The Point Solution Squeeze Is Priced In Before It Is Proven: 3 Escape Routes

Buyers spent 2026 promising to cut vendors. The average software portfolio shrank by 0.07%. Every point solution that was supposed to disappear is still installed. What fell was the price: median enterprise SaaS changed hands at 3.3x revenue at the end of Q1 2026, against 4.9x three months earlier.

What the point solution squeeze actually is

A point solution does one job. Scheduling, e-signature, transcription, contract redlining, expense capture. It is bought because it is better at that job than the suite already installed.

The squeeze has two sides. Platforms above absorb the feature into a bundle the buyer already pays for. Agents beside it call an API and skip the interface the product was built around.

Neither side kills the company on contact. Both remove the reason a buyer pays a separate line item. The mechanics of that absorption are covered in detail in the analysis of how agent ecosystems absorb software categories, so this post starts one step later. If you accept the squeeze, what do you actually do about it?

My answer is that there are three routes out and no fourth. Go deep enough into a vertical that a platform will not follow. Move down the stack and become infrastructure something else calls. Or sell into the consolidation while the buyers are active.

The rest of this post tests each route against 2026 evidence rather than against the pitch-deck version. Two of the three turn out to be weaker than the standard advice claims. One is stronger.

The consolidation nobody can find in the data

Start with the part that undercuts the narrative, because it undercuts mine too.

Vendor counts are flat. Spend is not.

Zylo's 2026 index puts the average organisation at 305 SaaS applications, with a median of 240. The year-on-year change in that count was a decline of 0.07%. That is not consolidation. That is a rounding error.

Spend went the other way. Average annual SaaS outlay rose 8% to $55.7 million, with a median of $20.6 million. Buyers are paying more for the same number of tools.

Read those two numbers together and the popular story inverts. Nobody removed the point solutions. They renewed them at higher prices. The rationalisation programmes are real, and their effect on the count is close to zero, which is the same pattern described in the earlier piece on SaaS sprawl and what rationalisation actually removes.

Shadow IT is running the other way. Torii measured the same question differently and got a much larger answer. By instrumenting browser activity, OAuth grants and direct sign-ups rather than contracts, it found an average of 830 applications per enterprise, 61.3% of them outside formal IT oversight.

Its conclusion was blunt. AI is not consolidating software, it is expanding the long tail. Over half of the fastest-growing unmanaged applications were AI-first tools arriving through OAuth grants rather than through procurement.

So the honest position on the demand side is that buying behaviour has not consolidated at all. Employees are adding single-purpose tools faster than procurement can retire them, and the tools they add are increasingly AI-first.

Where the squeeze is real: price, not headcount

The market is not removing point solutions. It is refusing to pay for them at the old rate. Three pieces of evidence point the same way.

The multiple compressed by a third in one quarter

FE International's mid-year report put the median enterprise SaaS transaction at 3.3x trailing revenue at the end of Q1 2026, against 4.9x three months earlier. Profitability is expected to move the other way: median EBITDA margins across the public SaaS cohort are projected at 22.6% for 2026, up from 20.0% in 2025.

That combination matters. Buyers are paying less at a time when software profitability is forecast to improve rather than deteriorate. The discount is not about execution. It is about the durability of the revenue.

Meanwhile AI infrastructure and cybersecurity assets held double-digit revenue multiples. The market has already decided which parts of software it expects to still exist in five years, and it has expressed that decision in price rather than in prose. The wider repricing of listed software is covered in the breakdown of the SaaS selloff.

Microsoft priced four categories into a 5% rise

On 1 July 2026 Microsoft 365 E5 went from $57 to $60 per user per month, an increase of 5.3%. In exchange, four capabilities moved into the suite: Security Copilot, Intune Endpoint Privilege Management, Enterprise Application Management and Cloud PKI.

Each of those had a standalone market. Endpoint privilege management is a category with established independent vendors. Cloud PKI displaces third-party certificate authorities. Security Copilot had been sold separately at $4 per compute unit hour, with annual bills that regularly passed $100,000 for enterprises running a security operations centre.

E5 tenants now get 400 compute units per month for every 1,000 licences, capped at 10,000. Overage is throttled, not billed. That is the absorption mechanic stated in dollars: four purchase decisions collapsed into one renewal, at a 5% premium.

The buyer did not cancel the incumbent tool on day one. The buyer stopped defending its budget line at the next renewal. That lag is why vendor counts look stable while pricing power drains, and it is why the count is a bad early indicator for a founder.

Chegg is what completed absorption looks like

The full sequence takes years and looks ugly at the end. Chegg reported Q2 2026 revenue of $51.8 million, down 50.7% from $105 million a year earlier.

The company did almost everything right on the way down. Operating expenses fell by roughly half to $32.3 million. Free cash flow turned positive at $6.4 million from negative $12 million. Adjusted EBITDA margin held at 17.4%.

Chief executive Dan Rosensweig described the position plainly in the Q2 presentation, saying AI created real headwinds and the company responded by rebuilding an AI-first cost structure. Cost discipline of that quality bought time. It did not buy back the category. Nobody at Chegg mismanaged the response, and the revenue still halved.

Escape route 1: go deep enough that the platform will not follow

The standard advice is to pick a vertical. It is good advice stated at the wrong altitude, because most vertical positioning is a marketing layer over a horizontal product.

Depth is not a landing page that says "for law firms". Depth is when the product holds data, workflow steps or regulatory obligations that a general platform would have to build a compliance function to touch. That is the version worth having, and it is the argument developed in the piece on vertical AI eating horizontal software.

Tomasz Tunguz makes the related point from the vendor side. He argues that AI is rebundling what SaaS unbundled, because a model release cadence measured in weeks pushes buyers toward a platform they can trust for three to five years. His examples run the other direction from the usual advice: Harvey expanded from legal automation into professional services covering more than 25 tax jurisdictions, and Glean moved from enterprise search into vertical solutions across healthcare, financial services and government. We covered one of those expansions in the analysis of Harvey against Thomson Reuters.

Read carefully, that is not an argument for staying narrow. It is an argument that narrowness is a starting position, not a destination. The companies cited as vertical winners all widened once the wedge held.

The vertical premium is thinner than the pitch decks say

This is where I disagree with most of what is written on the subject. The claim that vertical software commands a large valuation premium does not survive contact with current comps.

Multiples.vc data as of 19 August 2026 puts vertical software at 2.3x next-twelve-month revenue and horizontal at 2.2x. That is a gap of one tenth of a turn. Secondary write-ups quoting vertical premiums of 40% or more are generally comparing different samples across different periods, and a premium that large does not appear in any current public comp set I could open.

Vertical depth is still worth pursuing. It buys retention, pricing power and a slower competitive clock. It does not currently buy a valuation premium, and a founder choosing this route on the promise of a multiple is choosing it for the wrong reason.

Escape route 2: become infrastructure, not an interface

The second route runs downward rather than sideways. Stop competing for the screen and start being the thing other software calls.

The price signal here is unambiguous. Infrastructure software trades at 2.8x forward revenue against 2.2x for horizontal application software, and data infrastructure sits at 5.4x. FE International found the same shape in private transactions, where AI infrastructure and cybersecurity held double-digit revenue multiples while the enterprise SaaS median fell to 3.3x.

That is a market telling you where it thinks durability lives. An interface can be absorbed by a bundle. A dependency has to be replaced, and replacement is a project with a budget, a migration plan and a risk register attached. Bundling a competitor's feature is a pricing decision. Replacing a dependency is an engineering programme, and those get deferred.

The test is whether an agent would call you

Here is the check I would run on any single-feature product this quarter. If an autonomous agent had to complete the job your product does, would it call your API, or would it do the job itself?

If the answer is that it would call you, something in your product is not reproducible from a prompt. Proprietary data, a licence, a network of counterparties, a settlement rail, a regulator relationship. That is an infrastructure position and you should invest in it directly rather than continuing to spend on the interface around it.

If the answer is that a competent agent would just do the work, the interface was the product. That case is examined in the piece on what the wrapper insult actually gets right, and the practical response is to move whatever is genuinely hard to reproduce into an addressable surface. Making that surface callable is what the interoperability layer is for, which is the subject of the explainer on agent interoperability standards.

Escape route 3: sell into the consolidation

The third route is the one founders resist naming, so it is worth naming precisely. If your product is a feature, the highest-value buyer is a platform that wants the feature and does not want to build it.

Record volume, thinner prices

The activity is there. Technology M&A produced $649 billion of announced transactions in the first half of 2026, inside a global market tracking toward record annual volume.

The shape of it is less friendly. Deal counts fell about 9% year on year while value concentrated at the top, with transactions above $5 billion accounting for 48% of total value against 39% in 2025. Large deals are absorbing the capital. Small ones are competing for what is left, at 3.3x rather than 4.9x.

So the window is open and it is narrower than the headline volume suggests. A founder waiting for the multiple to recover before selling is making a forecast, not a plan.

Run this route when three things are true at once. Your growth has settled near the 15% median, your product solves one job that a larger vendor already lists on its roadmap, and your retention is holding. That last condition is the asset. You are selling a working revenue base, not a story, and the base is worth most while it is still intact.

Selling into consolidation is a legitimate outcome and most founders file it as failure. That framing costs real money, because it delays the decision into the quarter when retention starts slipping and the buyer sees it in diligence. The median private company in this band is growing at 15% with gross retention at 91%. Those are the numbers a buyer underwrites, and they get worse after absorption starts, not before.

Choosing between the three routes

The routes are not equally available to every company. They demand different assets and they fail in different ways.

There is a fourth path that is not an escape route: keep operating a profitable single-feature business and accept slower growth. Zylo's flat application counts suggest that is survivable for longer than the discourse implies. It is not a strategy, it is a decision to stop optimising for exit value.

Where this argument is weakest

Three problems, and the first is serious enough that it should change how you read everything above.

The best datasets disagree with each other

Zylo counts 305 applications per organisation. Torii counts 830. Both are 2026 figures from firms whose business is measuring exactly this. They differ by a factor of nearly three because they measure different things: contracted and discovered spend in one case, browser and OAuth activity in the other.

That gap is not a detail. If Torii is closer to the truth, the long tail of single-purpose tools is far larger than procurement believes and the absorption thesis is weaker than I have argued. Anyone quoting one of these numbers without the other is picking a side.

The same caution applies to the multiples. Public comps at 2.2x to 2.8x and private transaction medians at 3.3x are different populations on different bases, and I have used both. Read them as direction, not as a spread you can arbitrage.

One quarter of multiple compression is also not a trend. The 4.9x to 3.3x move is a single quarter measured by a single advisor, and transaction medians move on deal mix. Two quarters of small deals can produce that pattern without anything structural changing. I have treated it as a signal because it agrees with the direction of the public comps. It remains one data point from one dataset.

Rebundling may be a lab story, not a SaaS story

The rebundling evidence comes mostly from AI-native companies expanding scope: Harvey, Glean, ElevenLabs, the frontier labs building industry teams. Those are companies with capital and model access, not a signal about what happens to a $6 million ARR scheduling tool.

It is possible that the labs consolidate at the top while the application layer stays as fragmented as it has always been. The seat-level pricing pressure discussed in the analysis of seat compression would still bite, but the category count would not fall, and the three routes below would matter less than they appear to.

Where to start this week

Two exercises, both finishable before your next board update.

First, run the agent test on your own product and write the answer down in one sentence. Would an autonomous system call your API to finish the job, or would it do the job itself? If you cannot answer without qualifying it, the honest answer is the second one.

Second, open your largest platform vendor's most recent licensing announcement and read the list of what moved into the bundle. Microsoft absorbed four categories into E5 for a 5.3% price rise. Do the same reading for whichever platform your buyers already pay, and check whether your category appears. Whether your feature is on that list is a better predictor of your next three years than your win rate is. The platform side of that question is examined in the piece on whether Salesforce is the platform or the roadkill.

Frequently asked questions

Related reading