From Mihir Katiyar | Product & Market Analysis
Horizontal SaaS in 2026: The Bundle Already Priced Your Feature at Zero
On this page
Yes, for three shapes of product, and no for the rest. In January 2025 Google took a Gemini add-on that cost $20 per user per month and folded it into Workspace for a $2 increase. That is the actual risk to a horizontal SaaS company in 2026. Not a rival product. A repackaging you never got to compete against.
Key takeaways
- Suite owners removed the line item rather than beating the price. Google folded a $20 per user per month Gemini add-on into Workspace Business plans for a $2 per user increase in January 2025. Microsoft folded Copilot Chat into Microsoft 365 base packaging on 1 July 2026.
- Microsoft's biggest increases landed on the cheapest seats. Microsoft 365 F1 rose 33% and F3 rose 25%, against 5% for Microsoft 365 E5. Frontline workers who never had an AI budget now fund one.
- Horizontal still works when the artefact lives inside your product. Figma reported $370.1 million of second-quarter 2026 revenue, up 48%, with 136% net dollar retention, while Adobe left its competing design tool in maintenance mode.
- Horizontal versus vertical is the wrong question. Ask whether a suite owner can ship a good-enough version as a checkbox, and separately whether a general agent can do the job without your interface. Those are two different tests with two different answers.
What horizontal SaaS means now, and who this is for
This piece is written for founders and directors deciding where to put the next two years. The single question it answers is whether a horizontal product is still a defensible starting position.
Horizontal software sells the same product to every industry. Design, payroll, project tracking, notes, scheduling, customer support. Vertical software sells one industry a product shaped around that industry's records and rules, which is the subject of the evidence on vertical AI taking share.
The industry count is the wrong way to read the distinction. What decides your outcome is whether the buyer already pays somebody else for a bundle that could contain you. Horizontal is a distribution position, not a product category.
Direct answer for anyone who reads only this paragraph. Starting a horizontal SaaS company in 2026 is defensible when the work product is created and stored in your system, when users outside a single company collaborate in it, and when the buyer sits outside the suite renewal. Fail all three and you are building a feature that a suite will include at no extra charge.
The bundle came for the price, not for the company
The story founders tell each other is that a large incumbent clones your feature and outspends you on distribution. That is not what happened over the last three years. Something quieter and more final happened instead.
Four suite owners stopped charging for the category. Not discounted it. Removed it as a purchase decision.
Google put a $20 add-on inside a $2 increase
On 16 January 2025 Google made Gemini part of Workspace Business and Enterprise plans. Before that day, Gemini for Workspace was sold separately.
Google published the arithmetic itself. A Business Standard customer with the Gemini add-on had been paying $32 per user per month, and would now pay $14.
Read that from the seller's side rather than the buyer's. A product line priced at $20 per user per month was withdrawn as a line item and replaced with a $2 increase on the suite. The category lost roughly 90% of its price in a single announcement, decided by a company that never had to win it on merit.
My view is that this one change did more damage to standalone assistant startups than any competing product launched that year. A rival product gives you something to sell against. A repackaging gives you nothing to point at.
Microsoft repeated the move eighteen months later
On 4 December 2025 Microsoft announced a global pricing and packaging update effective 1 July 2026. Copilot Chat entered base packaging, including inbox and calendar awareness and access to Word, Excel and PowerPoint agents.
The list prices moved with it. Office 365 E3 went from $23 to $26. Microsoft 365 E3 went from $36 to $39. Microsoft 365 E5 went from $57 to $60. Business Basic went from $6 to $7 and Business Standard from $12.50 to $14.
The interesting part is the distribution of the increase. Frontline seats moved most, with Microsoft 365 F1 rising 33% and F3 rising 25%, against 5% at the top of the enterprise range.
Those frontline seats were the population least likely to have ever bought an AI add-on. They are now the population funding the largest proportional share of one. If you sell a per-seat tool to that population, your buyer's software budget just moved before you got a meeting.
Zoom and Atlassian got there first
Zoom announced AI Companion on 5 September 2023, included at no additional cost with paid Zoom user accounts. Atlassian began including Rovo in Jira, Confluence and Jira Service Management from 9 April 2025, starting with Premium and Enterprise plans.
Neither charges a separate AI subscription line. Atlassian meters heavy consumption through credits, which is a different thing from an add-on price, and it is the model I would expect the rest of the category to converge on.
Four companies, three different years, one behaviour. If your product's core job can be described in a suite's release notes, your price is not set by your value. It is set by the increment the suite owner chooses to add, which is the same force examined in the analysis of what agents do to per-seat pricing.
The squeeze from below is narrower than the pitch
The second pressure gets more attention and deserves less. A general assistant with tool access does the job without a visit to your interface.
The adoption is real. OpenAI announced on 5 November 2025 that it had passed 1 million business customers, with more than 7 million ChatGPT for Work seats. Those seats sit inside companies that still pay for point tools, which is the tension explored in the piece on point SaaS absorption into agent ecosystems.
What the agent actually replaces
An agent replaces an interface. It does not replace a system of record.
If your product is a screen over data that someone else owns, the screen is the replaceable part and you should assume it will be replaced. If your product is where the data is created, the agent becomes a client of your product rather than a substitute for it.
There is a single question that settles this for most founders. If a customer's agent called your API every day and no human ever opened your application, would you still get paid the same amount? A yes means you sell infrastructure. A no means you sell a screen, which is the distinction argued in the piece on whether wrapper is an insult or a business model.
Horizontal companies that are still winning, and why
The pessimistic case is easy to write and incomplete. Some horizontal companies are compounding right now against every one of the pressures above.
Figma is the cleanest counterexample
Figma reported second-quarter 2026 revenue of $370.1 million on 5 August 2026, up 48% year on year, which the company described as its third consecutive quarter of accelerating growth. Net dollar retention was 136% at 30 June 2026.
Customer counts moved with it. 15,964 customers pay more than $10,000 a year and 1,635 pay more than $100,000, growing 34% and 46% respectively. Full-year guidance was raised to a range of $1.463 billion to $1.467 billion.
Design is as horizontal as software gets. Every industry designs something. Microsoft, Google and Adobe all ship design surfaces, and the bundle did not take the category.
The incumbent walked away twice
Adobe agreed to buy Figma for about $20 billion, then terminated the deal on 17 December 2023 under regulatory pressure and paid a $1 billion reverse termination fee.
Adobe had already put its own competing product, XD, into maintenance mode. In early 2024 it confirmed it had no plans to invest further in it. An incumbent with two decades of relationships in that buyer's department tried to buy the category, could not, and then declined to compete for it.
That is worth sitting with, because the reason is the whole framework. The design file was in Figma, the version history was in Figma, and the people commenting on it did not all work for the customer.
| Shape. | Why a bundle struggles. | Supporting evidence. | Where it still fails. |
|---|---|---|---|
| The artefact is created and stored in your product. | Switching means moving the file and losing the history, not turning off a feature. | Figma at 48% growth and 136% net dollar retention while three suite owners ship design surfaces. | Most categories have no artefact. A summary is not a file. |
| Users span more than one company. | A suite licence stops at the customer's tenant boundary. | Microsoft's Teams settlement required interoperability and data export for rival tools, binding for years. | Two-sided products have the hardest cold start in software. |
| The buyer sits outside the suite renewal. | A different budget, a different approver and a different renewal date. | Figma's customers above $100,000 of annual revenue grew 46% year on year, bought by design and product teams. | Procurement consolidation pushes every budget back toward the suite eventually. |
The evidence column carries one strong public example per row, not a measured population. Treat the rows as hypotheses with supporting cases, not as a tested model.
Five tests to run before you start
Slogans do not help here. Below are the questions I would actually answer on paper before committing two years, in the order they matter.
| Test. | A pass looks like. | A fail looks like. |
|---|---|---|
| 1. Whose budget pays for it. | A buyer outside IT, with a budget line and a renewal date of their own. | The same administrator who renews Microsoft 365 also decides on you. |
| 2. Where the work product lives. | The artefact is created and stored in your system, with its history. | You read and write a record that lives in someone else's system. |
| 3. Whether it crosses the company boundary. | People outside the customer's tenant collaborate inside the product. | Every user is an employee of the one company paying you. |
| 4. Whether a checkbox can carry it. | The job needs a surface people work inside for hours a day. | The job is a panel a suite can add to a screen the user already has. |
| 5. Who signs for the outcome. | Somebody is accountable for the result and needs an audit trail. | The output is a suggestion nobody has to stand behind. |
Three passes out of five is not a business, and the combination matters more than the count. Failing tests 1 and 4 together is the profile a bundle absorbs without noticing, because the buyer is already in the room and the job fits on a screen they already own.
I would not start a company with that pair of failures in 2026, whatever the market size looks like. Test 5 is the one most founders skip, and it is the one that turns a tool into a purchase order.
Run the same checklist on the incumbent's incentives while you are at it. Adobe did not decline to compete with Figma out of politeness. Pricing a serious rival meant repricing Creative Cloud, and the arithmetic of defending a bundle is covered in the test for whether you are standing in a kill zone.
Pricing a horizontal product when the bundle costs nothing
Founders usually describe this as competing with free. That framing is wrong and it leads to the wrong pricing.
You are not competing with free. You are competing with already paid for, which is a much stronger position for the buyer. Nobody has to justify a renewal they did not initiate.
Two responses hold up. Price on the outcome or the consumption rather than the seat, so the comparison is against a result instead of a per-user line. Or price clearly above the bundle and refuse to be compared, which only works when the job is visibly different rather than better.
What does not hold up is pricing just under the incumbent. That invites a direct comparison with a product the buyer has already bought, and it caps you at the ceiling of a category somebody else set. It also concedes the argument covered in the analysis of application sprawl and rationalisation, where consolidation pressure is doing the incumbent's selling for it.
Where this argument is weakest
Three problems with everything above, stated plainly.
Vertical is not the safe answer either
Nothing here says go vertical instead. A vertical product trades bundling risk for a smaller market, harder distribution and the same agent exposure at the interface layer.
The five tests apply identically to a vertical idea. A vertical product that reads someone else's system of record and shows a panel is in the same position as a horizontal one, with less room to grow out of it.
This is a survivor sample
Figma is one company. It is used here because its numbers are audited and public, which is exactly what makes it unrepresentative. Companies that failed the same tests and died did not file quarterly results.
The honest version of the artefact argument is that it is a plausible pattern with one very strong case, not a measured effect across a population. I could not find a dataset that tests it, and I would treat anyone who claims one with suspicion.
Bundling has failed before
Bundled defaults do not always win. Regulators have twice intervened in exactly this pattern, and the Microsoft settlement over tying Teams to Office is the recent example. A default that is good enough is still a default that can be switched away from when the alternative is materially better.
Frequently asked questions
Should you still start a horizontal SaaS company in 2026?
Yes, if the work product is created and stored in your system, users from more than one company collaborate inside it, and the buyer sits outside the suite renewal. No, if the job fits on a screen the buyer already owns and the same administrator who renews their productivity suite also decides on you. That combination is what a bundle absorbs without a fight.
What is the difference between horizontal and vertical SaaS?
Horizontal software sells the same product to every industry, such as design tools, payroll or project tracking. Vertical software serves one industry with a product shaped around its records and regulations. The more useful distinction in 2026 is not the industry count. It is whether a suite the buyer already pays for could include your product at no extra charge.
Why did Google make Gemini free in Workspace?
Google folded Gemini into Workspace Business and Enterprise plans on 16 January 2025 and raised suite prices by about $2 per user per month. By its own example, a Business Standard customer with the add-on went from $32 to $14. Including the feature protects suite renewals and removes a purchase decision that competitors were trying to win.
Can AI agents replace horizontal SaaS products?
Agents replace interfaces more readily than they replace systems of record. A product that is a screen over data someone else owns is genuinely exposed. A product where the data is created becomes something the agent calls rather than something it removes. The practical test is whether you would still be paid if a customer's agent used your API and no human opened your app.
How do you price a SaaS product against a bundled free feature?
Do not price just below the incumbent, because that invites comparison with something the buyer already pays for. Price on outcomes or consumption so the comparison is against a result rather than a per-seat line, or price clearly above the bundle and make the difference in the job obvious. Competing on a discount against an included feature is a losing position.
Is horizontal SaaS dead?
No, and the strongest counterexample is public. Figma reported $370.1 million of revenue in the second quarter of 2026, up 48% year on year, with 136% net dollar retention, in a category where Microsoft, Google and Adobe all ship competing surfaces. What died is the standalone feature priced per seat, not horizontal software as a category.
Where to start this week
Two exercises, neither of which needs a deck.
First, write your idea across the five tests above and mark each one honestly. Then find the person who renews your prospective buyer's productivity suite and work out whether they would ever hear your name. If the answer is no, test 1 is a pass and you have more room than the headlines suggest.
Second, read the last two release note pages from the suite your buyers already own. Not the marketing pages, the release notes. Everything shipped there is a category that has stopped having a price, and the list is the most accurate competitive research available to you at zero cost.
Related analysis
The pressure from above and below is covered separately. Read the piece on whether the platform incumbents become roadkill and the evidence on vertical AI taking share for the two ends of the same argument.
References
- Google Workspace, The best of Google AI, now included in Workspace Business and Enterprise plans, 16 January 2025. Used for the $32 to $14 example, the withdrawn add-on price and the inclusion date.
- Microsoft, Microsoft 365 pricing and packaging updates, announced 4 December 2025, effective 1 July 2026. Used for every list price and percentage change, and for the Copilot Chat packaging description.
- Figma, Figma announces second quarter 2026 financial results, 5 August 2026. Used for revenue, growth rate, net dollar retention, customer counts and guidance.
- Microsoft 365 Blog, Evolving our productivity offerings to resolve European competition concerns about Teams, 12 September 2025. Used for the unbundling commitments and interoperability obligations.
- Zoom, Zoom introduces Zoom AI Companion, available at no additional cost with paid Zoom user accounts, 5 September 2023. Used for the Zoom inclusion date and terms.
- Atlassian Support, Manage your bill for Rovo. Used for the 9 April 2025 rollout date, the plans covered and the absence of a separate Rovo invoice.
- Adobe and Figma, Mutual termination agreement, 17 December 2023. Used for the termination date and the $1 billion payment.
- OpenAI, 1 million business customers: the fastest-growing business platform in history, 5 November 2025. Used for the business customer and seat figures.
The weakest part of this source base is that the positive case rests on a single public company. Figma is cited because its figures are audited and disclosed, which also makes it a survivor rather than a sample. No public dataset tests whether the five criteria in this post predict outcomes across a population, and the checklist is the author's judgement rather than a measured model.
Related reading