Every leader has a moment where they have to choose to do something that is the hard way instead of the easy way, sometimes it pays off, sometimes it fails, but leaders make calls. I lead our AI practice at Seer.
Going “all in” on AI sounds great, when it comes down from your CEO. Ours says “we want to be one of the leading agencies helping clients bridge the search to LLM future.”
The reality is that to go “all in” on a technology, with NO USER MANUAL, we have no roadmaps, and while that is a tremendously fun challenge, we underestimate the hundreds of microdecisions that lead up to going “all in” and the thousands of microdecisions we’ve made over the past year that influence what we do and don’t invest in, prioritize, and test. Only to find a new model has come out and we have to re-evaluate everything we thought was / wasn’t possible just a month ago.
Not all time savings is created equal.
1,000 pennies is technically 10 dollars. There’s a reason why you wouldn’t want to walk around with 1,000 pennies to make your purchases.
Saving someone 30 minutes across a week - 10 minutes here, 15 minutes there - isn't enough to fundamentally change a teams capacity for strategic work. Isn’t that what this whole automation thing is about?
These fragments of 'saved time' often get absorbed into email checking, quick meetings, or administrative tasks rather than translating into meaningful client value.
You can not improve what you don’t measure, well.
I learned this lesson the hard way in conversations with our CFO and leadership team about resource management. When we talked about AI saving 30 minutes to an hour per week per person, the CFO's response was telling:
"If someone's working seven hours on a project and we free up 30 minutes, they're still at 6.5 hours. Even across three or four clients, those freed-up hours don't necessarily translate to capacity for a new project."
Small time savings don’t meaningfully impact agency financials.
Those freed-up hours don't add up to meaningful capacity. It's like carrying around 1000 pennies instead of a $10 bill - technically the same value, but practically useless. We've never been hired or fired over meta descriptions, so why start there.
-png.png?width=1200&height=675&name=Large%20Title%202(1)-png.png)
While tools could help us tackle easy tasks at scale and 'save time,' that saved time would be too fragmented to translate into the kind of deep, strategic work our clients need to get results in this new world with all the changes occurring in AI & search.
Have you ever had a client hire or fire you over your meta description strategy?
No. That is why I decided to prioritize areas that could improve multiple divisions on the areas that are the most likely to help us grow and retain clients. Our priorities connect back to explicit client frustrations we’ve seen and heard. We will get to the meta descriptions, just not early because I want wins that impact the finances of the business, full stop.
We'd be optimizing for efficiency without actually improving effectiveness. It's why when I look back at our decision to focus on higher-impact initiatives, even if they were harder to execute, I stand by that choice.
The following post explores why these seemingly counterintuitive decisions about time management have shaped our tests. After all, all this is, is a test…
Starting Point: We’re not going to pull a Moderna
When Moderna announced they'd deployed 750 GPTs, everyone at Seer got excited, me included. Then I thought…wow could we build 200 custom GPTs in two months, one per person? Sure. Would it help clients, I’m not so sure.
Scale changes everything. With 200 people, we can't chase pennies - we need to mine for gold. Some of our teams are using ChatGPT for meta descriptions, and that's fine, they have the permissions, tooling, and my support with the AI task force.
But my job isn't to build tools that might be obsolete in six months (our CRM HubSpot writes every meta description for us using their AI) - what HubSpot doesn’t do is look at all our business processes, cross 6 years and 170 deliverables, it doesn’t take our time per task requirements, and find the highest leverage points, so it doesn’t transform how our entire agency delivers value.
Our teams spend 20x more hours on communications with clients than writing meta descriptions.
Communications in the form of Client presentations, reviewing those comms against our acceptance criteria and client briefs, and communicating our findings and strategies to clients.
Shouldn't our AI investments reflect where most time is spent and where relationships grow or wither, impacting our agency and our clients results?
Instead of racing to build 200 different tools, we stepped back and asked one question: what actually moves the needle for our clients?
Meta description writing takes such little time already that the small savings here and there, wouldn’t be nearly as helpful as if we developed something that reviewed team strategies against what our thought leadership is saying yes have taken everything our thought leaders say in video internally for the last 18 months and are building a thought leadership GPT to check our internal thoughts against deliverables for teams, allowing us to press on strategies that may seem counter to our recent thoughts.
Documentation is unsexy, but it is the key.
Through meticulous documentation, we didn't just list tasks - we quantified the time investment behind every workflow and deliverable across all divisions.
This wasn't just counting pennies; it was following the money trail to find where chunks of valuable time were being spent, or sometimes lost.
What emerged wasn't just a task list - it was a heat map of opportunity scored across multiple qualitative and quantitative data points. While others were surface-mining for quick wins, we were building a three-dimensional model of where to drill for maximum impact.
We focused on client communications & results first
Keeping with the idea of “what tasks do we do multiple times a week, across every client, across every division” as our true north we could leave behind small fixes like meta descriptions, and focus on areas that improve or erode relationships…communications.
We identified three foundational GPTs by analyzing common threads across our 170+ deliverables, yes we cataloged 170 deliverables before we started building enterprise wide AI roll outs. That step back kept us from knee-jerk reacting…like Wil did, to basic things right in front of us. These were our base GPTs:.
- Slide Deck Outline GPT: At its core, this helps teams communicate complex information to clients in the most compelling way possible. It's about structuring narratives that resonate.
- Project Brief Generator GPT: This addresses one of the most frustrating "unforced errors" in agency work - misaligned expectations. Instead of teams copying and pasting old briefs and missing crucial nuances or new stakeholders, this GPT guides them through a thorough process of alignment from step one.
- Meeting Recap GPT: While tools like Zoom provide transcripts, they often miss the mark on client-ready communication. This GPT transforms raw transcripts into actionable meeting recaps the Seer Interactive way. It helps us store critical bits of information that are easy to get lost in the shuffle and force clients to repeat themselves.
But here's where the "base-level" concept really shines: these GPTs serve as building blocks. Take the Meeting Recap GPT - while it works well as a general tool, we can customize it for specific clients.
For instance, if a client needs their recaps presented in a particular way, we have invested hundreds of hours in Custom GPT training, trial and error, so our whole team can take the base GPT's instructions (as a building block) and create a custom version, hard-coding their preferences, terminology, and communication style.
What was a generic transcript becomes a one-click solution that speaks the client's language.
This approach matters because what often loses client trust isn't the small tasks like meta descriptions - it's when they feel we have forgotten key elements about their go to market, or when we miss truly understanding their business goals & pivots.
The Results: Where We Stand Today
We've standardized how we communicate findings in nearly a dozen analyses, and the Search Landscape Analysis provides a perfect example of how our base-level GPTs are transforming our work. Here's where it gets interesting - and counterintuitive.
-png.png?width=1200&height=675&name=Large%20Title%202(2)-png.png)
The c-suite likes saving time for senior (expensive) talent
We never had 8-10 year SEO leaders working on meta description strategy, we preserve them for unique solutions and mentoring via deliverable reviews where their years of experience shine.
Our base GPTs provide the nudges that senior leaders used to when they would do deliverable reviews.
Built on classic consulting frameworks from firms like McKinsey and Deloitte, it helps guide our "scrappier" team in telling compelling stories that drive immediate value. It's not just about organizing slides - it's about answering the crucial "So what? Now what?" questions that turn insights into action.
They get to put all their ideas and findings into ChatGPT and our custom GPTs take into account the clients meeting notes & client brief to ensure alignment.
It helps save us from our nerdy selves sometimes, where we go so deep in the data, we can lose sight of the objectives.
Chaining the CustomGPTs together 1+1 = 3!
What makes this approach powerful is how these base-level GPTs can work together. An analyst might start with our Internal Link Analysis GPT (which streamlines what used to be a manual process of clicking between spreadsheets and running macros), then feed those findings into the Slide Deck Outline GPT to structure the presentation.
They could even loop in our SERP Data GPT to validate specific keyword opportunities in real-time - yes we have a SERP GPT that lets us pull 200 factors for a keyword, right inside Chatgpt with a basic prompt…we can take a set of keywords and ask in our natural language if any of the PAA’s contain a certain keyword relative to client briefs / client meeting notes. No hopping from tool to tool, to inbox, to tool we want to do it all in ChatGPT.
Give a curious team building blocks, training, and support and see what they do with it.
Here's the real magic: because these are base-level GPTs, we can customize them for specific clients. If a client has particular language preferences or business priorities, we can "hard code" those into the GPT. This means we're not just delivering faster analyses - we're delivering them in a way that resonates with each client's unique perspective and needs.
The result? We're shifting from a model where analysts might spend two months heads-down in data to one where we're constantly in a cycle of "analyze, execute, analyze, execute." It's not about doing less analysis - it's about doing the right analysis at the right time, presented in the right way, to drive actual client value.
Parting thoughts
Look, I get it. When you're running an agency of 200 people, there's immense pressure to show quick wins with AI. But I've learned that chasing those easy efficiency gains - those scattered pennies of time savings - is actually the less effective play, especially if you have all your tasks, deliverables, acceptance criteria, already documented, you can go bigger.
That's why we chose to tackle the hard stuff first. Yes, we could have built 200 GPTs in two months and made a splashy announcement.
Does it mean we sometimes look slower on paper than agencies rolling out a meta description automater? Maybe.
I'd rather be the agency that's thoughtfully revolutionizing how we deliver client value than the one celebrating a random number of GPTs built.
Sometimes the hard way is the right way - and I'll stand by that choice every time.
