Some operations do not have a category. They have a Monday-morning complaint. A spreadsheet that has grown three sheets too many. A document type your industry handles and nobody else does. These are the engagements where we get to do our best work — because the answer was never going to be a SaaS subscription.
When custom is the answer
Three signals tell us a custom build is the right shape. One: you have already evaluated the off-the-shelf tools and none of them fit, because the problem is too specific to your domain. Two: the bottleneck involves a document or workflow that requires judgment your team has spent years building. Three: the team has tried to fix it twice already and the fix did not stick.
When all three are present, the diagnostic is not "should we build something." It is "what is the smallest thing we can build that the team will actually open on a Tuesday morning."
What we will build
The shape varies, but the modules we end up with are usually some combination of:
- Document intelligence. Contract review, claim triage, spec extraction. Trained on your historical documents, evaluated on adversarial cases that already burned you.
- Internal cockpit. The screen your team opens first thing in the morning. Account briefs, risk signals, drafted comms, what to do today.
- Cross-system bridges. Where two systems should agree but never do. We write the bridge, evals included.
- Workflow automations. The "I do this same thing every Friday" tasks. Drafted, approved, and logged in one place.
A real engagement
A finance ops company asked for a "better way to manage accounts." Eight weeks later they had an Ops O.S. built from scratch — three modules (document IQ, KAM cockpit, finance bridge), all wired together, all owned by their team. One KAM now runs 20+ accounts. They have not added an ops hire since.
Bespoke is not the same as expensive. It is the same as honest. We build the smallest custom thing that actually fits.
How we scope without a category
Two-week paid diagnostic, always. Operator interviews, workflow shadowing, three nasty real examples we work through end-to-end. By day ten we have a scope, a price, and a signed-off KPI. We do not start coding without all three.
We turn down about 40% of custom inquiries. Not because the work is unappealing — because the buyer has already decided the shape of the answer and we suspect the diagnosis is one or two layers off. We say so up front.
The bottleneck nobody else has is also the bottleneck nobody else can fix off-the-shelf. That is a feature, not a bug. It is why the build is yours, not somebody else's product.
Where AI shows up — and does not
AI shows up in the parts that are mostly judgment with implicit rules: classifying, drafting, summarizing, extracting. It does not show up in the parts that touch money, identity, or compliance — those stay deterministic and audited. We build the rails AI runs on, not the other way around.
Engagement shape
- Two-week paid diagnostic. Always. No diagnostic, no build.
- Build, scoped per engagement. Typical 6–10 weeks. KPI-priced. Miss the signed number, the next sprint is on us.
- Code, prompts, and evals in your repo at handoff. No vendor lock-in. No opaque models. Two days of operator training included.
If your bottleneck does not have a category, it might have a Scooper engagement. Book a diagnostic and we will tell you in two weeks.