Will Your Practice Software Still Exist in Five Years? A Therapist's Due-Diligence Guide

In April 2025, Woebot's users received an email: the app would retire on June 30, and they had until July 15 to download transcripts of their conversations. After that, everything would be anonymized. Woebot was not a fly-by-night operation. It was one of the most studied products in digital mental health, with roughly 1.5 million lifetime users and $124 million raised. It still disappeared, on ten weeks' notice.

For a therapist, a vanishing tool is not an inconvenience. It can mean losing between-session records mid-treatment, rebuilding intake flows during a full caseload, and explaining to clients why the app you recommended no longer opens. Vendor survival is a clinical continuity question, and it deserves the same due diligence you would apply to any other part of your practice.

A 2026 study makes that diligence unusually concrete. It catalogued 542 digital mental health companies that shut down, went bankrupt, pivoted, or were absorbed, and coded why. The patterns are consistent enough to use as a checklist.

The business model predicts the lifespan

The single most practical finding is that how a vendor charges predicts whether it survives:

• One-time purchase apps died 85 percent of the time. A single payment funds the install, then an operating system update breaks the app and there is no revenue to fix it.
• Freemium products died 62 percent of the time. Free users cost money, and conversion rarely covers them.
• Subscription products died 21 percent of the time, and per-employee-per-month institutional products 14 percent.

The second predictor is who pays. Products paid for by individual consumers died at 53 percent; products paid for by institutions (employers, insurers, clinics) died at 21 percent. And counterintuitively, money did not buy safety: companies that raised over $100 million still died at 25 percent, with Pear Therapeutics falling from a $1.6 billion valuation to a $6 million asset auction in two years.

The third predictor is published evidence. Companies with published clinical data died ten points less often, and when they were acquired, the evidence base was often what the buyer wanted. Evidence is the asset that survives even the company.

Seven questions to ask any vendor

You do not need the vendor's financials to apply this. The signals are visible from outside:

1. Who pays you, and how? Subscription or institutional billing is the survivable pattern. A lifetime deal or a mostly-free product is a warning sign dressed as a bargain.
2. What is your data export path, today, without contacting support? You want your notes, client records, and history exportable in a standard format on demand, not as a favor during a shutdown window.
3. What happens to my clients' data if you close? Woebot handled its wind-down responsibly, with notice and transcript access. Ask the vendor to point at their commitment in writing before you need it.
4. Have you published evidence? Published data predicts survival and signals a company building for clinical buyers rather than app-store impulse purchases.
5. Does your product depend on replacing the clinician? Products positioned as autonomous treatment died at 53 percent in the study, and no payer or regulator rescued the category. We covered this in what 542 dead startups show about AI replacing therapists.
6. How locked in would I be? The less a tool demands you abandon your existing stack, the smaller the blast radius if either side disappoints. Our view on that architecture is in why side by side beats switching.
7. Who else depends on them? A vendor whose customers include institutions and group practices has payers with contracts and expectations, which is a different survival footing than a consumer app riding download charts.

How we answer our own checklist

Fairness requires that we take the test too. Citt.ai charges practices by subscription, with group practices on contracted billing, which is the surviving payer pattern in the data. Client records and notes are exportable, and our data commitments are stated on the trust page rather than negotiated during emergencies. We publish how the platform behaves and where its boundaries are, and the product is built to run alongside your existing systems rather than demanding a leap of faith, which caps your downside as much as ours.

No vendor can promise you a five-year future, including us. What a vendor can do is choose the business model and posture that the failure data favors, and make leaving safe enough that you never feel trapped. Those two properties, more than any feature list, are what due diligence should find.