sense_ was an internal venture inside a multi-award-winning London architecture practice: a bet that the same firm which designed great spaces could build the technology to measure and optimise them. I joined as Product Director from inception, made the hardware bet, built a cross-functional team of fourteen, and reached a £30m+ contract pipeline with six blue-chip pilots and no dedicated sales function.
Commercial real estate is one of the world's largest asset classes, yet it has almost no performance data. At any given moment, over a third of a building's space sits empty. In central London, every 1,000 sq ft of leased space represents roughly £1m over a ten-year lease, a decision made almost entirely on intuition. The sensor technology available to address this was crude, expensive, or locked into proprietary systems that couldn't talk to each other.
spacelab_, a multi-award-winning London architecture studio, saw the gap and recruited me to fill it. We set out to build what you might call Google Analytics for physical space: a platform giving real estate owners the occupancy, energy, and environmental performance data their buildings had always lacked. The core challenge was a hardware bet no off-the-shelf sensor could meet at the accuracy and price point required, so we built one. By early 2020 we had six live pilots with blue-chip clients, a £30m+ TCV pipeline, and our V2 sensor shipping an embedded ML model.
Then COVID emptied every office overnight. The pivot was real, but so was the learning: sense_ was my first serious encounter with deployed machine learning, and it set the direction for everything I've built since.
spacelab_ was seventeen years into delivering award-winning workplace projects for large corporates when the founders spotted a gap. Their clients had deep knowledge of how to design space, but almost none of how space was actually used. The occupancy data simply didn't exist.
The scale of the problem was striking. At any given time, over a third of a building's space sits empty. Buildings are responsible for nearly 40% of annual global greenhouse gas emissions, and the majority of that waste flows from a basic information gap: no one knows where people are, so no one can regulate temperature, air quality, or lighting against real demand. In commercial real estate, the consequence was both financial and environmental. A 10,000 sq m London office might carry a £10m+ lease, renewed on the basis of gut feel and walk-throughs because no analytics existed to say how the building was actually performing.
The opportunity was to build the measurement layer that didn't exist. spacelab_'s seventeen years of client relationships, deep understanding of workplace design, and trusted position with property directors gave it a route into the market that a pure technology startup would have struggled to open. The question was whether a design firm could make a credible technology bet.
Three legitimate problem spaces competed for limited resource. Choosing the wrong one to anchor on meant building a product the market wouldn't pay for. Choosing the right one meant making a hardware bet we weren't sure we could win.
Property directors could immediately quantify the cost of running a building at a third of capacity. In London, every 1,000 sq ft of underused space is roughly £1m wasted over a ten-year lease. The ROI story was quantifiable, immediate, and sat at CXO level, not just with facilities managers.
Solving occupancy meant making a hardware bet. No off-the-shelf sensor would do it at the accuracy and price point required, and at a privacy-compliance standard suitable for the workplace. We would have to build one, an unfamiliar discipline for a team inside a design firm with no hardware background.
CO2 levels above 900ppm reduce concentration by up to 15%; above 1,400ppm the drop reaches 65%. Most facilities managers had no visibility of air quality in real time. Conference room availability, temperature management, and sensor-driven HVAC control were all unaddressed. The day-to-day friction was real.
The buyer for comfort management was a facilities manager, not a CXO. Budget authority was lower, and the ROI story was harder to quantify directly on a balance sheet. The problem was real, but it didn't justify the level of investment required to build from scratch when occupancy offered a clearer path to value capture.
Commercial buildings account for 10% of the UK's greenhouse gas emissions, and businesses were estimated to be missing a £1.6bn cost-saving opportunity through energy inefficiency. Carbon footprint was moving from optional reporting to a regulatory obligation, and no granular measurement layer existed at the building level.
In 2018-2019, sustainability reporting was a growing agenda but not yet a financial forcing function for most property directors. The ROI case was real but longer-term and harder to tie to an immediate balance sheet decision. It was the right extension, not the right starting point.
spacelab_'s identity was built on human expertise and judgment. The proposition that a sensor platform could tell a client what a talented design team had previously interpreted through site visits and workshops raised uncomfortable questions internally: was sense_ complementing the design practice or beginning to replace it?
Getting alignment required building a product vision that held both the design heritage and the technology ambition. The framing that worked was that sense_ surfaced the data; the design team turned it into insight and action. The sensor platform couldn't read a space. It could tell you how a space was being used. What to do about it still required expert judgment. This kept sense_ positioned as an enhancement to the practice, not a threat to it.
In retrospect, this framing was also commercially correct. The combination of real-time data and spacelab_'s seventeen years of workplace knowledge was the differentiator no pure sensor vendor could replicate.
Two bets defined the product: build a proprietary sensor rather than wait for off-the-shelf technology to catch up, and build the platform mobile-first for the buyers who would actually use it.
The property director buyer had fifteen minutes, not fifteen hours. Building for a dashboard they'd sit in front of at a workstation was building for the wrong context. The person with authority to act on the data was often on-site, walking a building, or reviewing portfolio performance between meetings. Mobile-first was a bet on where the buyer's attention actually was, not where a typical B2B SaaS product would have assumed it to be.
The platform surfaced real-time and historical occupancy data, CO2 levels, energy consumption, and carbon footprint. Alongside it sat the UPV (Unutilised Potential Value) space modelling tool, which calculated the monetary cost of unused workspace in terms the CFO would recognise: this floor is running at 40% occupancy, and the lease cost of the unused space is £X over the remaining term. Making the insight legible to a property director in a fifteen-minute walkthrough was the design brief.
Peak and average occupancy across all floors and space types. CO2 levels measured continuously. Carbon footprint surfaced. Heatmaps showing how the space is actually used, updated in real time.
Utilisation data showing which spaces are under and over capacity. The Unutilised Potential Value (UPV) tool translates occupancy data into financial terms: the monetary cost of unused workspace over the remaining lease term.
A knowledge layer drawing on spacelab_'s seventeen years of workplace evidence. Surfaced key facts, trends, and findings from across the client base to give each customer context for their own data.
Profile data store letting customers update their space configuration to keep occupancy benchmarks accurate. Foundation for the personalised insights the platform delivered at scale.
Design partners were recruited before a single pilot ran at scale. Every organisation willing to deploy the technology in exchange for early access and a seat at the product development table became a named reference. The process was structured: a visioning session to establish business requirements, a 3D scan to map the space, a six-week live deployment with bi-weekly client workshops, and a structured insight report designed to make the case for full deployment. Every design partner converted to an early customer.
In March 2020, the core use case, optimising how space is used by the people in it, became temporarily redundant. The pipeline that had taken eighteen months to build went quiet. The question was whether the underlying product had a different job to do.
The pivot wasn't purely defensive. The pandemic made building health acutely visible, and sense_ already had the sensors. We moved quickly to build features around COVID-safe capacity management: real-time headcount against compliant limits, flow rate monitoring, and alerts directing people to underutilised spaces as buildings began reopening. It was a genuine product response, not just a messaging shift. But it was adaptation under pressure rather than the natural extension we'd planned.
The core strategic decisions paid off: the hardware bet, the design partner approach, and the mobile-first platform all worked. The harder lessons were about who the product was actually for once it was in the field.
Choosing space efficiency over comfort or sustainability as the primary problem was validated quickly. Every conversation with a property director lit up around the cost and utilisation story in a way that comfort management never quite did. We didn't lose a single design partner through the pilot process.
Recruiting design partners before building at scale was the right call. Real environments, real feedback, real names to reference with subsequent prospects. The 100% conversion rate was the signal: if the product is right, the pilot sells itself. If it isn't, you find out before you've scaled anything.
Platform usage data surfaced something we hadn't fully anticipated: the most engaged daily users were facilities managers, not the property directors who signed off the subscription. The ROI story that justified the spend lived at CXO level; the day-to-day value was operational and granular. We tried to serve both without fully committing to either.
The iteration from breadboard thermopile prototype to embedded ML camera took longer and consumed more resource than planned. Firmware complexity is a different discipline from platform development, and managing both tracks simultaneously with largely separate contracted teams created coordination overhead we hadn't fully priced in.
I'd have moved faster to surface the buyer-user tension and built more explicit product separation between the CXO value layer and the FM operational layer, possibly as distinct tiers rather than a single dashboard attempting to serve both. And I'd have built more buffer into the hardware roadmap for firmware complexity, which consistently surprised us despite being predictable in retrospect.
sense_ is the case study I return to most often when thinking about what product leadership actually means in an organisation making its first technology bet. The conditions were genuinely difficult: a professional services firm whose identity was built on human expertise, a product requiring proprietary hardware we had to invent, three legitimate problem spaces competing for limited resource, and a market that was upended at the moment of maximum momentum.
The fact that we reached blue-chip pilots, a £30m+ contract pipeline, and a working embedded ML sensor with no dedicated sales team reflects what's possible when the technical and commercial case is made carefully and the right design partners are recruited early. The harder lesson is about what product leadership looks like when the organisation isn't built for technology: the job is as much about building confidence as building the product.
sense_ was my first real encounter with machine learning in a deployed, physical context: not a dataset or a proof of concept, but an embedded model making inferences in real buildings with real consequences for accuracy. Working at that intersection of hardware and intelligence sparked a fascination with what machine learning could do and where it might go. That curiosity carried forward from the thermopile sensor on a ceiling in Waltham Forest through to the data services and ML work at Utility Warehouse, and eventually into the agentic systems and LLM work I do today. The thread runs directly, even if the medium changed completely.