I’ve been running a software agency long enough to notice a pattern. Product companies ship one product — their team knows the codebase, the domain, the performance patterns. When they estimate a feature, they’re estimating inside something they already understand.
An agency like ours ships twenty projects a year across five different industries. The stack changes, the client changes, the domain changes. Every time, the team is essentially starting from zero. We’ve had quarters where we jumped from an e-commerce build straight into a medical SaaS product, then an IoT dashboard — each one a completely new world.
I used to think we just needed better estimators. After watching margins evaporate on fixed-price contracts for years, I’ve changed my mind. The estimation problem agencies face is structural, not personal. And pretending otherwise costs real money.
The Four Things That Make Agency Estimation Brutal
Project Diversity Kills the Learning Curve
Product companies build up estimation accuracy because they work in the same context month after month. Their historical data is actually relevant to whatever they’re building next. When a product team says “this feature will take two weeks,” they’ve probably built something similar before, in the same codebase, with the same people.
Agencies don’t get that luxury. One project is an e-commerce platform. The next is a medical SaaS application. Then an IoT dashboard. The data from project A tells you almost nothing about project B — completely different stack, domain, and client expectations. Every estimate is, to some degree, a first-time estimate. And first-time estimates are the least accurate ones. There’s no way around it.
Pressure for Fast Quotes
Clients rarely give you the time an honest estimate requires. They’re comparing three or four agencies and they want answers in a week. Bid in a week and you might win. Bid in two and you’ve already lost. So you’re forced to estimate based on a three-paragraph brief — no discovery, no requirements workshop, just a few lines from a client who may not know what they actually need.
The spread between an optimistic and pessimistic estimate at that level of information can be 400 to 800 hours.
Fixed Price Without a Safety Net
When a product company’s feature takes longer than expected, nobody pays extra. The feature ships when it ships. The cost gets absorbed as part of the product investment. In an agency, every extra day eats into margin. The price was fixed at signing, the work took longer, and the difference comes out of your pocket.
We once quoted a project at 900 hours and finished closer to 1,400. The client got exactly what they asked for. We just didn’t charge enough for it — and there was no mechanism to correct course mid-project.
This creates a perverse dynamic. The worst thing that can happen to an agency is winning too many projects at too-low prices — you work twice as hard and lose money on every one. I’ve seen it happen more times than I’d like to admit.
That 900-hour project taught me something: the price wasn’t wrong, but the process was. We had no way of catching the gap before it was too late.
Scope Creep as Default Mode
In a product company, scope changes are just product decisions. The team weighs the trade-off and adjusts the backlog. No invoices, no renegotiation. In an agency, every scope change is a margin negotiation. And honestly, many agencies handle this badly — they fold with a “we’ll do it as a favor.” The problem is that a favor this week becomes an expectation next week. Before long, the scope has grown by thirty percent and the price hasn’t moved.
Five Things That Actually Help
We’ve tried a lot of approaches over the years. Some worked, most didn’t. Here’s what survived.
Build a Component Library
Every project repeats certain patterns — authentication, admin panels, dashboards, reporting, API integrations. We built an internal library with real data from completed projects. Things like “REST API integration: 3–5 days. Auth module: 5–8 days.”
When a new project comes in, we’re not guessing from scratch. We’re referencing data from similar components we’ve built before. It makes the estimate defensible in a way that intuition never can.
Sanity Check Every Estimate Before Sending
This one seems obvious but almost nobody does it consistently. Have someone who wasn’t involved in the estimate review it before the proposal goes out. They’ll catch things the estimator missed — missing cost categories, inconsistencies between scope and hours, pricing that’s unrealistically low.
It takes about thirty minutes and it has saved us weeks of margin erosion.
Offer Variants Instead of One Number
Instead of sending a single price, we started offering scope options — core, standard, full. The client picks the scope level that fits their budget rather than asking us to cut margin. It’s a small change but it reframes the whole negotiation. They’re not pushing back on our number, they’re choosing between options we designed.
Automate the Mechanical Parts
You can’t automate the judgment that goes into an estimate. But you absolutely can automate everything around it — generating proposal structures, recalculating costs when scope changes, exporting to professional formats. Every minute you spend copying numbers from a spreadsheet into a proposal is a minute you’re not thinking about whether the estimate is right.
Make Change Costs Visible
Every scope change after signing should be documented, priced, and approved. When the client sees “Add PDF export: +2 days, $1,200” written out, they understand that changes aren’t free. It reduces frivolous requests and protects your margin on the ones that do happen.
It’s not about nickel-and-diming the client. It’s about making the economics of the project transparent for both sides.
None of these things are magic. We still miss estimates, we still have projects that eat into margin. But the misses are smaller than they used to be, and when we do catch a problem early — because someone sanity-checked the number, or because the client saw a cost they hadn’t considered — it feels less like gambling and more like running a real business. That alone made the whole effort worthwhile.