Business9 min read

What Is a Software Development Company?

What a software development company actually does, how the engagement models differ, and how to evaluate one before you sign anything.

UL

Ullass Editorial Team

Editorial Team — Software Development & Digital Products

What a Software Development Company Actually Sells

Underneath the proposal, a software development company is selling one thing: the willingness to absorb uncertainty on your behalf. That is a real and valuable service, because the person who knows what is wanted and the person who knows how to build it are rarely the same person, and neither of them has complete information.

Knowing the model also makes the incentives legible, and the incentives explain most of the disappointing experiences. Three of them are worth stating plainly before you read a single proposal:

  1. Vendors price the risk, you absorb it. Under fixed price, the vendor is paid to be right about things nobody can yet know. Every ambiguity left unwritten in the scope document comes back as a change request.
  2. The industry is mostly an arbitrage business. Agencies employ engineers in lower-cost labour markets, mark them up, and sell certainty. That is a legitimate business model. It is also why "our team has ten years of experience" is a claim rather than a fact, and why the verification steps below exist.
  3. Cheap at proposal, expensive at month seven. A bid that wins on price has usually been priced on the most optimistic credible reading of your brief. Ask what happens when that reading turns out to be wrong.

In practice the work is the familiar sequence — discovery, architecture, build, quality assurance, infrastructure, handover — and the phases in more detail are worth reading regardless of who you hire.

The Four Engagement Models

ModelYou are buyingRisk sits withFitsTypical failure mode
Fixed priceA defined outcome for a defined feeVendor commercially, until a disputeWell-specified builds, internal tools, a prototype with a fixed briefDispute over what was "in scope"
Time and materialsBillable effort against a defined outcomeYou, entirelyDiscovery-heavy work, product iteration, anything with unknownsNo ceiling; scope expands quietly
Dedicated teamA named team working only on your productShared, but you control prioritiesOngoing product ownership over 6+ monthsSkills mismatch that takes months to surface
Staff augmentationIndividual engineers inside your processYou, including hiring and retentionYou have process and need specific skillsCoordination tax paid by you

Staff augmentation is not the same as outsourcing and the distinction matters. In augmentation the person works in your standups, your repository, your on-call rotation, and your employment-at-will decision applies. The vendor is a recruiting intermediary. That is a different product with different accountability.

Fixed Price: How It Actually Works

A fixed price is not a discount. It is a transfer of risk in exchange for a higher unit rate, and it only works when the scope is genuinely fixed. The mechanism is a contract, so the entire relationship is determined by one artifact:

text
SCOPE ITEM 3 — Supplier invoice export In scope: - Export invoices for a single calendar month as CSV - Columns: number, issue date, customer, subtotal, tax, total, paid date - Access restricted to users with the "billing.read" permission - Generated synchronously; max 90 day range - Loading state and empty state per the design system Out of scope: - PDF generation, scheduled email delivery, custom column selection - Multi-entity consolidated exports Acceptance criteria: - A user with billing.read exports 2026-03-01..2026-03-31 and the file reconciles to the on-screen invoice list to the cent, verified on staging - A user without billing.read receives 403 - Response time under 2s for a 90-day range with 50,000 rows Change control: Any requirement not listed above is a change request, priced before work starts. A change that cannot be scoped within 2 working days is escalated, not absorbed.

Three features of that document do the work. The out-of-scope list converts future scope creep from an argument into a lookup. The acceptance criteria are testable by someone who did not write the code. The change-control clause states a process, which is the only thing that works when a relationship is already under strain.

What fixed price cannot do is absorb genuine ambiguity. Here is the arithmetic of what happens when it is sold as if it can:

python
# Fixed-price build quoted at $180,000 over 16 weeks. quoted = 180_000 vendor_budget = 180_000 * 0.55 # target gross margin on delivery true_cost = vendor_budget + 30_000 # overruns funded from vendor margin scope_growth = 0.15 # you add 15% after the contract is signed delta_work = quoted * scope_growth # A change request prices the delta at vendor cost + margin, plus a # re-planning tax you do not pay when the work was planned from the start. change_order = delta_work * 1.45 print(round(change_order)) # 39150 print(round((quoted + change_order) / quoted, 3)) # 1.217 -> 21.7% over contract

So a 15% scope change costs roughly 22% more than the contract price, even when the vendor absorbs the cost overrun without complaint. That is not bad faith; it is the model working as designed.

Time and Materials

Time and materials moves the risk back to you, which is correct when the scope genuinely is not knowable — which is most first-time software work. The weakness is that it has no natural ceiling, and cost overrun reads as failure even when the spend bought exactly what you wanted.

python
# T&M cost model. Blended rate, not the senior engineer's list rate. team = [("engineer", 1, 65), ("engineer", 1, 65), ("designer", 1, 70), ("QA", 0.5, 45)] days, hours_per_day = 22, 8 monthly = sum(count * rate * hours_per_day * days for _, count, rate in team) print("monthly burn:", monthly) # 44540 print("months on a 300k budget:", round(300_000 / monthly, 1)) # 6.7 # A 15% overrun on a 6.7-month engagement is not a rounding error. # Write the ceiling down before month 1, not after month 7.

Three mechanisms keep it honest: a notional budget with a forecast refreshed monthly, a cap on team size so scope growth does not silently become headcount growth, and a velocity-based reforecast at the end of every cycle that states what the next quarter costs. None of these prevent overspend; they make it visible a quarter earlier, which is worth more than the prevention.

Dedicated Teams and Retainers

A dedicated team is a long commitment: your backlog, your standups, your on-call. The commercial question is what happens at month nine, when someone valuable leaves. Ask directly whether the named individuals are employees or subcontractors, whether you have any visibility into secondments, and whether the contract lets you replace a named person without renegotiating.

Most disputes about dedicated teams are not technical. They are about how long a ramp-up is tolerated, who owns knowledge that leaves with a person, and whether the rate is the rate when the composition of the team changes.

The Questions to Ask Before You Sign

Ask these in the sales meeting, not by email afterwards. The quality of the answers is the evaluation.

  1. "Show me the repository, not the portfolio page." A read-only walk through the actual codebase for a similar project tells you more in thirty minutes than any case study. Ask which parts were done by your proposed team.
  2. "What would make this project fail?" A vendor who cannot name two concrete technical or organisational risks is selling certainty they do not have.
  3. "Who is on the team, and what have they shipped?" Names, roles, tenure. Then check those people exist on the company's public profiles or repositories.
  4. "What is the acceptance criterion for this phase?" If the answer is not a testable statement, the vendor does not have a process.
  5. "What did the last project you cancelled go wrong, and who paid?" This is the most revealing question on the list. A vendor that has never cancelled anything has either never been challenged or does not discuss it.
  6. "Which third parties do you depend on, and what happens when they are down?" Payment provider, cloud, package registry. Every dependency is an outage you inherit.
  7. "How do you handle a security review from our side?" The OWASP Top 10 is the common baseline; a confident answer names the artefacts rather than saying "we follow best practice".
  8. "What does IP assignment look like, and when does it trigger?" See below.
  9. "Who owns the infrastructure after launch?" Accounts, domains, DNS, cloud tenancy. This is where handover quietly goes wrong.
  10. "What is the median project you have cancelled, and what was the refund position?" Nobody expects zero cancellations. They expect a straight answer.

Red Flags

  • No engineer will speak to you before you sign. Sales talks to sales.
  • Estimates arrive before requirements. A number produced in week one is a guess with a currency symbol on it.
  • The rate is suspiciously below the market. Someone is being underpaid, the scope is wrong, or the margin depends on change orders.
  • "Agile" with no demo cadence. Describe the last cycle and what got shipped. If the answer involves a slide deck, there is no cadence.
  • The proposal has no out-of-scope section. Everything is in scope, and you will pay for that.
  • Testing is a line item at the end. Delivered in bulk at the end, testing is a phase; delivered continuously, it is a practice. Ask which.
  • No mention of handover, documentation or your access to accounts. A vendor that keeps credentials is not a partner, it is a hostage situation waiting for a contract dispute.
  • Pressure to sign before the reference call. There is always another client.

Verifying Claims Independently

Take three concrete steps. First, run a live code walkthrough with two of your own engineers present. Ask them to explain why the code is shaped the way it is — a team that can defend its own architecture is worth more than one with a nicer case study. Second, call two references of similar size and shape, and ask what was hard rather than what was good; specifically ask when the vendor caused a delay. Third, verify the security posture against something published and specific, such as the NIST Secure Software Development Framework, rather than accepting "we take security seriously".

Then read the contract specifically for four clauses:

text
INTELLECTUAL PROPERTY All deliverables created in the course of this engagement, including source code, design files, infrastructure definitions and documentation, vest in [CLIENT] on payment. Assignment is not conditional on any further agreement. Third-party and open-source components are listed in the SBOM at [url], each under a licence permitting commercial use and redistribution. WARRANTY AND REMEDY 90 days from acceptance. Remedy is repair or replacement within 10 business days, or fee refund for the affected deliverable. No limitation of liability applies to breach of this clause or to confidentiality. SECURITY AND ACCESS Client retains ownership of all cloud accounts, domains, source control and third-party accounts from day one. Contractor access is via named accounts with MFA, removed within 24 hours of a named person's departure. TERMINATION Either party may terminate for convenience on 30 days notice. Client pays for work completed plus non-cancellable commitments. Contractor delivers work in progress in a usable state, including repositories, credentials and documentation, within 5 business days.

The termination clause is the one that is negotiated hardest and matters most. A provider who refuses to deliver current work on termination is telling you, plainly, what a dispute would look like.

The Checklist

Before signing, confirm each of these in writing:

  • Named team, roles, and whether they are employees or subcontractors
  • Written scope with an explicit out-of-scope list and testable acceptance criteria
  • Which artifacts exist at the end of each phase, and who signs them off
  • IP assigned on payment, with an SBOM covering third-party and open-source code
  • Source control, cloud and domain accounts owned by you from day one
  • Security baseline referenced by name — OWASP Top 10, or the NIST SSDF
  • A support and defect window after launch, with a stated response time
  • A termination clause that hands over live work in a usable state
  • A named reference you have actually spoken to
  • A budget ceiling or a reforecast cadence, whichever the model requires
  • A rollback plan for the launch, not just a launch date
  • An operational model after handover — SLOs, alerting and on-call, in the sense the SRE book and SRE workbook use them, not as a slide

If a vendor resists three or more of these, the disagreement is not about process. It is about who is holding the risk when something goes wrong, and you should find out now rather than in month four.

We build this kind of work as a software development company, and the questions above are the ones we put in front of clients during a first scoping conversation. If you would rather see what a finished engagement produced than read another page, the POSNova case study walks through a live deployment. The phases referenced throughout are set out in the development process explained.

How Ullass can help

This guide covers the theory. If you would rather have it implemented and operated for you, these are the services that cover it.

Related Articles

Ullass — Software Development Company

We build web apps, SaaS platforms, and digital products

Ullass designs, engineers, and scales software products for ambitious businesses. Also try our free online tools at tools.ullass.com. Questions? hello@ullass.com