The premise sounds almost too simple: ask someone to listen to a short story, then retell it. Ten minutes. A standard web browser. No lab, no specialist, no needles.
Yet somewhere in those rambling retellings—the pauses, the forgotten details, the slight unraveling of syntax—Emil Fristed and Jack Weston believe they've found something that has eluded neurologists for decades: a scalable way to catch Alzheimer's disease before memory clinics and brain scans confirm what families already fear.
In July 2020, their London-based startup Novoic raised £2 million in seed funding to prove it. Anthemis Exponential Ventures led the round, joined by APEX Ventures, Entrepreneur First, Plug and Play Ventures, and a handful of angels through Stanford Angels & Entrepreneurs. The company called it "strongly oversubscribed"—though in venture parlance, that phrase gets tossed around liberally enough to warrant mild skepticism.
Still, the timing was revealing. Even then, as the world fixated on a very different kind of health crisis, a quieter shift was underway in Alzheimer's research.
When Detection Becomes the Bottleneck
Fristed and Weston, both products of Oxford and Cambridge's neuroscience programs, launched Novoic in 2019 around a straightforward insight: speech patterns degrade before traditional cognitive tests catch them. Not dramatically. Subtly—the linguistic equivalent of hairline fractures.
Their flagship product, Storyteller, administers what they call an Automated Story Recall Task. Patients hear a narrative, repeat it back, and Novoic's algorithms dissect the response for markers invisible to human ears: degraded memory recall, fractured syntax, diminished semantic richness.
It's the kind of technology that sounds futuristic until you consider that speech has always been a diagnostic tool—doctors just couldn't scale their own pattern recognition across millions of at-risk adults. Automating the ear, in essence.
By 2022, peer-reviewed validation arrived via JMIR Aging, showing the test reliably distinguished mild cognitive impairment and early Alzheimer's from healthy controls. Later that year, Novoic reported that its models detected amyloid positivity—a biological hallmark of Alzheimer's that appears years before symptoms—and outperformed the standard PACC5 cognitive composite used in U.S. and UK clinical trials.
Those results mattered. Not because they were perfect—no early-stage diagnostic ever is—but because the healthcare system was about to need them.
The Drug Approval That Changed Everything
Fast forward to 2023. The FDA granted traditional approval to Leqembi, Eisai and Biogen's antibody therapy targeting amyloid plaques. Medicare coverage expanded. For the first time, Alzheimer's patients had access to a treatment that actually slowed cognitive decline, not just masked symptoms.
Eli Lilly's donanemab followed in July 2024 with similar results. Both drugs share a critical limitation: they work best—perhaps only meaningfully—in early-stage disease. Patients with advanced Alzheimer's gain little. The treatment window is narrow.
Which creates an uncomfortable math problem. Millions of Americans likely harbor early amyloid buildup but show no symptoms. Traditional screening—PET scans, spinal taps—costs thousands per patient and requires specialized facilities. You can't screen an entire aging population that way, not economically, not logistically.
Remote speech tests that run in web browsers? Maybe you can.
Anthemis Exponential Ventures clearly thought so. In its post-investment thesis, the firm positioned speech biomarkers not just as research tools but as early risk indicators for insurers and health systems grappling with long-term care costs. APEX Ventures, the digital health-focused arm of the larger APEX fund, echoed the sentiment: decentralized trials and value-based care models need precisely this kind of scalable, objective measurement.
Follow the incentives, and you see why Novoic's oversubscribed round probably wasn't marketing spin.
From Lab to Real World
Since that 2020 raise, Novoic has embedded itself in the Alzheimer's research infrastructure. The company now supports over 50 research institutions, according to its website—a claim that's harder to verify independently but aligns with the proliferation of digital cognitive assessments across academia.
More tellingly, Novoic's Storyteller test was deployed in ADNI4, the fourth phase of the Alzheimer's Disease Neuroimaging Initiative, a massive longitudinal study tracking cognitive decline. Results published in late 2024 showed the test generalized across diverse populations, addressing one of the field's most persistent challenges. Alzheimer's research has historically suffered from homogenous study cohorts; validation across demographics matters if this technology ever scales beyond clinical trials.
Corporate filings from 2024 show fresh share allotments, suggesting additional capital or equity grants, though Novoic hasn't announced follow-on rounds publicly. That silence could mean anything—negotiations in progress, strategic partnerships being inked, or simply a preference for staying out of the funding news cycle.
The Wider Field Is Watching
Novoic isn't alone. The Alzheimer's Drug Discovery Foundation launched SpeechDx in early 2025, a dataset initiative aimed at standardizing speech biomarker research. Siemens Healthineers licensed it shortly after—a signal that established medical device players see commercial potential.
Which raises the question Novoic will eventually face: does speech analysis remain a research tool, or does it become clinical standard of care?
The answer depends partly on continued validation—more studies, more populations, more evidence that a ten-minute browser test can reliably stratify risk. But it also depends on something murkier: whether healthcare systems, insurers, and regulators are prepared to screen millions of people who don't yet know they're at risk.
That requires infrastructure. It requires reimbursement codes. It requires clinicians comfortable acting on algorithmic risk scores, and patients willing to learn they might develop Alzheimer's in a decade when interventions remain imperfect.
Perhaps more than Fristed and Weston anticipated when they closed that seed round in summer 2020, the technological problem was the easy part.
