I cancelled twelve SaaS subscriptions in one evening. Four months later I haven't gone back to a single one
In January, accounting showed me $9,760 a year in software. I opened the list and cancelled two thirds in one sitting. What I cut, what stayed, and why.
In January 2026 our accounting team sent the annual expense report — the usual thing, end of fiscal period, taxes, all that. I scrolled through it. Page three: the "Software & Subscriptions" line. Total: $9,760.
Nine thousand seven hundred sixty dollars a year on tools that I — one human in one small company — use to do my job. That's $813 a month. For comparison, that's more than I pay for rent in one of the cities where my team and I worked over the last two years.
I opened a spreadsheet, sat down, and listed every subscription. Came out to eighteen active, plus four I'd forgotten to cancel (including one at $39 a month I hadn't touched since last May — a separate kind of pain). Then I did what I'd been meaning to: tagged each subscription with one of three labels — "opened this week", "opened this month", "haven't opened at all".
Opened this week — five. Opened this month — four. Haven't opened — nine.
Nine subscriptions I cancelled that evening. Three more two weeks later, after a short "what if I was wrong" test. Out of those twelve, I have not gone back to a single one in four months. Not one.
First, the ones I cut, and why. The point isn't that the tools are bad. Most of them are excellent. The point is that I had stopped using the function I subscribed for. And the shift happened quietly, in the background.
Calendly Pro, $20/mo. Subscribed two years ago when I was doing 8 meetings a day. Now I do 2-3, and Cal.com (free, open source) covers them with room to spare. Minus $240/year. Nobody noticed.
Grammarly Premium, $30/mo. A separate story. I still write in three languages, and Grammarly was once indispensable for English. But in 2026 I drop in a piece of text and ask Claude to clean it up — works better, because Claude reads meaning, not just grammar. Grammarly went invisible. Minus $360/year.
Otter.ai Premium, $20/mo. Call transcription. Stopped using it the moment it became clear that after a call it's better to scribble five bullets myself than read a thirty-minute transcript that's 95% filler. Transcription helps when you weren't on the call. If you were — it hurts. Minus $240/year.
Mixpanel, $40/mo. Analytics for one of my sites. Switched to free Plausible because I needed exactly three numbers — visitors, sources, path. Mixpanel is a powerful tool for a team making Big Product Decisions. I have no Big Product Decisions right now. I have three numbers. Minus $480/year.
Loom Pro, $15/mo. Video messages. Stopped recording, because nobody watches video messages longer than 90 seconds, and free Loom caps at five minutes — which is more than enough for me in 2026. Minus $180/year.
Webflow, $39/mo. Separate story — I had one landing page hosted there. Rewrote it in Next.js in half a day (Cursor did the heavy lifting) and moved to Vercel. Minus $468/year, plus $0 for hosting.
HubSpot Starter, $50/mo. CRM. Realised that at any given time I have 8-12 clients in the active pipeline, and HubSpot for that is a tank for crushing ants. Moved to a single Notion table. Minus $600/year.
Buffer, $15/mo. Social posting. Realised my "content strategy" is "when something gets written, I publish it". I don't need a scheduler — I need to press "publish". Minus $180/year.
Pitch, $10/mo. Slides. I had two pitches last year. Made the slides in Figma and exported to PDF. Minus $120/year.
Notion AI, $10/mo. Subscribed a year ago out of curiosity. Used it exactly three times since, because I already have Claude, and walking over to AI in a different window is one extra step. Minus $120/year.
Linear, an extra workspace seat, $8/mo. Pure admin debt. Minus $96/year.
Figma Editor seat, $15/mo. Stopped designing in Figma — restructured the process so I look in viewer mode (free) and the designer touches the file. Minus $180/year.
Total cut: $3,264 a year. A third of the original. No pain. No hesitation. No "let me try it and think it over".
What stayed — five tools I live in every day. Each one and the reason it doesn't get cut.
Cursor, $20/mo. Code editor with built-in AI. This is my main tool. I open it daily. I write not just code in it, but a serious chunk of text (this post included). Alternatives exist, but I'd pay $40 or $60 for Cursor. Not advertising — diagnosis.
Claude Pro, $20/mo. Chats with the model. Mostly used for editing, parsing long documents, prepping for meetings. I tried living on the API alone once — didn't work, because the chat interface is faster when the question is unstructured.
Anthropic API, ~$70/mo. This is for everything that's automated — handling emails, generating responses, classifying tasks. It's fuel. Pay-as-you-go, never been excessive. Last year I paid more — switched to smaller models for routine work, the bill dropped by a third.
Notion, $10/mo. Documents, notes, tables, weak CRM. One person, one workspace, zero appetite for complication. Notion in 2026 is the place where all my written memory lives. Nothing replaces it.
Vercel, $20/mo. Hosting for all my sites and tools. Cheaper alternatives exist. I checked — savings of $5-10 a month, hours a year lost on configuration. Not worth it.
Total: $140/mo = $1,680/year. It used to be $9,760. Savings — $8,080 a year. In place of that money I now have a simple operation: I know what I'm paying for every month and can describe each line in a single sentence. That, as it turns out, is a rare state for a founder in 2026.
Three non-obvious takeaways I pulled out of this exercise.
One. Most SaaS tools are specialised for a workflow you've already changed. Calendly is great when you do 8 meetings a day. HubSpot is great when you have 200 leads in the pipeline. Webflow is great when you don't have a developer and you're making landing pages. If your process changed and the tool stayed — it runs on inertia, not utility.
Two. AI swallowed a serious slice of niche SaaS. Grammarly, Otter, Notion AI — these are all "specialised models" from an era when the general model didn't cope. In 2026 the general model copes. Not perfectly, but — enough to save three specialised subscriptions.
Three. Subscriptions "for $10 a month" are the most dangerous category. Easy to approve, easy to forget, easy to stack a dozen. When you have twelve subscriptions at $10 each, that's $1,440 a year, and each one in isolation feels like nothing. I now look at new subscriptions in annual money. "That's $120 a year" is a different conversation than "that's $10 a month".
I now keep this list in Notion (where else, of course). Run through it once a quarter. In April I cut two more — a Google rank tracker and an SSL monitor (a cron job handles it for free). Down to seven left. Or eight, actually — I added a new one for one specific task that neither Claude nor Cursor could really handle.
Good subscriptions are working tools. Bad subscriptions are an archive of intentions you pay for every month. A year of not separating them is $8,000 of archive that needs to go.