How to build an AI brain from your customer data in less than 2 hours

Your own customer knowledge center is an accelerant, allowing you to get to the best version of your marketing more quickly. Marketers who embrace AI are already seeing substantial benefits: Those who use AI say it saves them 13 hours a week at work.

It’s more within reach than you may think and takes only an hour or two to set up.

The ingredients here are:

1. Customer inputs (30–60 minutes)
2. A capture mechanism (30–60 minutes)
3. An AI project (10–30 minutes)

Our goal is to cast a wide net and feed our AI project as many data sources as possible so it can find signals in the noise that we may have missed or, more likely, don’t have time to sift through.

Here’s how to get started.

Step #1: Track down your customer inputs

To search through your customer feedback, you’ll need to, well, collect it. The source here depends on the kind of business you run, but it is only possible if you have some sort of digital layer to your business. AI can’t listen in on our in-store conversations—at least not yet.

Some example inputs to gather:

• Customer reviews — Google Local, Angie, Yelp, industry-specific sites like G2, and others.
• Customer support software — Zendesk, Intercom, Help Scout
• Email communications - Especially shared support inboxes.
• Social media software - Anything you use to reply to your customers on social media, such as Buffer, Hootsuite, etc.
• Video calls - Zoom, Google Meet, or other video chat or note-taking software.
• VOIP services — Aircall, Google Phone, etc
• Surveys and feedback forms — Post-purchase survey or automated feedback form that is emailed to customers.

If you use a service like ActiveCampaign, you also have an easy repository of email responses.

Step #2: Capture them

One of the best uses of an AI tool is to turn unstructured data into structured insights. As a result, you don’t need to be shy about collecting raw data—the more the better. Whatever service you are using, find its export functionality (most but not all services should offer something like this). Then, turn your customer feedback into text or csv files that can easily be added to the AI’s project knowledge.

Some approaches:

• Google Takeout: If you own a local business, you can export all of your Google Local reviews.
• A call recorder: Granola, Fathom, Gong, or built-in features of your video call service allow you to turn your calls into transcripts that you can gather and paste into your AI project. (Remember to adhere to your company’s recording policies and your state or country’s call recording laws).
• Software exports: Depending on what you use, most tools have some export function where you can have a large text or csv file of customer feedback. Examples include:
  • Reviews: Trustpilot, Yelp
  • User surveys: Typeform, SurveyMonkey
  • Ecommerce: Shopify

Think through your most common workflows and the tools you use each week. Export as much feedback data as you can to your computer.

Note: not every tool provides export options.

Step #3: Create your AI project

With your data in hand (or on your desktop), let’s build your brain. There are dozens of AI tools, but we can keep it simple. Whatever you use needs:

1. A ​“project” functionality where you can upload data to serve as a central repository and training data for your AI agent.
2. Team functionality so you can share results with your coworkers, as well as allow them to upload their own customer reviews.

We suggest two options: ChatGPT Projects and Claude Projects. These are commercially available, already approved by most organizations, reasonably priced, and come supported with vast amounts of education and training. For the purposes of this example, we’ll use ChatGPT, which, as of this writing, is more widely used. 

• Step 1: For most services, the project’s functionality requires a paid tier. Sign up for ChatGPT here. Or if you are already a user, you can upgrade in the upper right-hand corner. Click on your user avatar and then ​“Upgrade.” You’ll want to select the ​“Pro” tier.

Step 2: In the left sidebar, select ​“New Project” and give your project a name.

Step 3: With your new project created, select ​“Add files” and upload your exported customer feedback.

Step 4: Prompt your new customer knowledge center or ​“customer brain.” 

In this example, I run a roofing business, and I uploaded my transcripts of client phone calls to help ensure my response is comprehensive yet not too time-consuming — so I can focus on driving new business.

Or, based on my call transcripts, I’ve asked my customer brain for my insurance business to generate headlines for a new landing page.

How to get sharper, more reliable outputs

Use custom instructions. As you get a feel for your customer brain, you can further modify outputs with ​“custom instructions.” 

In ChatGPT, that’s the ​“Add instructions” button beneath the input box. You can ensure that it uses an uploaded style guide, or never mentions a competitor, or a myriad of other guidance that helps the output be as useful as possible. 

One suggestion: Include a ​“cite all quotations and customer feedback with the source document and name of the customer.” This helps you double-check output and prevent hallucinations from making their way into your marketing.

Upload stakeholder feedback. Take transcripts from internal feedback sessions and use them to hone your model. For example, if your manager offers constructive criticism of your latest campaign, upload that information into your knowledge center to help inform output.

Upload campaign results. Include charts, metrics, and post-mortem notes from campaigns to help inform asset creation. For example, if your audience does not respond well to casual subject lines, that should guide future output. You can also include this guidance in your ​“Add instructions” field.

Create a persona bot. When you feel confident that your knowledge center is mapping well to actual customers, you can ask how a customer may respond or how a campaign may perform based on past behavior and feedback.

Don’t forget ​“Deep Research.” Both Claude and ChatGPT have ​“Deep Research” functionality that produces research report-grade outputs. This is best for serving as a starting point for deeper strategic analysis, versus one-off or tactical questions. Ask your knowledge center to:

• Create customer analysis for quarterly reviews
• Rewrite your marketing KPIs based on previous input
• Suggest new or overlooked strategic adjustments you could make

Hallucinations, security, and the limitations of AI: What you need to know