A founder rang me in November 2024, two weeks before Christmas, absolutely furious. He'd had a web agency quote him £14,000 for a bespoke logistics dashboard. He'd signed. Six months later the final invoice landed at £61,000. Nobody had lied to him, technically. But nobody had told him the truth either — because nobody had actually done the work of estimating properly before taking his money.
I've been building software for over twelve years. Seahawk alone has shipped more than 12,000 sites and applications. And the single most consistent failure point I see — with founders, with freelancers, with agencies quoting projects — is that nobody has a rigorous way to estimate cost before a line of code gets written. People guess. They anchor on vibes. They use the last project as a reference point without checking whether it was remotely comparable.
So. Here's the framework I actually use. Not a spreadsheet template with placeholder cells. A real mental model, with current 2026 numbers, specific tools, and the caveats that matter.
---
Why Most Software Quotes Are Fiction
The problem isn't dishonesty (usually). It's that software estimation is genuinely hard, and most people underestimatehowhard — so they reach for shortcuts that feel like rigour but aren't.
A developer gives you a gut-feel number. An agency multiplies their day rate by some estimated sprint count. A freelancer looks at a similar gig on Upwork and works backwards. None of these are wrong exactly, but none of them account for the real cost drivers: integration complexity, decision latency on the client side, scope creep on features that seemed minor, testing overhead, and the silent killer — environment setup and DevOps that nobody prices in.
Back in 2021 I was running a project for a property-tech client in Manchester. Simple enough on paper: a tenant portal with document uploads, rent tracking, and a maintenance request workflow. Initial estimate was £28,000. We'd done similar builds. But this client was using a legacy property management system that had a proprietary API nobody had touched in four years. Integration alone blew out by three weeks. Final cost: £47,500. The client was fine about it because we'd flagged it the moment we found the API docs — but the initial estimate was still fiction, and I owned that.
The Cone of Uncertainty Is Real
The cone of uncertainty, popularised by Steve McConnell, describes how project estimates get more accurate as you get closer to delivery. At ideation, you might be off by 4x in either direction. After detailed design, maybe 1.25x. Most founders are getting quotes at the ideation stage and treating them like signed contracts.
The practical upshot: any quote you receive before detailed specs are written should be treated as arange, not a number. If an agency gives you a single figure before they've seen your data model, user flows, and third-party dependencies — that number is decorative.
---
The Four Real Cost Buckets
Before any estimator can work, you need to split the project into the right categories. Not "frontend, backend, QA" — those are roles, not cost drivers. The actual buckets:
1. Greenfield vs. Integration Work
Greenfield — building something from scratch against your own database and business logic — is the cheaper, more predictable half of almost every project. It's the integration work that gets you. Every external API, legacy system, payment processor, or third-party auth provider adds non-linear complexity. I typically add a 30–40% contingency on any scope line that touches external systems.
2. UI/UX Design (Don't Skip This Line)
A lot of founders treat design as optional or try to roll it into the development estimate. Mistake. Design done properly — wireframes, interactive prototypes inFigma, a tested component library — typically runs 15–25% of total project cost. Skip it and you'll pay twice: once in developer rework when requirements turn out to be ambiguous, and again in user research later when the product doesn't convert.
3. Infrastructure and DevOps
Nobody quotes this properly. Staging environments, CI/CD pipelines (we use GitHub Actions on almost everything now), containerisation with Docker, cloud hosting on AWS or GCP — these are real costs that don't disappear because nobody itemised them. For a medium-complexity SaaS, budget a minimum of £3,000–£6,000 for initial infrastructure setup, plus ongoing monthly costs that typically run £200–£800 depending on usage.
4. Testing, QA, and Launch Overhead
Automated test suites, manual QA passes, performance testing, security review for anything handling payments or personal data — this is typically 20% of total development time if done right. Most agency quotes include "testing" as a line item and mean "a dev clicked around for an afternoon before the handover call."
---
A Working Estimator: The Module Method
Here's the actual framework. I call it the Module Method because it forces you to break a project into discrete, independently estimable chunks rather than treating it as a monolith.
Step 1: List every feature as a user story.Not "user management" — that's too vague. "A user can register with email and password, verify their email, reset their password, and update their profile photo." Four stories. Each one has a cost.
Step 2: Rate each story by complexity.I use three tiers:
- Simple(S): Pure CRUD, no external dependencies, standard UI patterns. Think: a settings page, a profile update form, a data table with sorting.
- Medium(M): Some business logic, one external integration, or non-standard UI. Think: a filtered search with saved queries, a webhook handler, a Stripe subscription flow.
- Complex(C): Multiple integrations, real-time features, algorithmic logic, or heavy infrastructure. Think: a live chat system, a recommendation engine, a multi-tenant permission model.
Step 3: Assign hour ranges, not point estimates.
In 2026, working with a mid-tier UK agency or a strong nearshore team, here's what I'd budget:
- Simple story: 4–8 hours
- Medium story: 12–24 hours
- Complex story: 30–80 hours (yes, that wide — complex is genuinely unpredictable)
Step 4: Apply your rate.
Current market rates worth knowing:
- London-based senior developer (freelance): £90–£140/hr
- UK agency (full team, project managed): £80–£120/hr blended
- Strong nearshore (Eastern Europe, LatAm): £35–£65/hr
- Offshore (South Asia, Philippines): £15–£35/hr — and yes, you can get excellent work here, but communication overhead is real and needs to be priced in
Step 5: Add overheads explicitly.
Don't absorb these. Line-item them:
- Project management: 10–15% of dev hours
- Design (if not already scoped): 15–25% of total
- DevOps/Infrastructure setup: fixed £3,000–£8,000 depending on complexity
- QA: 20% of dev hours minimum
- Contingency: 20% on greenfield work, 35% on anything with significant integration
---
A Real Example: SaaS Dashboard, 2026 Pricing
Let me walk through something concrete. A founder comes to me wanting a B2B analytics dashboard — a SaaS product where their clients log in, see their performance data pulled from Google Analytics 4 and a custom event database, can export reports, and manage their own team's access.
Here's how I'd break it down:
- Auth system (email + Google SSO, role-based access):2 Medium stories = 48 hrs
- GA4 integration + data pipeline:1 Complex story = 55 hrs
- Custom event database schema + API:2 Medium stories = 40 hrs
- Dashboard UI (charts via Chart.js or Recharts, responsive):3 Medium stories = 65 hrs
- Report export to PDF/CSV:1 Medium story = 18 hrs
- Team management (invite, remove, role assignment):2 Simple + 1 Medium stories = 28 hrs
- Billing via Stripe (subscriptions, upgrade/downgrade):1 Complex story = 45 hrs
Raw development total: ~299 hours
Apply a blended UK agency rate of £95/hr:£28,405
Add:
- Design (20%): £5,681
- DevOps setup: £4,500
- QA (20% of dev hours, same rate): £5,681
- PM (12%): £3,409
- Integration contingency (30% on GA4 and Stripe work): £3,000
Total estimate: ~£50,676
That's a real number for a real product. Not shocking if you understand what's in it. Absolutely shocking if you went in expecting £18,000 because that's what someone told a founder in your network last year for "something similar."
---
The Hidden Costs Nobody Mentions
Licences and Third-Party Services
Mapbox for mapping features. Twilio for SMS. SendGrid for transactional email. Algolia if you need proper search. These are ongoing costs that founders routinely omit from their software budget entirely because they think of them as "just APIs." A mid-scale SaaS with 10,000 active users might be paying £800–£2,000/month in third-party service fees before a single server bill lands.
Security and Compliance
If you're handling personal data in the UK, GDPR isn't optional. If you're touching payments,PCI DSS complianceadds overhead. If you're in health or finance, you're looking at additional audit costs that can run £5,000–£20,000 just for initial certification work. I've seen founders get genuinely blindsided by this. One fintech client of ours in 2023 had budgeted zero pounds for compliance work and then needed £14,000 of it before they could launch.
Handover and Documentation
Good documentation — API docs, deployment runbooks, onboarding guides for future devs — costs real time. Budget 5–8% of total project hours for documentation if you ever want to be able to maintain, extend, or sell the thing.
---
How to Pressure-Test a Quote You've Received
You've got a proposal in front of you. Here's how to audit it before signing:
- Ask for the feature breakdown behind the number. If they can't give you a story-level breakdown, the quote is a guess.
- Check whether DevOps, QA, and PM are line items or hidden in a blended rate. Hidden usually means undercooked.
- Ask what the contingency policy is. Do they cap overruns? Time-and-materials or fixed price? Each has real implications.
- Find out what assumptions they've made about third-party integrations. Ask them directly: "Have you read the API documentation for [specific service] before quoting?"
- Ask what happens if requirements change after sprint 2. The answer tells you almost everything about how the agency operates.
Basecamp's Shape Up methodologyhas a genuinely useful framing for this: fixed time, variable scope. It's worth reading before you enter any agency negotiation.
---
AI Tools in 2026: What They Change (and What They Don't)
Everyone wants to know whether GitHub Copilot, Cursor, or the new wave of agent-based coding tools (Devin, Replit Agent) have materially reduced software costs. Honestly? Yes, a bit. But less than the hype suggests.
My rough experience: strong AI-assisted developers are working about 20–30% faster on greenfield CRUD work. That middle chunk of standard features — forms, tables, basic auth flows — does come out faster. But complex integration work, architectural decisions, debugging subtle race conditions, and anything requiring deep product thinking? AI doesn't help there, and in some cases it generates plausible-looking code that introduces new problems.
I'd apply a 10–15% efficiency discount to simple and medium stories in 2026 if the team is confirmed to be using AI tooling seriously. Not more than that. Anyone quoting you 50% lower "because we use AI" is either running a very different kind of project than you think, or telling you something too good to be true.
---
FAQ
How accurate can a software estimate really be before specs are written?
Not very. Expect ±40–60% variance at the idea stage. Once you have detailed user flows, a data model, and a list of all third-party dependencies, you can get to ±20%. After a discovery sprint with wireframes and technical architecture: ±10–15%. Pay for a proper discovery engagement before you commit to a full build budget — it typically costs £2,000–£6,000 and is the best money you'll spend on the project.
Should I go fixed-price or time-and-materials?
Fixed price gives you budget certainty but shifts risk onto the agency — which means they'll pad the estimate and protect themselves with change-order clauses. Time-and-materials is honest but requires you to manage scope actively. My preference for most founders: fixed price per sprint (typically 2 weeks), with scope defined at the start of each sprint. You get predictability without the padding game.
Is nearshore or offshore development actually cheaper once you factor in management overhead?
Often, but not always. The overhead is real — more PM time, more asynchronous communication, occasional rework from misaligned expectations. For a well-scoped project with strong specs, nearshore (Eastern Europe particularly) offers genuine value. I've run successful projects at £40/hr blended with Polish and Ukrainian teams. For something ambiguous and fast-moving, I'd keep it closer to home.
What's a red flag in a software proposal?
A single-line quote with no breakdown. A timeline that doesn't include QA or deployment. No mention of how change requests are handled. And honestly — any agency that doesn't ask you hard questions about your existing infrastructure before they quote. If they're not curious about your constraints, they haven't thought about your project seriously.
How do I budget for post-launch maintenance?
Standard rule: 15–20% of initial build cost per year for ongoing maintenance, bug fixes, and minor feature work. A £50,000 build needs a £7,500–£10,000 annual maintenance budget. If someone tells you software is "done" at launch, find different software people.
---
The logistics founder from November? He went back to his agency, they refactored their working process, and the product shipped in March. It's running well. But he told me he wished someone had handed him a framework like this before he signed anything. So. There it is.