AI Will Not Save Your Broken Operations
by Jared Novack, Chief Data & Information Officer @ Upstatement
The path to an AI-enabled agency runs through four important stages.
The next version of your agency is being founded right now, AI-native by default. This isn't a new story. You likely founded your own agency as a digital native and ate the lunch of the incumbents of the day. This new batch, even if they embrace a traditional services framework and human-centered craft, gets to start fresh on a stack of AI-native apps and approaches that drive operational overhead to near zero.
That overhead is exactly the part of the business clients resent paying for. The fall campaign your creative team is building? Worth every penny. The reports your finance team owes the parent company? Not their problem.
"But we're launching agents!" you say. And you might — but if those agents are bolted onto the operational patterns of the last twenty-five years, they're woefully under-equipped for what the moment actually requires.
Before COVID and the rise of remote work, the machinery of our business was built around an office-centric assumption. In that world, the oral tradition ruled: the status of a deal lived in someone's head or a whiteboard. The office was the database. When COVID forced a quick pivot away from that database, we did our best to rebuild through the CRM and other systems — but we never really escaped the human-to-human core. It's not just us. I've had the chance to compare notes across sister agencies I had assumed were well ahead of me, and found the same thing, often worse than I expected. Largely, operations are still running on word of mouth, emails, Slacks and incomplete spreadsheets. It's a perfect mix of unclean data and mushy process. AI can't save these broken operational stacks — it will merely expose and amplify them.
We all have a chance to re-found our agencies in this AI-native context. Yes, we're carrying baggage up this mountain — but the climb is non-optional. We'll need to work through distinct stages that unlock AI's full operational capacity. If you jump straight to deploying agents, those tools will just get stuck on the same problems our teams have been circling for years.
Stage 1: Oral Tradition
This is where we're likely all starting (or emerging from). Here, the source of truth lives in people's heads, or in systems only they have access to. The humans in this mix might even relish the arrangement, as it protects their value as knowledge holders — as if being a living S3 bucket is one's true calling. Chatbots can help these humans do their thing slightly faster, but it's still a human thing. AI merely watches.
Stage 2: The Dangerous Middle
This is where I've seen many groups declare victory and go home. We've graduated to a number of half-trusted spreadsheets and systems that look like progress and provide a shared source of truthiness. But these systems still require oral tradition or years of accumulated context to truly use. Who's the main contact for that important client? Is that project really open or closed? "It depends," you might hear — or say. These groups have carried the value of creative ambiguity (which can be good!) over into operational ambiguity (which rarely is). This is the worst place to add AI, and also the most common. Here, an agent or chatbot will fluently work against stale or dirty data. Introducing AI at this stage is like introducing 100 junior employees into the mix: now we just have even more entities to contend with, each carrying or amplifying their own version of the information.
Stage 3: Fully Digitized
Now we have a shared source of truth. The CRM and other systems rule, and they're interlinked: the deal in HubSpot connects to the invoice in QuickBooks. Humans utilize this operational data in their work — they know what deal sizes are likely to land with a client, the average margin for a type of project, or the last time we talked to a particular contact. The data is there and it's reliable. What's still human is the connective tissue: a person decides which threads to pull together and acts on what they find. AI, through chat and coding agents, finally has enough trustworthy, known data sources to reliably answer the questions we put to it. But it's still we who are asking, connecting, and executing. The challenge at this stage is to be precise about who is responsible for those connections so everything isn’t a jump ball.
Stage 4: AI-Enabled
Here, actions on the business come from actions on our databases. These are well documented and consistently executed. Indeed, that documentation is what informs those actions: as the rules and guidance for an AI agent. Instead of a human remembering to compile a weekly pipeline report, an AI agent draws on the fully digitized data to draft it, and even distribute it, automatically. We're still discovering the upper limits of what's possible here, from Zapier-like automations to something that feels like a teammate who can accept and execute an assignment.
AI value compounds at this stage. One agent flags when a project is over budget. Another drafts a recovery plan. Another compares that plan against the success or failure of past plans. Yet another prepares the resourcing changes needed to support it. The chain keeps going as far as your definitions do.
But asking an agent to perform these steps requires a clear definition: What does it mean to be over budget? When do we monitor? How do we intervene? To simply drop a machine's instructions onto a group of humans invites resistance. Instead, building these rules and systems requires human involvement.
Done right, this doesn't just reduce costs; it frees humans to focus on the truly creative and strategic layers. When our operations director was responsible for so many execution pieces, there wasn't time or energy left for anything else. Now they can consider the whole of our system, down to the fundamental questions. What does it really mean to be over budget? How do we better share risk with clients? What new interventions or offerings should we propose?
Advancing through these stages should be an urgent priority for every operations team. It requires both a clear mandate from leadership and the buy-in of a broad team willing to do the work. Technical skill helps, but this is fundamentally a political problem. How do you help everyone share in the upside? Make them the authors of the change? Reward small actions with positive feedback loops? Start with an easy, stakeholder-driven problem, solve it, earn the next system, and repeat.
Does that sound like a general description of a healthy business operation? Ding ding ding! That's the truth of it. A healthy operation is the thing being amplified and enabled here — the thoughtful adoption of AI simply accelerates a team toward where it was always heading.
Which brings us back to the baggage. It might seem we're at a disadvantage against the upstarts, weighed down by years of accumulated process while they travel light. But that baggage holds the one thing these new groups can't match: experience. If expertise is the thing we sell, then it's because of that accumulated knowledge, relationships, and creative capital that we hold the inside track on the value that matters most. The upstarts can stand up a clean stack in a weekend. They can't stand up decades of knowing what “good” looks like. That's precisely why this transition is non-optional — the baggage is worth carrying, but only if we do the work to make it usable.
Over the coming years, it will be a pleasure to welcome new AI-native groups into SoDA and the industry. And if we make the right changes now, we'll be right there to create alongside them.
Jared Novack, Upstatement
