Exclaimer is a good product. It manages email signatures for Microsoft 365, it costs a couple of bucks per mailbox per month, and plenty of companies should just pay for it. I rebuilt its core functionality anyway, because I wanted a real answer to a question I keep getting asked: can AI-built software actually replace a SaaS subscription, and what does that take in practice? I demoed the finished system at our September AI workshop, and this article is the honest accounting.
How long did it actually take?
About 4 to 5 focused hours, spread over one weekend. I started on a Friday, worked in spare moments on Saturday and Sunday, and by Monday I had a working signature system with a web designer, dynamic campaigns, and group-based rules. Maybe a little more than five hours if I am being strict. But it was never a grind, it was a conversation I kept picking back up whenever I sat down at the computer.
That number only makes sense once you see how the work broke into phases, because each phase answered a different question.
What were the four phases?
Proof of concept, real application, refinement, and multi-tenant onboarding, in that order. That sequence matters more than any individual feature.
Phase 1 was proving it could work at all. A couple of hours on Friday: a test tenant, a small VM, the Microsoft 365 transport rules that route mail through the system, and a basic signature stamped onto an outgoing email. Nothing pretty. The only goal was confirming the core mechanic before investing anything else.
Phase 2 was turning a trick into an application. The AI built out the database, the web application, and a signature designer, and we worked through the ordinary bugs that surface when a proof of concept becomes real software.
Phase 3 was refinement, and it was where the real lessons lived. Drag and drop editing in the designer. Embedded images, because Outlook blocks externally linked images by default and I wanted signatures that displayed as trusted, the way polished commercial signatures do. And deliverability, which deserves its own section.
Phase 4 is the one I have not fully built yet: onboarding other tenants. The plan is for the system to check a new tenant’s DNS records and report exactly what needs to be added, validate those records once they exist, then log into the tenant and create the transport rules automatically, since we already have permissions to do that. There is also a gotcha I know is coming: what happens when a user already has a signature set up in Outlook? Do we keep both or override theirs? I do not have that answer yet, and I am fine saying so.
What went wrong along the way?
My test emails kept landing in junk mail, and fixing that was the most instructive part of the whole build. I did not diagnose it myself. I just told the AI, plainly, that messages were hitting spam folders. It asked me for information, and when I did not know how to get it, like pulling the full email headers out of an Outlook message, it walked me through the clicks.
Then it asked for DNS changes. It told me which records it needed, I logged into GoDaddy and added them, and it verified my work afterward, which mattered because a typo in a DNS record fails silently. The same pattern fixed the blocked images: I asked why our signature images were blocked when other companies’ were not, learned that trusted signatures embed their images in the message, and had the AI make that change. The images displayed cleanly from then on.
Notice the division of labor. I never stopped being necessary. I was the one with the GoDaddy password, the Outlook mailbox, and the judgment about what looked right. The AI was the one who knew why things were failing and what to do about it.
When does replacing a subscription with a custom build make sense?
When you have a technical partner, you want features the vendor does not offer, or per-seat pricing scales badly against your headcount. All three applied to me. As an MSP we maintain software for a living, so upkeep is not a burden. I wanted things like scheduled campaign banners tied to specific groups and, eventually, one system onboarding many client tenants, which is exactly where per-mailbox pricing multiplies. A couple of dollars per mailbox is cheap for one company and a real number across every mailbox we manage.
If that describes you, the economics have genuinely changed. This category of tool used to be a months-long development project, which is why buying always won. Now it is a weekend of focused attention plus a partner who builds custom software this way.
When should you just keep paying for the SaaS?
When you need a vendor’s compliance certifications, or when nobody in your orbit will maintain the software after the excited weekend ends. Some industries need to point auditors at a certified vendor, and a homegrown tool cannot answer that requirement no matter how well it works. And software is never finished. Microsoft changes something, a certificate expires, an edge case appears, and someone has to care. If that someone does not exist, the subscription is not an expense, it is cheap insurance.
Exclaimer earns its price for most of its customers. The point of this build was never that the vendor is overcharging. It was measuring how far AI-assisted development has moved the line between “buy it” and “build it.” That line used to sit at months of developer time. It now sits at a weekend, a clear description of what you want, and someone accountable for keeping it running. Where it sits for your specific subscription is a conversation worth having before the next renewal.