From AI Prototype to Production: How Savannah Became a Platform Its Sales Team Can Sell
Massive kudos to you and your team. Outstanding.
Tom Chapman, Advisor to the Savannah board![[object Object]](/images/customers/logo-savannah.png)
See the impressive results
- Took an AI recruitment platform from AI-built prototype to production, ready for the scrutiny of enterprise customers
- Delivered the full feature build inside a transparent estimate, with AI-assisted effort shown against the traditional equivalent and approved before work began
- Proved the experience at production scale before launch, loading 50,000 anonymised candidates, 100 jobs and five million match-score rows and testing that reviewing candidates stays fast at full volume, not just that the backend holds
- Re-engineered the AI shortlisting on those measurements until results were fast, stable and repeatable
- Completed the transition ahead of schedule under a fixed-price statement of work, formally closed with a closure package
- Independent security firm penetration-tested the delivered platform end to end, web, mobile, APIs and the AI agent, with zero infrastructure issues and no high or critical findings
- Built a contextual help system, help centre and in-product alerts so the product explains itself
- Moved straight onto base2Services 24/7 managed operations with continuous security assessment through Secure Compass, roadmap still moving
A bit about Savannah
Savannah is an AI recruitment platform born inside a specialist recruitment business. AI shortlisting, a mobile-first candidate experience built around video rather than resumes, and a real head start from being built fast with AI-assisted development. The product was capable and the idea proven with customers.
What the founders wanted next was bigger: enterprise customers, investors and scale.
What Savannah needed
Savannah needed engineers who could evaluate what had been built, match it to what the business needed it to become, and close the gap between the two: the requirements their largest prospects would test, performance at real data volumes, an AI layer they owned rather than rented, and a team to design, recommend and build alongside their own.
Their destination was a production platform their sales team could sell with confidence. The plan had to get them there without losing what already worked.
How base2Services delivered
Evaluate first
An architecture review, then a fixed-scope code and IP review, established the ground truth: the platform architecture was solid and worth keeping, while the AI layer relied on third-party services and could be re-engineered without disrupting the rest. That distinction, keep what is sound and re-engineer only what needs it, shaped the whole engagement.
Match to requirements
The feature catalogue and an 18-month roadmap were worked through with the leadership team, tested against the customers Savannah wanted to win. We wrote the questions their toughest prospect would ask and planned against them, and roadmap choices were made with the obligations of their market in view, including walking away from features the EU AI Act prohibits.
Built correctly, and provably
The work ran in stages under a signed statement of work: codebase brought under controlled ownership, CI/CD pipelines stood up, versioning and release discipline introduced, new infrastructure built. Correctness was demonstrated, not asserted. The platform was loaded with 50,000 anonymised candidates, 100 jobs and five million match-score rows, and the real test was not the backend, which was engineered to hold. It was the experience: could a recruiter load, search and review candidates at that volume and still move quickly? Demo data never answers that question, and the prototype had never faced it. The review experience and the AI shortlisting were re-engineered on those measurements until both were fast, stable and repeatable, and quality assurance moved from anecdotal to quantitative.
Features delivered, not just fixed
This was a build, not a clean-up. The feature catalogue was worked through to completion for launch: the candidate and recruiter experience finished across web and mobile, a contextual help system, help centre and in-product alerts built into the product so it explains itself, white-labelling capability scoped for multi-brand growth, and the AI sourcing pipeline redesigned for the roadmap ahead. Each feature was demonstrated live in a weekly walkthrough as it landed.
Professional to the week
Every Monday, a written progress report with the hours attached. Every week, a live demo of working software rather than a status slide. Readiness tracked as numbers against named lists, blockers counted down in the open. Estimates given before work started, AI-assisted effort shown against the traditional equivalent, and held. The transition was formally closed with a closure package, ahead of schedule, and handover between our own leads was planned and announced rather than discovered.
Shipped proven
Before go-live, an independent security firm penetration-tested the platform we delivered: web application, mobile app, APIs and the AI agent. Zero infrastructure issues, no high or critical findings, remaining minor items tracked to closure. Incident readiness and service governance were in place before launch day.
Kept running
Savannah moved onto base2Services 24/7 managed operations under a single services agreement, with continuous security assessment through Secure Compass, and the roadmap, voice interfaces next, continuing under a standing engagement structure.
This is first class… massively helpful.
Paul Garrett, Chief Executive Officer, Savannah