SalaryRight

AI agents handle your entire salary negotiation — market research, email drafts, call prep, closing. £99 upfront with a money-back guarantee: £5,000+ uplift or full refund.

Comment

Inspiration

Three things came together to push us toward SalaryRight.

First: 55% of working professionals never negotiate their salary. The ones who do average an 18.83% increase. Most people walk away from thousands of pounds in compensation every year because they don't know what to ask for or how to ask.

Second: existing solutions are bad at the price points that matter. A negotiation coach at Levels.fyi costs $1,500+. A "negotiation prompt pack" on Gumroad costs $9. There's nothing in between — no professional-quality service that someone earning £50k a year can actually afford.

Third: the Build with Gemini XPRIZE asks for businesses operated by AI agents, with real revenue, in 90 days. A salary negotiation service is the cleanest possible expression of that brief — a service where AI can genuinely deliver expert-quality work, where customers will pay upfront, and where outcomes are measurable in real money.

So: build it in 90 days. Charge £99. Money-back guarantee. Make expert negotiation help accessible to everyone.

What it does

SalaryRight is a salary negotiation service operated entirely by AI agents.

Customers arrive with a new job offer or an upcoming pay review. They upload their offer letter and answer a few questions. Within minutes, a free analysis tells them exactly how much they're likely leaving on the table — with cited market data from Levels.fyi, Glassdoor, and other sources.

If they upgrade to the paid service (£99 flat fee), a team of specialised AI agents takes over:

  • Market Analyst researches comp at this specific role, company, and location
  • Strategist builds a personalised negotiation plan — which lever to push first, what to anchor at
  • Drafter writes every email to the recruiter in the customer's voice
  • Quality reviews every output before it reaches the customer
  • Negotiator reads each recruiter reply and decides the next move
  • Coach prepares the customer for phone calls with scripts and role-play
  • Closer evaluates the final offer and recommends accept, counter, or walk
  • Risk flags cases needing human attention
  • Refund Adjudicator handles refunds automatically when the £5,000+ uplift threshold isn't met

The customer never has to draft a difficult email themselves. They review what the AI has produced and send it from their own account. Total customer time per case: typically 30-60 minutes spread across 2-8 weeks.

If they don't end up at least £5,000 better off, every penny is refunded automatically. No forms, no haggling.

How we're building it

The architecture is intentionally simple: Next.js 15 (PWA) frontend, Cloud Run for the agent orchestrator, Firestore for case state, Gemini API for the AI work, Stripe for payments. All hosted in Google Cloud.

What's not simple is the agent layer. A multi-agent system has more failure modes than a single chatbot — race conditions, stale context, agents disagreeing, infinite loops. We've designed around this with:

  • Sequential pipelines for most workflows; concurrent execution only where it pays off
  • A Quality gate between every customer-facing output and the customer
  • Structured outputs (JSON schema enforcement) so failures are catchable
  • Explicit case state as single source of truth — agents don't pass context informally
  • Logged decisions for every agent action, human-readable and auditable

Every customer case generates 30-100 logged agent decisions. By submission, we expect to have run roughly 10,000 production agent decisions across all cases. The decision log is both the trust mechanism for customers (they can audit why the AI recommended what it did) and the evidence trail for the hackathon judges.

Model selection per agent: high-capability tier for strategic reasoning (Strategist, Drafter, Negotiator, Closer), balanced tier for analysis (Market Analyst), fast/cheap tier for routine tasks (Intake, Operations).

Challenges we're running into

The single hardest design challenge: how do you build a B2C AI service that doesn't feel like a wrapper around ChatGPT?

Every potential customer asks themselves: "Why pay £99 when ChatGPT is free?" If our product looks like a chat window with a salary skin, we lose that question immediately.

We answered it with five structural defences:

  1. Workflow, not chat. Every interaction is a structured step. No blank text boxes. No free-form prompting.
  2. Real domain data. Our agents query live market sources directly. General-purpose LLMs can't do this with current accuracy.
  3. Persistent state. A negotiation plays out over 2-8 weeks. ChatGPT loses context within a session. We maintain the full case file across weeks.
  4. Outcome guarantee. £99 with refund if the customer doesn't end up £5,000+ better off. ChatGPT doesn't carry financial risk for your results.
  5. Done-for-you, not done-with-you. The customer reviews drafts; they don't drive them.

A second challenge: in 2026, every obvious AI startup idea has 5-15 competitors. We spent the design phase exploring what felt like dozens of candidate markets — memoir services, refund recovery, divorce assistance, medical bill negotiation — and concluded that every space we'd consider has multiple credible AI-native players already. The winning strategy isn't to find an empty market; it's to out-execute in a contested one.

Accomplishments we're proud of

Before writing a line of product code, we built a complete operating manual:

  • 37 design documents covering vision, customer, differentiation, success metrics, product design, technical architecture, distribution, business model, legal posture, and risk
  • 6 self-contained execution milestones for the 90-day build, designed for an AI-assisted CLI workflow where each milestone is a self-contained "what to do next" document
  • A multi-agent architecture specification with 15 named agents and explicit coordination patterns
  • A go-to-market plan with concrete content templates, per-channel strategy, and launch sequence
  • A risk register with the top 10 risks scored, mitigated, and monitored — plus crisis playbooks

This level of upfront design might look excessive for a hackathon. We don't think so. The 90-day window is unforgiving; the cost of building the wrong thing for 30 days is far worse than spending one week designing it right.

What we've learned

The biggest lesson from the design phase: in 2026, execution and distribution matter more than novelty.

Every "obvious" AI idea has credible competitors. The winning strategy isn't to find an empty market; it's to out-execute in a contested one. That changes how we think about the hackathon. SalaryRight isn't competing on "we thought of this first." We're competing on: did we build the thing better, distribute it more effectively, and operate it as a real business with real customers and real revenue? Those are the actual judging criteria.

The second lesson: AI as feature vs AI as workforce is a real distinction. A chatbot is a feature inside a product. SalaryRight isn't a chatbot — it's a business where AI agents do the work humans used to do. That's the architectural shift the XPRIZE is reaching for, and it shows up in product design, in cost structure, and in how customers perceive value.

What's next for SalaryRight

The 90-day build:

  • Milestone 1 (now active, ends 30 May) — foundation, free analysis MVP, infrastructure
  • Milestone 2 (to 14 June) — paid tier MVP, first 10 customers, Stripe integration, full drafting and reply flow
  • Milestone 3 (to 5 July) — phone call prep, role-play, distribution engine, 30-50 customers
  • Milestone 4 (to 20 July) — true multi-agent coordination, Marketing Content Agent, 100 customers
  • Milestone 5 (to 5 August) — viral push (Show HN, Product Hunt, PR), 150-250 customers
  • Milestone 6 (to 17 August) — submission polish, shipped by 14 August, buffer days for final push

Beyond the hackathon: SalaryRight isn't built to be a hackathon-shaped demo. It's a real business that operates after 17 August regardless of prize outcome. The roadmap includes a Premium tier for executive-level offers, regional pricing for non-UK/US markets, a contributor program where customers can opt in to share final accepted offers back to our market data pool, and eventually employer partnerships for outplacement and benefits.

The hackathon prize is downstream of the business working. We'd love to win it. We're building as if we won't.

Built With

Share this project:

Updates