In "the old days" - a long, long time ago, in 2023 and before ... professionals who worked alongside me often thought that I have a tool obsession. An unhealthy tool obsession to be precise. They were (are?) right that I do indeed have a tool obsession, but I'd make the argument it's pretty darn healthy. This obsession of mine has helped me hone in on my superb competitive intelligence research and analysis capabilties. Having had this obsession for so long, it's amply prepared me to cope with the AI-rage. And, it's also made it incredibly easy for me to evaluate and choose tools when I need a new solution.
Today, businesses are still struggling no less than before, and perhaps losing even more cash in the process thanks to the invisible AI spend, often unable to limit that stack and keep track of the tools they're subscribing too.
Many might think that that's an executive leadership problem, or an Operations team problem. But not theirs.
I believe that true ownership and accountabiliity for each and every employee contributing to the business is also protecting it though, and this means being sensitive to the money spent in order to run that busienss.
So - when you need to choose a tool, or an entire infrastructure, how can you also take responsiblity for the insane tool sprawl costing your business unnecessary costs and still build a stack that supports you and your team? Let's get into it.
What is tool sprawl and why should I care?
Most tool decisions run in the wrong order. The shortlist arrives first, a demo follows, and the criteria get written afterwards to justify whatever won the demo. By then it's landing in a stack nobody has audited in three years, beside two tools that did most of the job.
Why does that keep happening? Because the vendor hands you a feature grid built around what that vendor happens to do well, and reacting to a grid is far easier than starting from a blank page, so the grid quietly becomes your criteria. You measure the doorway before you buy the fridge, and tool selection is the one purchase where we skip the measuring.
The tool is never the point. The fix is mostly sequencing, and it starts well before anyone opens a pricing page.
What do we own already, and does it still earn its place?
Start with an audit: what we own, what each thing is for, who actually opens it, what it costs, and what it connects to. It takes a week, and it's the cheapest week in the process.
Zylo's 2025 SaaS Management Index found that organizations waste an average of $21M a year on unused SaaS licenses, a 14.2% increase year over year! The same report puts 70% of SaaS spend with lines of business, against IT's 26.1%, and that second number explains the first. When most buying happens outside the one function that can see the whole stack, five teams buy five tools for one job and nobody reconciles them.
Then ask the harder question: should anything you own be replaced rather than supplemented? Adding is politically easy and structurally expensive: every addition is another integration, another login, another place for the truth to live, and another renewal date. Should we keep the tools people love? Well, of course. But "we already pay for it" is not a reason, and neither is "the team is used to it" when the habit in question is a workaround.
Decide what counts before you meet a vendor
Write the must-haves, the nice-to-haves and the disqualifiers down in advance, while nobody is charming you. Once you've seen a good demo the criteria bend to fit the product, and they bend one way: what you just watched becomes a must-have, and what it lacked becomes a nice-to-have. So which criteria decide it? Four, and only one of them is on the vendor's grid.
Integrations belong in the must-haves. Every time. A tool that doesn't talk to the systems you already run is a second copy of your data with its own opinions, and "we'll build the connector later" is how that copy becomes permanent.
Compliance is the criterion that quietly disqualifies, and it's cheaper to check now than in legal review. When we built Coro's compliance hub across 20+ global regulatory frameworks, most of the work was mapping which requirement applied to which buyer in which market. That same mapping tells you which tools are already out.
A worked example: documentation tooling, 2021
In March 2021 I ran a category-level evaluation of documentation tooling for a team that needed to buy something, and every vendor was answering a different question. The unit of comparison was the category, not the product, because products change quarterly and categories change slowly, and once you know which category fits your environment, your stack, your in-house roles and culture, the shortlist writes itself.
CategoryExamplesStrengthsWeaknessesSupport platformsZendesk, IntercomHosted, built for support pipelinesNo single-sourcing. ExpensiveSingle-source CMSsPaligo, MadCap FlareOne source of truth, many outputsHosting costs extra. API docs are awkwardHybrid KB platformsElevio, HelpDocsHosted, good integrations, versioningOutput is locked in. ExpensiveDocs-as-code, static generatorHugo, GatsbyCheap, scales, snippets single-source itHard for non-developers to troubleshootDocs-as-code hybridsArchbee, Gitbook, ReadmeEasy to use, Git and markdown, API supportLittle single-sourcing. Weak image managementAPI-first platformsRedocly, Swagger, PostmanOpen API specs, try-it tools, versioningNo snippets, nothing single-sourced
The costing came out in the same shape: Paligo at $269 per writer to publish into Zendesk, or $170 hosting free on AWS or Git, plus a $15 Zendesk admin license, for $284 a month. Naming the number turns a preference into a decision, and it's the step that gets skipped most. That deck is five years old and some of those products barely resemble their 2021 selves, so the verdicts have aged. The shape hasn't, and it's the shape of any competitive comparison built for a market: categories first, named examples second, honest failure modes third.
Why should you review the complaints too?
The directory we publish holds 2,433 tools, and 1,337 of them carry a G2 or Capterra rating. The schema is the interesting part, because every column in it exists to answer a question somebody asked during a selection.
JTBD comes first, because the job precedes the category, and integrations and price model sit right behind it, since those are must-haves pretending to be metadata. Likes and complaints get separate columns, and the complaints column earns its place, because support quality shows up there. Direct competitors are linked, so a shortlist can be built out of a single row, and company status flags an end of life or a company that appears closed, the failure a feature grid never warns you about.
A note on the limit: I build AI-assisted systems on top of models other people train, so "has MCP" is a column we can evaluate and model architecture isn't. The directory records what the market says about a tool, and it can't rank anything for your situation, because your must-haves aren't in it. The curated toolkit of 53 resources is a different thing: what's good, rather than what exists.
Baby steps
Test before you choose, using your own messy content, your own permissions model, and one person who wasn't in the buying room. Demo data is clean and yours isn't, so a demo tells you the tool works and a trial tells you whether it works here.
Where you can, sign up for a single month first and commit once you're convinced, because monthly pricing is worth a premium while the decision is open. For enterprise or business-critical purchases that option usually doesn't exist, since procurement, security review and annual terms arrive together. Then the pilot carries the weight the trial month would have, and the exit terms matter more than the discount.
Key takeaways
Run it in this order and most of the argument disappears.
None of this is specific to AI, which is rather the point: an AI purchase made without steps one through eight fails the way a documentation platform failed in 2021, only faster, with a bigger invoice. If you're mid-selection, or looking at a stack nobody has audited since before the last reorg, let's talk it through. Bring the list of what you already pay for, because that's usually where the answer is hiding.
