Any serious AI implementation has a plan for this -- human oversight, fallback processes, error detection. If they describe their AI as reliable and accurate without any mention of how they handle failures, they haven't deployed AI in production at scale. ### 4. What does the implementation timeline look like, and what do you need from us? Real AI projects take time. They require access to your existing processes, data, and systems. They require input from your team to understand how things actually work -- not how they're supposed to work on paper. If an AI consultancy says they can start tomorrow and have something running in a week, they're not planning a real implementation. ### 5. How do you measure success? They should define success metrics before they start, not after. If they can't tell you at the proposal stage what a successful outcome looks like and how they'll measure it, they don't have a clear plan. ## What Good AI Implementation Actually Looks Like Good AI projects start small. They target a specific, bounded problem where AI can reliably improve outcomes. They have measurable success criteria from day one. They involve your team in the process, because no outside consultancy understands your business better than you do -- they just understand the AI. Good AI projects also accept that AI makes mistakes. The implementation includes processes for catching errors, escalating uncertain cases to humans, and continuously improving over time. Good AI projects are honest about limitations. They'll tell you what AI can and can't do in your context. They won't promise that AI will solve problems that are actually process problems or product problems. ## The Difference Between AI Integration and AI Theatre AI theatre is when a company adds an AI chatbot to their website, calls it AI-powered customer service, and then ignores the fact that the chatbot gives incorrect information half the time. It looks like innovation. It doesn't create value. AI integration is when you identify a process that takes your staff significant time -- processing incoming enquiries, generating quotes, classifying incoming documents -- and replace it with an AI system that does it reliably, faster, and at lower cost, with human oversight for the edge cases. One is a marketing line. The other is a business improvement. Before you hire anyone to "do AI" for your business, understand which one you're being sold. --- Looking to understand what AI could actually do for your business? [Book a free diagnostic](/contact). We'll tell you what's realistic, what isn't, and what it would actually take to implement.