The Agency Tax: Red Flags Before You Sign

I’ve spent the last few years talking to CEOs, CTOs, and COOs who’ve been burned by development agencies. Seven-figure contracts. Products that broke on contact with real users. A rewrite bill on top of the original one.

Software delivery has a poor record in general. In a 2024 BCG survey of more than 1,000 C-suite executives across 59 countries, more than two-thirds of large-scale tech programs were not expected to land on time, on budget, and within scope. In another BCG survey that year, nearly half of executives said more than 30% of their technology development projects run late or over budget.

Most of that damage is visible before you sign. Here are six red flags, the question that exposes each one, and what a good answer sounds like.

Red flag 1: Discovery with no end date

You sign the contract. They schedule a kickoff. Then come three to six months of “discovery”: workshops, wireframes, stakeholder interviews, and scope documents. Ten to twenty percent of your budget is gone before anyone writes a line of code.

Discovery at that length protects the agency. It buys time to staff the team and builds a paper trail for when timelines slip. Meanwhile your market window closes and a competitor ships.

Ask: “In what week does code reach production?”

A good answer names a week, and it’s measured in weeks. Understanding the business matters, so good partners time-box it: a short, fixed-fee assessment with a written recommendation at the end.

Red flag 2: The pitch team leaves after the pitch

The senior architect who impressed you in the sales meeting disappears after signing. The work goes to whoever is free on the bench: mid-level developers, managed by someone juggling three other clients. The engineers who have built what you need are on the Fortune 500 account.

Ask: “Name the engineers who will write the code. What have they shipped, and will they be on the keyboard every week?”

A good answer is a list of names, with résumés you can check, and the senior people you met on it.

Red flag 3: The overhead tax

You think you’re paying for engineers. In a typical agency, 45% of the budget goes to people who don’t build: project managers, account managers, scrum masters, business analysts, design leads who don’t design, and tech leads who don’t code.

Then comes the markup. Agencies typically bill at twice their cost, so an engineer who costs them $100 an hour costs you $200. You’re paying for their bench, their sales team, and their office.

Out of every dollar you spend, about 55 cents reaches the people building your product.

Ask: “What share of our fee pays for engineers who write code?”

A good answer is a number. A partner who can’t give you one is telling you something.

Red flag 4: Demo-grade code

They deliver the MVP. It works in the demo. You launch, you get traction, and the system starts to fall over. The database chokes at 1,000 concurrent users. API timeouts spike. The authentication layer has race conditions. You’re looking at a codebase that’s 70% technical debt and a rewrite that costs three times the original build.

This happens when a contract is measured by delivery date. Teams build the happy path, fix bugs as they surface, and add complexity with no plan to maintain it.

Ask: “What’s in the handoff? What’s your test coverage bar, and what load have you tested to?”

A good answer lists a production system, automated tests with a coverage bar, a software bill of materials (SBOM), and runbooks your team can operate from.

Red flag 5: AI with no evidence

Every agency now sells AI. Many of them wrap a model call in a demo and call it a product. In a regulated business, that fails the first time an auditor asks why the system made a decision.

Ask: “How will you measure the model’s accuracy on our data? Who reviews its output, and how do we trace a decision back to its source?”

A good answer covers evaluations on your own data, a person who signs off on consequential decisions, and a record of every output, the model behind it, and the evidence it used.

Red flag 6: You don’t own what you paid for

The code lives in the agency’s repository. The infrastructure runs in their cloud account. The only people who understand the system work for them. Leaving means starting over, so you stay and keep paying.

Ask: “Who owns the code, the infrastructure, and the documentation? Could we run this without you tomorrow?”

A good answer is yes: your repositories, your cloud accounts, and a handoff your own team can run.

Why this keeps happening

The traditional agency model makes money from billable hours and bench utilization. Long discovery fills hours. Junior staff on senior rates widen margins. Layers of management make the org chart billable. Fragile code creates follow-on work. Each of these follows from the incentives.

The checklist

Before you sign, get a straight answer to each of these:

1. In what week does code reach production?
2. Which named engineers will write the code, and what have they shipped?
3. What share of our fee pays for engineers who write code?
4. What’s in the handoff: tests, coverage bar, SBOM, runbooks?
5. How will you measure and audit the AI’s output?
6. Who owns the code and infrastructure, and could we run it without you?

How SAIL answers them

We built SAIL around these six questions.

• Production by week 4. Architecture is settled by week 2 and the first deployment ships by week 4. When a client wants a map first, we run a two-week audit: one or two engineers shadow the people who run your key processes and deliver an assessment and a recommendation. The fee is waived if it leads to a build.
• Two or three senior engineers, on the keyboard. Ex-Google and ex-Oracle engineers who’ve shipped products to millions of users. The people you meet are the people who build.
• 100% of capital goes to engineering. Every dollar pays for engineers who write code.
• Production-grade handoff. A production system, tests, an SBOM, and runbooks, built with the discipline we use on our own products.
• AI proposes. People sign. Every output carries its provenance: the data, the model, and the reviewer behind it.
• You own everything we build. Code, infrastructure, and documentation.

When we built SDC’s clinical intelligence platform, a strike team collapsed 12 disconnected tools into one platform in three months and cut trial timelines by 36%. Senior engineers who knew what to build made the architecture calls early and spent their time building.

Before you sign that SOW, ask yourself: am I paying for engineering, or funding someone’s org chart?