Table of Contents
- Why standard agency evaluation produces the wrong shortlist
- What actually predicts a strong software and AI partner in London
- What separates a real AI partner from an agency that added AI to its website
- When a software and AI partner in London is not what you need
- A five-step evaluation process that produces better decisions
- The six questions that reveal an agency’s real character
- Frequently Asked Questions
- Conclusion: the decision that compounds in both directions
Quick answer: Foundry 5 judges a software and AI partner in London on four observable signals: how discovery is structured, how scope change is handled commercially, who runs your project day to day, and whether post-launch support exists in writing. McKinsey found large IT projects run 45% over budget. Process predicts delivery, presentation does not.
You have shortlisted three agencies. All three sent proposals within 48 hours. All three used the same words: end-to-end delivery, agile methodology, scalable architecture. All three showed polished screens, confident timelines, and client logos you half recognise. Now you are holding three documents that read like one person wrote them, facing a decision worth somewhere between £60,000 and £300,000, and you cannot tell them apart.
This is the real problem for London founders, CTOs, and operations directors in 2026. Not a shortage of agencies. Not even a shortage of competent ones. The problem is that the evaluation process most businesses run is designed to find agencies that present well, rather than agencies that deliver well. In software, that gap is the difference between a platform that drives growth and a rebuild that eats your next 18 months.
The evidence on what goes wrong is unambiguous. In research with the University of Oxford covering more than 5,400 IT projects, McKinsey found large IT projects run 45% over budget and 7% over time while delivering 56% less value than predicted, with software carrying the highest overrun risk of any category. Those failures are decided long before the first commit. They are decided at selection.
What follows is a decision framework built on evidence rather than first impressions: what a genuine partner looks like under the surface, which questions separate performers from presenters, and where conventional hiring advice will cost you more than it saves.
Why standard agency evaluation produces the wrong shortlist
Most businesses choose a software partner the same way: review the portfolio, read the proposals, take a discovery call, check some reviews, pick whoever felt most confident. Each step has value. None of them predicts delivery quality with any reliability, because every one of them measures salesmanship.
Portfolios show finished outputs, not the path to them. A case study with a beautiful interface and a happy client quote tells you nothing about whether the project ran six weeks late, whether scope was quietly cut to hit a date, or whether the relationship survived the month after launch. Portfolios are curated by the agency to show its best work in its best framing. That is marketing, rather than evidence.
Proposals reward proposal-writing skill. The agencies investing most in sales, polished decks, rapid turnaround, granular breakdowns, are demonstrating sales capability. Useful, but not the same as build capability. Teams that are exceptional at winning clients and mediocre at keeping them produce impressive proposals every single time.
Discovery calls are built to grow confidence and lower resistance, not to surface anything disqualifying. Ask yourself: across the last round of calls you took, how many agencies told you they were not the right fit? None did. Identifying genuine misfit is not what a sales call is incentivised to do. The best way to assess a partner is not to watch how they present. It is to probe how they think.
What actually predicts a strong software and AI partner in London
Choosing a software and AI partner in London comes down to four observable signals: the structure of their discovery process, how they handle scope change commercially, who manages your project after signing, and what their post-launch model says in writing. Foundry 5 uses the same four. Partners strong on all four are predictable. Partners strong on two create risk exactly where it hurts.
Discovery: the strongest single predictor of outcome
Discovery is where technical constraints surface, where business requirements become architecture decisions, and where the assumptions governing every later sprint get made explicit. A compressed discovery does not save time. It defers cost into a stage where changes are exponentially more expensive.
Ask every agency to walk you through discovery step by step, from signing to first sprint. Answers split cleanly into two categories. The first describes a documented process: stakeholder interviews, technical architecture review, user journey mapping, and a written specification both parties sign before code is written. The second describes something impressionistic: a few calls, a shared sense of direction, and then we begin. The second category is where the overrun statistics come from.
Consider a Shoreditch SaaS startup that signed with an agency describing discovery as an intensive kickoff week. Seven weeks into development, the team found the data model could not support the multi-tenancy the product needed at any real scale. Redesigning mid-build cost £38,000 and pushed launch back eleven weeks. Proper discovery would have caught multi-tenancy in the first conversation. Watch for agencies offering to discount the discovery fee to win the signature. That is not a favour. It is a warning.
Scope change: how they handle it reveals the operating culture
Every software project changes scope. That is not a failure, it is what happens as the picture gets clearer. The question is never whether scope will move. It is what happens to the relationship, the timeline, and the commercial terms when it does.
Scope movement is the norm, not the exception. In its Pulse of the Profession research, the Project Management Institute found 52% of projects experienced scope creep, up from 43% five years earlier. Any agency claiming your project will be the exception is either inexperienced or managing your expectations rather than your risk.
Ask for a specific example of a project where scope changed significantly mid-build. How was it communicated? Who owned the cost impact? How was the timeline adjusted? Mature agencies answer with detail and ownership. Weak ones give a vague answer, redirect to contract terms, or quietly blame the client. The contract model matters here too, and the honest trade-offs between fixed price and time and materials are worth understanding before you sign either.
Day-to-day contact: the team that sells is rarely the team that builds
This is the most consistent and costly failure pattern in the London market, and almost nobody asks about it before signing. The senior partner closes the deal. A mid-weight manager runs the engagement. Junior developers write the code. Three sprints in, the person who convinced you has moved to the next pitch and you are working with a team you never evaluated.
Ask directly: who is my primary contact once we begin, can I meet them before we sign, and how many other engagements will they run alongside mine? The partners worth hiring answer with names rather than titles, and put the delivery team in front of you before contracts are exchanged. Agencies that defer this, answer in organisational abstractions, or treat team access as a concession are showing you how they will behave all the way through. That pattern does not improve once the invoice clears.
Post-launch: a partner who leaves at launch is a contractor
Software is not a deliverable. It is the start of a system that needs iteration and maintenance as real users meet it. An agency treating launch as the endpoint is a contractor with a handoff clause, rather than a partner.
Ask for the post-launch model in writing: what the standard retainer includes, what costs extra, the response commitment for critical production issues, and what happens when a feature they built is not driving the behaviour it was meant to. Picture a Canary Wharf logistics business that launched an operations platform with no structured post-launch arrangement. Three weeks after go-live, a routing calculation error was mispricing roughly 9% of orders. The agency, off contract, quoted four weeks at day rates. By deployment, 600 orders had gone out wrong and two enterprise relationships needed repair. Post-launch support is not a premium extra. It is the difference between a platform that compounds and one that erodes.
Already know which of these four signals worries you most? Talk it through with Foundry 5 in 30 minutes, no deck and no obligation, or keep reading for the AI-specific tests.
What separates a real AI partner from an agency that added AI to its website
The London AI market expanded fast between 2023 and 2025, and much of that expansion was nomenclature rather than capability. Conventional software agencies added AI services pages, hired one or two specialists, and started pitching AI transformation to clients who deserved a more honest conversation.
Adoption is not the same as competence. In its State of AI research, McKinsey found 65% of organisations were already using generative AI regularly, roughly double the share a year earlier. When everyone has adopted something, the label stops carrying information. You have to test the substance underneath it.
The real question is not whether an agency can build with AI tools. Most can, to some degree. It is whether they treat AI as a different category of delivery, one where data quality, model evaluation, failure-mode design, and ongoing monitoring matter as much as the code. Agencies that understand this scope the work around business outcomes. When you describe a problem, they ask what decision you are trying to improve, what data you hold and in what state, how you will measure whether it works, and what an incorrect output costs you. Agencies that do not understand AI ask which model you want and how many integrations are in scope.
Consider the data question specifically, because AI systems perform in proportion to the data behind them and most London businesses are less data-ready than they believe. Picture two companies starting similar builds: one runs a structured data audit first and finds its records inconsistent across three systems, the other skips straight to model selection and discovers the same problem in month five, after the architecture is set. Same destination, wildly different cost of arrival. That is an illustration rather than a study, but the pattern is one every experienced team recognises.
So ask any AI agency what their data assessment looks like before a build begins, and what they have actually found when they have run it. Then evaluate their portfolio for business outcomes rather than technical demonstrations: reductions in manual processing time, accuracy rates under production conditions, cost savings attributable to the AI component. Agencies describing AI work in purely technical terms are building systems. You need a partner who understands the difference.
When a software and AI partner in London is not what you need
Intellectual honesty requires saying this plainly: hiring a software and AI partner in London is not always the right call. Foundry 5 turns down work that fits a cheaper model better, because three specific situations are served well by something other than a premium London agency, and a partner worth hiring will tell you so.
If you are pre-revenue and building a first MVP to test one core assumption, a London agency billing £700 to £950 per developer day is almost certainly the wrong structure. That capital is better deployed after the assumption is validated. A well-specified build with a strong offshore team or a senior freelancer can reach validation for a fraction of the cost, and our freelancer, offshore, and London agency comparison maps where each one genuinely wins. That is not a compromise. It is appropriate capital allocation.
If your internal team already has strong technical leadership and you need execution capacity rather than direction, staff augmentation will outperform a full-service agency. Agencies charge for strategy, architecture, and delivery combined. If you already have the first two, you are paying a premium for a layer you do not need. The same logic applies to a straight migration with no material change to the underlying logic: a specialist migration partner beats a generalist at lower cost.
The honest framework: a London partner makes sense when you need strategic contribution alongside delivery, when you want someone owning outcomes rather than scope, and when the cost of getting the architecture wrong exceeds the premium you pay to get it right first time.
A five-step evaluation process that produces better decisions
Run every candidate through a structured process rather than trusting impressions formed in one call. Five steps, in order, each designed to surface information a sales conversation is built to hide.
- Send a written brief before any call. Cover business context, the problem, timeline, budget range, and how you will measure success. Partners who reply with specific, probing questions have read it. Partners who reply with a generic deck are showing you their default.
- Ask for a case study in your problem category, not their general portfolio. Structurally similar, verifiable results, relationship intact beyond launch. If they cannot produce one, you have your answer before the proposal stage.
- Request a paid technical scoping session before committing. It produces a written technical approach and a realistic estimate, and it is the closest thing to real project work you can buy before signing.
- Speak with two clients you identify yourself, not references they select. Ask what surprised them, what they would do differently, whether final cost matched the estimate, and whether they would hire again.
- Evaluate the proposal for evidence of listening. If it could have been written before you spoke, they were not listening. Listening quality in the sales process predicts listening quality in delivery.
Step three deserves particular weight. A paid scoping session costs a fraction of a build and reveals how a team thinks under conditions that resemble the real thing. Agencies confident in their process offer it readily. Agencies selling certainty they cannot back tend to resist anything that tests them before the contract is signed.
The six questions that reveal an agency’s real character
Six questions cut through presentation quality and surface delivery quality. Ask all six on every first call, then compare answers side by side rather than agency by agency. The gaps between them will tell you more than any portfolio.
- How do you structure discovery, and what document comes out of it? This reveals whether discovery is a real investment or a formality on the way to signing.
- Walk me through a project where something went wrong mid-build and how you handled it. Every serious agency has one. The ones worth hiring tell you what they owned and what they changed.
- Who is my day-to-day contact after we sign, and what is their current workload? This separates responsible staffing from oversold senior access and junior execution.
- What does your post-launch model include, and what sits outside the standard retainer? Specificity reveals whether it is an operating model or proposal decoration.
- How do you handle scope change commercially, and can you show me the contractual mechanism? Mature agencies have solved this and can show you in writing without hesitation.
- What is the most common reason a project like mine underperforms, and what does your process do to prevent it? This tests pattern recognition, and pattern recognition is what experience actually buys.
That last question is the one that separates the field. Experts see patterns beginners cannot. The best teams will name precisely what kills projects in your category and explain how their process is built to prevent it. That answer tells you more than any case study, because it cannot be prepared in advance by someone who has not lived it. For the wider checklist, our guide to evaluating a software development agency covers the contractual questions that follow these six.
Working through a shortlist right now? Bring your brief and the three proposals, and Foundry 5 will pressure-test them against this framework in a 45-minute project feasibility review. Book a free feasibility review We will tell you honestly if we are the wrong fit and point you to someone better. No pitch, no obligation, two minutes to schedule.
Frequently Asked Questions
How much does it cost to hire a software and AI development company in London in 2026?
London agencies typically charge £600 to £950 per developer day for mid-weight engineers, and £900 to £1,400 for senior architects and technical leads. A mid-market custom platform runs £75,000 to £280,000 depending on scope and integration complexity. AI builds carry roughly a 15% to 30% premium over equivalent conventional projects, because of the extra data engineering, model evaluation, and monitoring work involved. Scope moves the number far more than the day rate does.
What should I look for when choosing a software agency in London?
Prioritise four things above everything else: a documented discovery process, a clear commercial mechanism for scope change, named delivery staff you can meet before signing, and a written post-launch model with defined response commitments. Foundry 5 assesses any software and AI partner in London the same way. Agencies strong on all four deliver more predictably than those strong on two or three, and the difference shows up in month eight rather than month one.
How long does a custom software build typically take in London?
A well-scoped MVP takes 12 to 20 weeks from kickoff to first production deployment. A full custom platform with complex integrations and AI components runs 6 to 18 months depending on scope and team structure. The single biggest variable is discovery: projects that compress it consistently overrun their original estimate, because the decisions skipped early resurface later when they cost far more to change.
How do I know if an agency’s AI capability is genuine?
Ask them to describe the data requirements for a project like yours before any technology comes up. Genuine AI teams raise data quality, volume, and structure as the primary constraints ahead of models or integrations. Then ask for a case study where an AI system is producing measurable business outcomes in production, and what their process is when a model returns incorrect output in a live environment. Those three answers separate capability from a services page.
When does it make sense to choose a London agency over an offshore team?
Choose a London partner when the project needs real strategic contribution alongside delivery, when post-launch iteration is part of the model rather than an optional extra, and when the cost of an architectural mistake in the first build exceeds the premium you pay to avoid it. For early-stage MVP validation, a strong offshore team or senior freelancer is often the better use of capital at that stage.
Conclusion: the decision that compounds in both directions
This framework is not built to find you the cheapest option or the most impressive one. It is built to find the right one, and Foundry 5 argues that choosing a software and AI partner in London is a decision about process rather than presentation. The partner whose discovery, delivery culture, and post-launch model match what your project actually needs is rarely the one with the best deck.
Software infrastructure compounds in both directions. A platform built on sound architecture by a team that understands your business enables every decision that follows it. A platform built quickly by the wrong partner constrains every decision that follows it. That difference is invisible at proposal stage. It becomes visible in months eight through eighteen, when you are either accelerating on a foundation that scales or managing a rebuild on one that cannot.
The five steps and six questions here will not guarantee a perfect outcome. Nothing does. What they will do is replace impression management with evidence, and move your decision from which agency presented best to which one has genuinely solved this problem before. If you want to run your specific project through this framework before committing to anyone, book a free 45-minute project feasibility review with Foundry 5. No pitch. No obligation. Just a direct read on what your project needs.
Choose on evidence, not on presentation.
