The agency that runs on a toolset it built itself
Here’s the conclusion up front: the biggest change AI made to our agency wasn’t speed. It was that “did anyone check?” stopped being a question. Every account, every day, the same rigorous sweep — and our senior people now spend their hours on the calls only experience can make.
This piece is about what that actually looks like inside a marketplace agency — what we automated, what we deliberately didn’t, and the one rule we treat as non-negotiable.
We didn’t start with AI. We started with a data problem.
TESMO has been a data company since 2008 — our first product was pricing and demand-planning tooling built on per-category demand curves correlated against sales rank. We hired a machine-learning PhD in 2013 to push the models further. It failed; the tooling was a decade away. The instinct stayed.
So when modern AI tooling arrived, we didn’t have to invent a data culture. We had to connect one. The foundation was already there: a unified data warehouse spanning every account we manage — orders, advertising, inventory, subscriptions, competitive pricing, both 1P and 3P.
What runs as code now
On top of that warehouse, our engineering team built a set of servers and tools that our AI layer works through. The practical effect: a library of production routines that run across the whole portfolio —
- Diagnostics: weekly ad-efficiency reviews for every brand, with a dedicated variant for 1P businesses.
- Anomaly detection: a daily variance monitor that flags pattern breaks across every account — hours after they happen, not at the quarterly review.
- Reporting: weekly advertising workbooks, pacing reports, and dashboards that build themselves and get reviewed by humans, instead of the other way around.
- Planning: replenishment models that account for seasonal curves and lead times, refreshed continuously.
One concrete number, because numbers are the point: a recent catalog update touched 28,407 cells across 4,923 SKUs. It ran in about fifteen minutes, with a reusable script as the artifact. The old version of that task was a week of someone’s life and a nonzero error rate.
What deliberately stays human
Three of our four operating layers are human on purpose: strategy, audit, and execution. The machine is the substrate underneath them — it sweeps, surfaces, and recommends. It does not decide.
And one rule has no exceptions: anywhere a decision touches money or price, the guardrails are enforced in code, and a human sits at the commit point. Not “the AI is usually right.” Not “we spot-check.” Enforced, deterministically, every time. We think any agency claiming full automation across pricing and spend is describing a risk, not a feature.
The tradeoff nobody mentions
Building your own toolset is slower and more expensive than buying a dashboard subscription, and for the first year it looks like a distraction. The payoff is compounding: every routine we ship runs across every account we’ll ever manage, and the marginal cost of rigor drops toward zero. A boutique with a dozen premium brands can now sweep its whole portfolio with the consistency of a firm ten times its size — without hiring the ten-times staff or diluting the senior attention clients actually came for.
That’s the honest pitch for an AI-run agency: not magic, not headcount reduction. Consistency, compounding, and senior judgment spent where it’s worth something.
The takeaway
If you’re evaluating an agency — ours or anyone’s — skip “do you use AI?” Everyone says yes. Ask instead: what runs automatically across every account, every day? What’s enforced in code when money is touched? And what do your senior people do with the hours that frees? The answers separate infrastructure from stickers.
Want this pointed at your channel? Let's talk.