I had a lead meeting recently that perfectly sums up where most businesses are going wrong with AI right now.
They reached out to us from a marketing campaign. They’re in the property space,primarily construction, and they were incredibly keen to chat about AI. We sat down, I got to know them, and then I asked a very simple, general question:
“So, what brings you here? What do you want AI and automation to actually do for your business? What are the top two or three projects you’re thinking about?”
Her reply?
“I don’t know. What do you offer?”
I started scratching my head. I said, “Well, we offer a lot of things, but it depends entirely on your needs and wants. Have you thought about where the bottlenecks in your business are? Or the really simple, repetitive tasks that your team might do 10 or 20 times a day?”
Her response: “No, we haven’t thought about that.”
“Okay,” I said. “Have you mapped out your sales process step-by-step to see what routine tasks have to get done every time?”
“No, we haven’t done that either. We’ve got a baseline idea of it, but it’s certainly not documented well.”
I tried one more angle. “From a customer service perspective, have you listed out the top 10 questions that get asked by email or phone? Do you know how many inquiries actually come in, or how long it takes your team to respond?”
Her response: “No, we haven’t done that either.”
The “Garbage In, Garbage Out” Reality
This conversation really got me thinking. It all comes back to a fundamental truth in business: you must have absolute clarity on what your problems are, what your challenges are, and an intimate understanding of exactly what is transpiring in your day-to-day operations.
It’s that old saying: “garbage in, garbage out”
If you don’t have clarity around your processes, your systems, your FAQs, your tracking, your response times, and where your human labour is actually being spent, it becomes practically impossible to define where automation and AI can genuinely impact your business.
Look, it’s really like anything in life. If you don’t have absolute clarity on the outcomes you’re trying to achieve, you are never, ever going to get there.
The No. 1 Place to Start
The number one place to start your AI journey isn’t with picking a software tool or asking an agency what they sell. It starts with clarity.
You need a crystal-clear understanding of what is currently happening in the business and what the top priorities are for automation.
Once you have clarity about exactly what you are trying to achieve you can finally start working backwards. That’s when you ask the right questions:
- How are we going to achieve it?
- What is the process we need to build around it?
- How are we going to measure it?
- How are we going to track it?
- How do we get better and better and better at what we do?
Stop looking for a magic AI wand. Document your processes, find your bottlenecks, and get crystal clear on your problems. Once you do that, the solutions will become blatantly obvious.
The 5 Core Steps to Build AI Agents for Sales & Customer Service
Once you have done the hard yards, and you have clarity, your strategy, and you know exactly which bottlenecks you are targeting, it’s time to build.
If you are implementing AI agents into your sales or customer service workflows, don’t just wing it. Follow these 5 core steps to ensure you actually get a return on your investment:
- Define the Structured Output
If your AI agent is qualifying a sales lead or resolving a customer service ticket, you don’t just want to have a polite chat. You want it to take action. Force the AI to output structured data (like strict JSON) so it can automatically update the right fields in your CRM without a human having to copy and paste.
- Build a Proper “Knowledge Base/Brain” (RAG)
Don’t just dump all your company PDFs into a prompt window and hope for the best. That leads to massive API costs and hallucinations. Build a robust Retrieval-Augmented Generation (RAG) pipeline. This ensures your customer service agent retrieves the exact right policy or pricing sheet before it ever answers a customer’s question.
- Engineer the “Human-in-the-Loop”
Never give an AI agent unconstrained autonomy. If a customer is furious, or a deal is worth $100k, a bot shouldn’t be flying solo. Program confidence thresholds into your agents. If the AI is 95% confident, let it reply. If it drops to 80%, have it draft the response and route it to a staging queue for a human to approve with one click. AI should assist your team, not blindly replace them.
- Implement Strict Observability
AI fails silently. It doesn’t always throw a neat error code; sometimes it just gives a prospect the wrong pricing tier. You must integrate observability tools to track every single prompt, response, and API call. You can’t manage what you can’t measure, and you need to review the agent’s work weekly to continually optimise its instructions.
- Stop Boiling the Ocean (The Step Approach)
Don’t try to build an AI agent that completely replaces your entire sales and customer service departments on day one. Paralysis by opportunity will kill the project. Pick one simple, high-volume bottleneck, like automating the answers to your top 5 FAQ emails or extracting data from incoming web forms. Get that one quick win. Prove it works. Then use that momentum to fund and build the next step.
Keep it simple. Get 1% better every day, and let the compounding results do the heavy lifting.
