What we would tell any agency leader about using AI today

We hosted a webinar on AI for Agencies. It wasn’t a product pitch, it was a behind-the-scenes look at how we have been using AI at Whitesmith over the past three years.

We have helped lead teams through the messy, non-linear process of adopting AI, including the experiments, rewrites, and wins that didn’t come from a fancy tool, but from a simple Claude prompt on Google Sheets that just worked.

This post is about what has actually worked for us, especially for agency leaders who know they should be leveraging AI but aren’t sure where to start.

Start with the problem, not the tools

Most people jump straight to tools. We did too. Result? A pile of unused subscriptions and frustrated team members.

What works: Pick one workflow that’s costing you time or quality. Test AI on that specific problem. Don’t try to automate everything at once.

Our process is simple:

• Find the repetitive, annoying stuff
• Pick something with clear wins
• Run small experiments first

Sometimes the answer is just a good prompt in ChatGPT, not a fancy new platform.

AI can save hours if you give it the right inputs

AI isn’t magic. Tools might help, but the quality of your input determines everything. 

Here are some workflows that we improved:

• Insightful revenue notes in seconds:
  I used to spend 90 minutes every week analysing numbers and writing notes. Now I feed structured data and previous examples into a reasoning model. It generates insights in seconds that align with my thoughts.
• Smooth content creation:
  We turned internal calls into LinkedIn posts, blog drafts, and social content. Transcripts, along with our style guide, allowed us to generate multi-platform content with consistency without starting from scratch every time.
• Faster proposal drafting:
  Instead of writing from scratch, we transcribe qualification calls with Granola, feed in successful proposals, and let the model build a solid v1 draft. From there, it’s just editing and refining, not reinventing.
• Better LinkedIn filtering + messaging:
  Using Gumloop and GPT-4o, we automated profile filtering and message generation. It felt like cheating: better targeting, better messages, and hours of manual work saved every week.

Pick the tools that fit your workflow by considering what matters for your situation:

• Cost vs quality
• Speed vs accuracy
• How much context you need
• Privacy requirements

We start with powerful models to test ideas. If it works, we will experiment with cheaper options. More than the model, it’s essential that you understand if it aligns with how you actually work.

Common problems

AI can be powerful but only if you manage the common friction points:

• Data privacy: Check how tools handle your data. GDPR matters, especially with client information.
• AI hallucinations: Polished writing doesn’t mean accurate content. Always review outputs before they go to clients.
• Team resistance: Some people worry about being replaced. Get them involved early and show value fast and frame AI as an assistant.
• Tool overload: New AI tools launch weekly. We’ve made peace with not knowing everything. Instead, we run time-boxed experiments and share learnings internally.

Final reflection

Start by picking one annoying, repetitive task. Feed it examples. Test with internal work first. Evolve from there.

We didn’t adopt AI because it was trendy.
We did it to stay lean, move faster, and build smarter. And it worked: we write faster, outreach better, and spend less time on things that don’t move the needle.

If you’re curious about how AI could work for your agency, whether you’re in design, content, legal, or dev, let’s talk. You don’t need to reinvent everything, you just need to start smart.