--- title: The integration tax: what five separate tools really cost canonical: https://getcyril.com/blog/the-integration-tax/ published: 2026-08-25 author: Sam Akbari language: en --- # The integration tax: what five separate tools really cost Every growing company arrives at the same stack by the same reasonable steps. Sales picks a CRM. Support picks a helpdesk. The team picks a wiki because the CRM's notes are hopeless. Delivery picks a project tool. Finance picks something that does invoices properly. Every one of those decisions was correct on the day it was made. What nobody prices at the time is the connective tissue between them, because it never arrives as an invoice. It arrives as a slow tax on everything. ## The model, and its assumptions A 40-person services or software business. Five categories of tool: CRM, helpdesk, wiki/docs, project management, finance and invoicing. Not everyone needs every seat — assume 40 CRM-adjacent seats, 12 support seats, 40 docs seats, 30 project seats, 6 finance seats. These are illustrative mid-market list-price assumptions in US dollars per month. Substitute your own; the conclusion is not sensitive to being twenty per cent out on any single line. ### Line 1 — Licences Assume a blended $22 per seat per month across the five products, at the seat counts above (128 paid seats in total). **≈ $2,800 per month.** This is the number that appears in the budget, and the number every renewal conversation is about. ### Line 2 — Connective software Two or three of those five need to talk to each other, so you buy an automation or iPaaS product to move records between them, plus one or two paid connectors. Assume a mid-tier plan with enough task volume for a company of this size. **≈ $400 per month.** ### Line 3 — The administration of the seams Someone maintains this. Field mappings drift, a connector breaks after a vendor's release, a duplicate rule needs adjusting, onboarding a new hire means five accounts and five permission sets and offboarding means remembering all five. Assume 0.35 of a full-time equivalent at a $70,000 fully loaded salary. This is the assumption most worth arguing with, and in my experience it is more often too low than too high — it is rarely one person's job, which is exactly why it is invisible. **≈ $2,040 per month.** ### Line 4 — Reconciliation and error The same customer exists five times. The versions disagree. Somebody quotes a renewal date from the CRM that finance already changed. Support promises something the project team cannot deliver because the two systems do not share a status. Assume 30 minutes per employee per week lost to looking things up in the wrong place, checking which system is right, and re-entering what a sync did not carry — 40 people, at a blended $34 per hour. **≈ $2,950 per month.** ### Line 5 — The AI tax This is the new line, and the reason the arithmetic has changed since 2023. Each of those five vendors now sells an AI add-on, typically $15–30 per seat per month. Each of them can only see its own silo. So you either pay five times for five assistants that each answer a fifth of your questions, or you buy one external assistant and pay again to give it access — an integration per tool, each with its own auth, rate limit, data shape and latency. Assume conservatively that you buy AI on only two of the five, at $20 per seat, for the 52 seats where it is most obviously useful. **≈ $1,040 per month** — and this is the line that buys you the least, because the questions worth asking are precisely the ones that cross the silos. ### The total | Line | Monthly | Share | |---|---|---| | 1. Licences | $2,800 | 30% | | 2. Connective software | $400 | 4% | | 3. Administration of the seams | $2,040 | 22% | | 4. Reconciliation and error | $2,950 | 32% | | 5. AI add-ons that see one silo | $1,040 | 11% | | **Total** | **$9,230** | | About **$110,000 a year**, of which the subscriptions everyone argues about are under a third. ## The part the table cannot show Lines 3 and 4 are real money and they are still not the worst of it. The compounding cost is the questions that never get asked. "Which accounts are at risk this quarter, and why?" requires the pipeline, the tickets, the delivery status and the invoices in one place. In a five-tool stack that is a project — someone exports four things and builds a spreadsheet — so it happens quarterly at best, usually when something has already gone wrong. The question that would have been most valuable weekly gets asked four times a year. That is a strategic cost with no line item, and it is the one that AI has made suddenly visible. A model is very good at synthesis and completely dependent on what it can see. Give it one silo and it will confidently answer a fifth of your question. Line 5 is not just poor value for money; it is the line where you pay to be told an incomplete answer in a fluent voice. ## What actually removes the tax Not "fewer tools" as a discipline — companies try that and drift back within a year, because each individual decision to add one is correct. What removes it is the thing underneath: **one data model, one permission model, one audit log**. When sales, support, docs, projects and finance are rows on the same graph rather than five products with connectors between them: - Line 2 goes to zero. There are no seams to bridge. - Line 3 collapses to ordinary administration — one account, one permission set, one offboarding. - Line 4 shrinks to the residue of ordinary human error, because there is one version of each customer. - Line 5 changes character entirely. The AI is not an add-on per silo, it is a layer under all of them, and it answers the cross-company question in one query because there is only one place to look. That is what Cyril is: sales, projects, support, docs and financials on a single graph, with the AI underneath rather than bolted to a corner of it. Per seat, no base platform fee, and data export always free — because a platform that is genuinely cheaper should not need a lock to keep you. ## Run the model on your own numbers Two adjustments matter far more than the rest. **Your line 3.** Ask who fixed the last broken sync, and how long it took. Multiply honestly. **Your line 4.** For one week, ask the team to note every time they had to check a second system to trust the first. Most companies find the 30 minutes a week assumption conservative. If those two lines together come to less than your licence spend, your stack is unusually well behaved and consolidation is a preference rather than a saving. If they come to more — which is the common case — then the subscription line you have been negotiating was never the thing costing you money. ## Common questions ### Is this based on a survey? No, and it should not be read as one. It is an open model with every assumption stated so it can be argued with, which is more useful than a survey average that hides its inputs. Change any number and the shape holds: connective cost exceeds licence cost. ### Does this apply below 40 people? The absolute number falls and the ratio gets worse. A team of ten has the same five seams and no one whose job includes maintaining them, so line 3 is paid in founder time — the most expensive hour in the business. ### What about companies that have already built good integrations? Then you have converted line 2 into a capital cost and kept an ongoing maintenance liability. Well-built integrations reduce line 4 and increase line 3. They cannot fix line 5, because integrations move records — they do not create a shared model for a question to be asked against. ### Is one platform not just a different kind of lock-in? It is, unless leaving is genuinely free. That is why the honest version of this argument has to come with free data export as a standing commitment rather than a plan feature. Ask any vendor making this pitch, including this one, what it costs to get everything out. --- If you want to be among the first to use Cyril, [join the waitlist](/waitlist/).