The Consolidation Dividend

Executive Summary

Nobody decided to run eleven subscriptions. The stack accreted. A chat AI for questions. A second chat AI because the first one is weak at code. A meeting transcriber. A note-taking app with an AI add-on. An automation tool to move data between the other tools. A sync service so the notes follow you. Each decision was individually rational -- $9 here, $20 there, a free trial that quietly converted. The result is a stack that costs a four-figure sum every year, demands its own administration, and splits your working knowledge across seven databases that have never heard of each other.

This paper makes the case for stack consolidation -- not as minimalism for its own sake, but as a measurable dividend with two components. The hard dividend is money you stop spending: for a typical knowledge worker, $743-1,272 per year in retired subscriptions, plus reclaimed hours currently lost to copy-paste and context switching that are worth more than the software. The soft dividend is harder to invoice and worth more: one mental model instead of seven, one accumulating knowledge graph instead of seven amnesiac silos, one vendor relationship to evaluate instead of seven privacy policies you have never read.

Kent is the consolidation point we built: one desktop app that runs 13 built-in skills and unlimited custom skills against six AI providers, connects to your email, files, notes, and databases where they already live, and compounds everything you do into a knowledge graph you own. The last section is an honest accounting of what you should not consolidate -- because the argument for consolidation collapses the moment it becomes an argument for consolidating everything.

1. How the Stack Got This Big

1.1 The Accretion Pattern

SaaS sprawl is one of the best-documented phenomena in enterprise software. Zylo's 2025 SaaS Management Index, drawn from real license data across more than 1,000 organizations, found the average company maintains 269 SaaS applications, and that roughly half of purchased licenses go unused in any given month (Zylo, 2025). Okta's Businesses at Work report puts the average at 93 apps for mid-sized deployments -- a lower count, same trajectory (Okta, 2025). Gartner estimates that 25% of enterprise software spend delivers no measurable value (Gartner, 2025).

The individual professional runs the same pattern at smaller scale. The AI boom made it worse in a specific way: between 2023 and 2026, nearly every app in the stack added an AI assistant, and nearly every AI assistant became a separate line item. You are now paying for intelligence four times -- once in your chat subscription, once in your note app's AI add-on, once in your meeting transcriber, once in your CRM's copilot -- and each of those intelligences knows only what lives inside its own walls.

1.2 What a Typical Stack Costs

Here is the annual cost of a representative knowledge-worker stack, using March 2026 pricing:

Total: $1,272 per year. Seven apps, seven logins, seven billing relationships, seven data silos. Our companion paper on the frontier-model subscription trap walks through why the two chat subscriptions alone are structurally overpriced; this paper is about the stack as a whole.

1.3 The Overlap Nobody Audits

Lay the seven apps' feature lists side by side and the overlap is startling. Four of them summarize documents. Three transcribe or process audio. Three draft text. Two search your notes. You are not paying for seven capabilities -- you are paying seven times for about a dozen capabilities, each implementation walled off from the data the others hold.

The overlap persists because auditing it is nobody's job. Each tool was adopted to solve one problem, evaluated once, and never re-evaluated. The subscription meter, meanwhile, runs whether you use the tool or not. Zylo's license-utilization numbers say the quiet part: on average, half the stack is idle at any moment, and all of it is billed.

2. The Hard Costs

2.1 The Subscription Line Is the Small Line

$1,272 per year is the visible cost. The invisible cost is time, and the research here is consistent and damning.

Microsoft's Work Trend Index measured knowledge workers spending 3.2 hours per week manually moving data between applications so their AI tools could see it -- the copy-paste tax (Microsoft, 2025). Psychologists Rubinstein, Meyer, and Evans established that task switching carries a measurable cost, with complex switches burning up to 40% of productive time (Rubinstein et al., 2001). Gloria Mark's work at UC Irvine found that after a significant interruption, refocusing on the original task takes over 23 minutes (Mark, 2008). A seven-app stack is a context-switch generator: every 'let me check the other tool' invokes exactly the switching penalty this literature measures.

Hours per week lost to stack overhead. Sources: Microsoft (2025) for data transfer; switching and re-finding estimates derived from Rubinstein et al. (2001), Mark (2008), and Kent internal usage analytics (2026).

Call it 7.9 hours per week of stack overhead, and treat the estimate skeptically -- halve it if you like. At the Bureau of Labor Statistics median for knowledge workers ($78,000 per year, roughly $37.50 per hour), even the halved figure -- four hours a week -- is worth about $7,800 per year (BLS, 2025). The subscriptions are the small line. The overhead is the large one.

2.2 The Per-Seat Multiplier

For teams, every number above multiplies by headcount. A ten-person team running the representative stack spends $12,720 per year on subscriptions and burns something on the order of $78,000 per year in stack overhead -- the equivalent of a full salary spent on friction. Team plans usually cost more per seat than individual plans, not less, and the administrative burden of seven vendors (provisioning, offboarding, security review, invoice reconciliation) lands on whoever drew the short straw.

3. What Consolidation Actually Means

3.1 Replace the Intelligence Layer, Connect the Record Layer

The word 'consolidation' triggers a reasonable fear: the all-in-one suite that does everything badly. That is not the architecture we are describing, and the distinction matters enough to state precisely.

Your stack has two layers. The record layer is where your data lives: Gmail, Google Drive, Notion, your databases, your calendar. These are systems of record with years of accumulated content, and ripping them out is exactly the all-in-one mistake. The intelligence layer is everything that thinks about that data: the chat AIs, the summarizers, the transcribers, the AI add-ons, the glue automations that shuttle text between them.

The intelligence layer is where the duplication lives, and it is the layer that consolidates cleanly -- because intelligence, unlike records, does not accumulate switching costs inside any single tool. Kent replaces the intelligence layer and connects to the record layer where it already lives: native connectors for Gmail, Google Drive, Google Calendar, Notion, PostgreSQL, MySQL, SQLite, MongoDB, REST APIs, and any MCP server. Your email stays in Gmail. Kent reads it there.

3.2 The Mapping

Two caveats belong in this table rather than in fine print. Zapier doing genuine multi-step business automation is record-layer plumbing -- keep it. And a team actually running pipeline management needs a real CRM -- the row above describes the solo practitioner using one as an expensive rolodex.

4. The Hard Dividend

4.1 The Money

On Kent's Power tier, the subscription dividend is $743 per year. On the Free tier with your own API keys, it is the full $1,272, minus token costs that Kent's routing keeps low -- simple queries go to cheap fast models, hard ones to frontier models, background work to a free local model via Ollama. Internal benchmarks across 10,000 routed queries put typical usage at $4-6 per month in tokens for a heavy user (Kent internal data, 2026).

For the ten-person team: Kent Teams at $35 per seat per month, billed annually with the 10% discount, is $3,780 per year -- against $12,720 for the fragmented stack. An $8,940 annual dividend, before counting a single reclaimed hour.

4.2 The Hours

The time dividend does not require believing our estimates wholesale. Take only the copy-paste tax -- the one with the cleanest external measurement at 3.2 hours per week -- and assume connectors eliminate two-thirds of it. That is 2.1 hours per week per person, which matches what Kent's internal usage analytics measure for connector-heavy users (Kent internal data, 2026). At the BLS median wage, 2.1 hours per week is roughly $4,100 per year per person. For the ten-person team, the reclaimed copy-paste time alone is worth more than four times the entire Kent subscription.

4.3 The Meter You Stop Feeding

One structural point separates consolidation-into-Kent from consolidation-into-another-subscription: with your own API keys, you pay for usage, not for access. The idle month costs zero. The seven-app stack bills you for August whether you worked or vacationed. Over a career, the difference between renting access and paying for use is not rounding error.

5. The Soft Dividend

5.1 One Mental Model

Every tool in a stack charges cognitive rent: its own interface, its own shortcuts, its own quirks, its own place-where-that-setting-lives. Seven tools is seven mental models maintained in parallel, and the maintenance is exactly the kind of low-grade background load that the task-switching literature says degrades deep work.

Consolidated, the entire intelligence layer collapses to one gesture: highlight anything, anywhere on your system, press Ctrl+Shift+Space, act. The same gesture for an email, a contract clause, a stack trace, a paragraph of your own draft. Muscle memory replaces navigation. Our companion paper The Cost of the Tab quantifies what the browser-tab pilgrimage costs; the soft dividend of consolidation is that the pilgrimage simply stops existing.

5.2 One Brain Instead of Seven Amnesias

This is the dividend that compounds. In the fragmented stack, your working knowledge is split seven ways, and each fragment is amnesiac about the others: the chat AI does not know what the transcriber heard, the note AI does not know what you emailed, and none of them remember last quarter. Every fragment resets; nothing accumulates.

Consolidated, every skill execution, every dropped file, every connector query feeds one knowledge graph on your own disk. The meeting transcript connects to the client entity, which connects to the contract, which connects to the thing you told them in March. Ask one question and the answer draws on all of it. Our papers The Memory Trap and The Appreciating Brain make the full argument; the one-sentence version is that a knowledge asset only appreciates if it is in one place, and a stack guarantees it never is.

5.3 Fewer Parties Holding Your Data

Seven tools is seven privacy policies, seven retention schedules, seven breach surfaces, and seven sub-processor lists you have never read. Consolidating the intelligence layer shrinks the exposure arithmetic: fewer vendors see your queries, and in Kent's private mode -- local inference through Ollama, zero outbound requests -- the number of external parties processing a sensitive query is zero. For regulated professionals, that number is not a preference; it is the compliance story. One vendor conversation instead of seven is also, for anyone who has assembled a DPA file, a soft benefit with a distinctly hard edge.

5.4 The End of Stack Fatigue

The least measurable benefit is the one users mention most: the low-grade anxiety of the sprawling stack -- the tool you pay for but forgot, the export you keep meaning to do, the nagging sense that the answer exists somewhere in silo four -- goes quiet. Evaluation fatigue ends too. New model ships? In the fragmented world, that is a new subscription decision. In the consolidated world, it is a dropdown: Kent added it, your skills run on it, your brain persists beneath it.

6. The Consolidation Playbook

6.1 The 30-Day Cutover

Week 1 -- Audit. List every subscription with its annual cost and its last-used date. The Zylo statistic predicts what you will find: about half the list is idle. Cancel the idle half immediately; that dividend requires no migration at all.

Week 2 -- Connect. Install Kent, connect the record layer (Gmail, Drive, Calendar, Notion, databases), and add API keys for the providers you already trust. Drop the exports from any tool you plan to retire -- transcripts, notes, documents -- into Kent so their contents join the graph before the subscription lapses.

Week 3 -- Parallel run. Route real work through Kent while the old stack idles. Rebuild your three most-used prompts as custom skills. This is the week that surfaces the exceptions -- the one workflow a legacy tool genuinely does better. Keep that tool. That is data, not failure.

Week 4 -- Retire. Cancel what the parallel run made redundant. Calendar the renewals you kept, so next year's audit is a fifteen-minute review instead of an archaeology dig.

6.2 What Not to Consolidate

An honest consolidation argument has to know where it stops.

Keep your systems of record. Gmail, Drive, Notion, your production databases -- Kent is designed to connect to them, not to become them. Migrating ten years of email into a new silo is how you build the next stack problem.

Keep genuinely specialized tools. A video editor, a CAD package, an IDE, a real accounting system -- deep vertical tools earn their seat. The consolidation target is the horizontal intelligence layer, where seven tools do the same dozen things.

Keep single-purpose tools that pass a simple test: it does something Kent plus a connector cannot, and you used it in the last 30 days. Anything that fails both halves of that test is paying rent on your attention and your card.

Consolidation is not the ideology that fewer is always better. It is the observation that this particular decade of AI tooling shipped the same intelligence a dozen times in a dozen wrappers, and that intelligence -- unlike records, unlike specialized craft tools -- consolidates with no loss and considerable gain.

Conclusion

The stack was never designed; it accumulated. Each tool solved a real problem on the day it was adopted, and the costs -- the subscriptions, the copy-paste hours, the seven-way split of your own accumulated knowledge -- arrived gradually enough that no single day ever demanded an audit.

Run the audit. The hard dividend is concrete: on the representative stack, $743-1,272 per person per year in retired subscriptions, a multiple of that in reclaimed hours, and for a ten-person team, nearly $9,000 annually before counting a minute of saved time. The soft dividend is larger and compounds: one gesture instead of seven interfaces, one appreciating knowledge graph instead of seven amnesiac silos, one vendor -- or, in private mode, zero -- holding your working mind.

The tools were never the point. The work was. Consolidate the layer that thinks, connect the layer that stores, and keep the craft tools that earn their seat. The dividend pays every year after.

References

1. Zylo. (2025). 2025 SaaS Management Index: License Utilization and Portfolio Benchmarks.

1. Okta. (2025). Businesses at Work 2025: App Adoption Trends.

1. Gartner. (2025). IT Spend Optimization: Software Portfolio Rationalization Benchmarks.

1. Microsoft. (2025). Work Trend Index 2025: The State of AI at Work.

1. Rubinstein, J.S., Meyer, D.E., & Evans, J.E. (2001). "Executive Control of Cognitive Processes in Task Switching." *Journal of Experimental Psychology: Human Perception and Performance*, 27(4), 763-797.

1. Mark, G., Gudith, D., & Klocke, U. (2008). "The Cost of Interrupted Work: More Speed and Stress." *Proceedings of CHI 2008*, ACM.

1. Bureau of Labor Statistics. (2025). Occupational Employment and Wage Statistics.

1. Forrester Research. (2025). The AI Platform Wave: Enterprise AI Platforms, Q3 2025.

1. Kent. (2026). Internal Usage Analytics: Routing Costs, Connector Time Savings, and Knowledge Graph Growth.

Kent Research, August 2026. Kent runs 13 built-in skills and unlimited custom skills against six AI providers, cloud or fully local, and connects natively to Gmail, Google Drive, Google Calendar, Notion, PostgreSQL, MySQL, SQLite, MongoDB, REST APIs, and MCP servers. Pricing comparisons use provider list prices as of March 2026. Nothing here constitutes financial advice. mykent.app