We Built a CRM. The Bond Market Will Finish the Job.

Our CRM cost us almost nothing to build. The company selling you one borrowed heavily for the privilege, and that debt is repricing.

Oh, the blasphemy!

As a former Microsoft Business Applications MVP, I like Power Platform. I like Dynamics 365. I like SharePoint. Heck, I even liked Viva Goals, despite apparently being one of the few people who actually used it.

But guess what?

For TSG Technology, we built our own CRM.

Then we kept going and grew it into an end-to-end operating system covering CRM, OKRs, project management, timesheets, invoicing and the various bits of connective tissue normally spread across a collection of SaaS products.

And we actually use it to run the business. No more vendor selection, licences, implementation partners, configuration, integration and then years of living with somebody else's architectural compromises.

"Nobody ever got fired for buying..."

Large organisations don't buy Dynamics, Salesforce, ServiceNow or any other major SaaS platform simply because they are unable to build software. They buy institutional safety.

I heard a finance podcaster rationalise this as "at least there's someone to sue", defending SaaS against the emerging buy-versus-build economics. An odd landing spot for someone whose day job is pricing risk. I mean, no analyst would let that near a valuation, yet said podcaster expects it to clear a purchasing decision unchallenged.

You may be thinking this is just a rehash of "nobody ever got fired for buying IBM". It is, and that is precisely the point. That line has outlived mainframes, outsourcing and cloud, because it never needed to be true, only useful enough to settle the dissonance my "build your own" blasphemy provokes.

Narrative can only outrun arithmetic for so long, and next year's budget is where they meet.

Yet I get it. Enterprises need contracts, security controls, support, audit evidence, continuity arrangements, escalation paths and somebody accountable when things go wrong.

But let's not romanticise what "someone to sue" actually looks like in enterprise technology. If you've ever sat on the customer side of an incident, tell me I'm lying.

• The vendor says "talk to the partner".
• The partner says "it's the product".
• Support just wants to close your call to meet their SLA.
• Your SLA gets you meagre service credits assuming you can weave your way through the liability caps and exclusions in contractual minutiae.

Nobody was ever really buying someone to sue. They were buying a scapegoat.

We selected a reputable product, from a reputable vendor, through a defensible process, so we can't be blamed.

Separate software from stewardship

This is where we think the technology service model changes. The traditional build-versus-buy argument assumes the alternatives are an enterprise SaaS product, or some clever home-grown application that Kevin from Accounts built.

AI now makes that a false choice. Software can be created, tested, instrumented and maintained at dramatically lower cost. Low code taught us how to govern development once the ability to build became democratised. Agentic engineering takes that same problem to another level.

The scarce capability shifts from manufacturing software to responsibly standing behind it.

Value moves to the judgement, controls, evidence and accountability that make an organisation comfortable depending on it. That is stewardship in practice: not building the system and handing it over, but staying responsible for how it operates, how it changes, and whether it still does its job.

This is our thesis at TSG Technology. A customer can benefit from software designed around their organisation, in partnership with a specialist local provider who takes responsibility for its stewardship.

And stewardship is not just a managed service with a nicer name. A managed service is principally about operating the technology: uptime, incidents, patches, backups and support. Stewardship exercises care over it like a trustee manages an asset portfolio, staying accountable for the outcome.

That changes the buy-versus-build equation again. Much of what enterprise SaaS bundles as institutional assurance is real enough on paper: certifications, audit evidence, contracts, escalation paths. What it rarely includes is somebody answerable for whether the system still does the job it was bought to do. That part was always the veneer, and it is precisely the part stewardship supplies. Once it does, the allure of "nobody ever got fired for buying IBM" evaporates.

Once that moat weakens, one of the strongest remaining arguments for the SaaS premium weakens with it.

"Everybody got fired for buying..."

For more than a decade, SaaS had almost the perfect financial characteristics for the cheap-money era.

• Recurring revenue.
• High gross margins.
• Sticky customers.
• Predictable cash flows.
• Pricing power.
• Capex lite.
• Strong reported growth.

Those characteristics attracted enormous amounts of private capital. The playbook was simple: buy a recurring-revenue business using borrowed money, streamline it, grow it, buy a competitor or two, then eventually sell the repackaged product for more than you paid.

An awesome trade when borrowing was cheap. But this is where bond yields enter the story.

That inconvenient risk-free rate

Government bond yields are effectively the starting price for money, because lending to a government is about as close to risk-free as it gets compared with lending to a crypto exchange in the Bahamas, or a leveraged buyer of SaaS providers. Think about it: if you can get 5 percent interest virtually risk-free, you are obviously going to demand considerably more than 5 percent to lend to a leveraged borrower.

So when bond yields rise, and they have, that awesome playbook looks decidedly less awesome. The cheap debt used to buy those SaaS companies has to be refinanced at higher rates, at exactly the moment the SaaS company itself is facing margin pressure as AI eats away at its value proposition.

That's a squeeze, and those two changes compound brutally against leveraged equity.

That leaves two levers, and both point at the customer: charge more, or cut deeper. Streamlining that once removed genuine waste starts reaching into support, account management and roadmap, which is where it stops being good management and starts eating the institutional safety the customer was buying.

Back to our little CRM. Sure, we might be early, but do you think we are the only ones doing this? In investing terms we are a weak signal: one data point that means very little on its own, and a great deal if it turns out to be the first of many. That is how something as mundane as us building our own CRM starts connecting to the long end of the bond market.

There is another uncomfortable layer

The same issue potentially applies further down the AI stack too.

The technology industry is investing extraordinary amounts of capital in datacentres, GPUs, power, networks and frontier models. That is also done on the same borrowed money, and it now carries a higher hurdle rate too.

Yet the product being produced by all of that capital, intelligence itself, appears to be getting cheaper.

• Model prices are falling.
• Open models are improving.
• Inference efficiency is rising.
• Workloads can increasingly move between providers.

Brilliant for us consumers, and precisely why TSG Technology could create so much software capability ourselves. But that does not mean the companies providing the underlying AI captured the economic value that we created. In our case, most of it stayed with us.

• We avoided licences.
• We avoided implementation costs.
• We avoided somebody else's architectural compromises.
• We avoided a backlog whose timeline and priorities are set by someone else.

The model and infrastructure providers received a relatively small amount of revenue in exchange for enabling considerably more value.

Multiply that across an economy and you arrive at a perfectly plausible outcome.

AI creates enormous economic value while disappointing some of the investors funding it.

The bond market adds another ingredient: it is making capital expensive at exactly the same time. Our CRM is a tiny example of the first change. Higher long-term yields tell us why the second one matters. Put them together and you arrive at the problem facing SaaS, and leveraged SaaS in particular.

Cheap software, expensive money, and a stranded multiple.

The screenshot above is our own back office, running the business it describes. This piece extends an argument we made earlier in AI Works. The Returns Might Not.