How to audit the subscriptions a team already pays for, which ones to keep as exceptions, and what disappears when an account is closed
Aug 18, 2026 · 13 min read
A team paying for two or more overlapping AI products should audit them before cancelling anything, then move to one main workspace and keep a short list of exceptions. The exceptions are the accounts where the native product is part of the work, such as Gemini inside Gmail and Docs, or a project full of history that cannot be exported usefully. The eight workspaces compared on the same criteria below are Playgram, WorkLLM, nexos.ai, Langdock, TeamAI, Aymo, Magai and TypingMind.
The spending is hard to see before you look for it. Zylo's 2026 index found business units controlling 81 per cent of software spending while IT directly manages 15 per cent, and an average of 36 per cent of licenses sitting unused. The same index puts expensed software spending up 267 per cent year on year, with ChatGPT the most expensed application in its dataset32. BetterCloud found unused or underused applications and licenses given as the reason for consolidating by 21 per cent of the organisations it surveyed33.
This guide sets out how to run the audit, how to sort every account into keep, replace or review, and what is actually lost when one is closed. It compares eight multi-model workspaces on the same criteria and ends with a two-week pilot plan.
You own the renewals and the access lists, and neither of you can see the whole picture.
People claim their own subscriptions back, so nobody knows what the company actually pays for.
Marketing bought one provider and engineering another, and the work moves between them.
Fewer than five people on one provider, with nothing overlapping to remove.
Separate accounts spread the same work across four bills, four histories and four owners, and each split has its own cost.
Separate subscriptions multiply per person even when only a few people need every provider. As an example, four single-vendor team plans came to about $101 per person a month at July 2026 list prices. That total takes ChatGPT Business at $25, Claude Team at $25, Gemini Business at $21 and Grok Business at $301, 2, 3, 4. Read it as an illustration rather than a rate, because a cheaper mix is easy to assemble. The list price is only part of it, though, because minimum-seat commitments, duplicated capabilities and seats bought just in case all sit on top. Zylo's 2026 index puts average unused license volume at 36 per cent, and BetterCloud found 21 per cent of surveyed organisations naming unused or underused licenses as their reason to consolidate32, 33.
People switch tabs, copy prompts, download a file from one product and upload it to the next, then rebuild the same instructions in each one. A single research task can start in one vendor, move to another for drafting and finish in a third for analysis. Every one of those transfers is manual, because separate accounts have no way to hand work to each other.
Chat history belongs to one product and one account, so the project instructions, uploaded files, corrections and decisions inside it do not follow the person to another model. Even a move inside one vendor can be blocked, and Anthropic states that an individual account stays separate from a Team organisation with no way to transfer content between them36. That matters most at the moment of consolidating, because the knowledge you are about to cancel is the same knowledge the new product needs on day one. Exports help, and they arrive as archives rather than as conversations another tool can resume.
Separate products mean separate invoices, owners, access lists, privacy settings and usage reports, so nobody can answer which models the team uses or which seats are idle. Expensed personal accounts make it harder still, because they never appear on the vendor list at all. When a person leaves, their chats and prompts leave with them, and access has to be removed one console at a time. Zylo's 2026 index found business units controlling 81 per cent of software spending against 15 per cent managed directly by IT32.
Five checks on any product you are about to consolidate into. The first four decide whether the old accounts can close at all, and the fifth decides what the move costs.
The product should carry suitable models from OpenAI, Anthropic, Google and xAI, and say plainly whether their usage is included or billed through your own provider keys. A person should also be able to change model inside a thread and keep working, which is the behaviour vendors document least.
List what the old subscriptions did beyond chat: web research with citations you can open, document and spreadsheet work, image generation, video generation, code review, and chats that leave nothing behind. Anything on that list the new product lacks is a subscription you keep.
Files, instructions, approved outputs and decisions should belong to projects and folders the team owns, with shared chats and a named owner for each. The exported material lands there, and without somewhere to put it a consolidation only moves the same scattered work into one login.
An admin should see usage by person, model and billing period, cap spending before an overage rather than after one, restrict models, and set who can reach which knowledge source. Adding and removing a person should happen once rather than in four consoles.
Check what you are buying, because the shapes are pooled credits, included usage, your own API keys billed by the provider, or per-seat limits that match real activity. Price your mix under each of them, count the people who barely sign in, and check what happens when the included usage runs out.
The multi-model workspaces a team is most likely to consolidate into, on the same criteria and to one standard. Where a vendor does not document something, the cell says so.
This table compares multi-model team workspaces with each other. The single-vendor plans a workspace replaces are priced further down, under 'Priced per seat', and are not rows here. Pricing is the lowest-priced paid plan that covers five users, at the monthly rate, so a product whose entry plan holds fewer than five people is shown on the plan that holds them. Each cell cites the page that documents that cell rather than one pricing page per row. Figures checked August 2026 against each provider's own pages, and cells marked 'Manual test required' could not be confirmed from public documentation.
The same products on the criteria that decide daily use: what each one does besides chat, what it connects to, what an admin can see and cap, and where your data goes.
These criteria decide daily use more than the model list does, and vendors document them very unevenly. 'Not publicly documented' means the official sources checked did not state it, and it does not mean the feature is absent, so read those cells as questions to put to the vendor. Checked August 2026.
The published per-seat price of each major single-vendor team plan, billed monthly. Most teams start a consolidation after two or three of these are already on the invoice.
Buying all four for one person came to about $101 a month at July 2026 list prices. Read that as one example stack rather than a going rate, because a team can assemble a cheaper mix and the four plans do not buy the same amount of use. A Google Workspace fee is a different case, because it pays for email, storage and documents alongside its AI features, so only the part you would stop paying for belongs in the comparison. Figures checked July 2026, and the estimator further down compares this shape with usage pricing on your own numbers.
Two of these show up on an invoice and two are found only by reading expense claims, which is the order most teams meet them in.
Six destinations for the spending you are about to reorganise. The workspace options come first because they are what most consolidations move towards.
The team moves generic chat, drafting, research and file work into a single product and keeps a short exception list. Pricing shapes differ, so WorkLLM and Langdock charge per seat while TeamAI and Aymo sell workspace capacity and Playgram sells credits6, 13, 17, 21.
Best for: Teams of five and up paying for two or more products.
Strengths
Trade-offs
The same move, plus the standing decisions and reference material saved where the whole team can retrieve them. It is the version that keeps what the closed accounts knew, and it is also the one with the most to verify before you trust it.
Best for: Teams that reapply the same context across projects.
Strengths
Trade-offs
Standardise on a single vendor and cancel the rest. It is the simplest structure to buy and to explain, and it stops the context fragmenting immediately because only one product holds it.
Best for: Teams whose work fits inside one model family.
Strengths
Trade-offs
Keep the plans and fix only the ownership, so procurement moves the personal accounts onto team plans and assigns an owner to each. The spending stays where it is, and the governance problem gets solved without a migration.
Best for: Teams that use each vendor's own tools every week.
Strengths
Trade-offs
People keep expensing their own accounts. It needs no decision and no project, which is exactly why the spending grows without anyone seeing it. Zylo found expensed software spending up 267 per cent year on year, with ChatGPT the most expensed application in its dataset32.
Best for: One or two independent users with no shared work.
Strengths
Trade-offs
Engineers replace the subscriptions with an interface of your own, so routing, retention and logging are decisions the team makes. The subscription line disappears and reappears as API consumption, hosting, security, evaluation and maintenance.
Best for: Engineering-led teams with a defined application.
Strengths
Trade-offs
The audit is the first project the new workspace does, and it runs across models with one shared context, so finance and IT read the same record afterwards.
Each stage reads the same audit project, so nobody retypes the vendor list or the evaluation criteria. A person approves the classification before anything is cancelled and sends weak reasoning back to the classify stage. The approved audit, the migration checklist and the exception list then stay in the project for whoever runs the cancellations.
The accounts you are closing hold instructions and decisions the team still needs, so the replacement has to be able to keep them and to say who can read them.
A context window is how much text a model reads in one request, and it empties when the conversation ends. Chat history keeps that conversation so a person can reopen it. Memory is context stored outside the thread and retrieved later, and it is the only one of the three that survives a closure.
Some keep history and nothing more. Some let you upload files and write instructions, where retrieval is automatic once you have built it by hand. Some learn automatically and keep it to one person. Only the last shape, saved where teammates can retrieve it, replaces what the old accounts knew.
Shared memory needs boundaries: what belongs to one person, what belongs to a project and its members, and what the organisation should see. Products draw these lines differently and some draw only one, so ask which boundaries exist before you move a closed account's knowledge into them.
Will the replacement hold the instructions, decisions and reusable facts that live in the accounts you are closing, and can you verify who retrieves them? Check that someone can see what was saved, correct a wrong entry and delete one, because a migrated assumption nobody can edit becomes permanent.
A vendor-neutral plan that starts with an inventory and ends with a cancellation you can defend. It runs about two weeks.
For each subscription record the plan and billing cycle, the renewal date and cancellation window, and the owner and payment source. Add assigned against active users, the models and tools actually used, and any integrations or API keys. Pull card statements and expense claims too, because reimbursed personal accounts are the ones missing from the vendor list.
Keep an account when a measured workflow depends on a native integration, a specialist model, a contractual deployment or content that cannot be reproduced. Replace it when the work is generic chat, drafting, research or file analysis. Review it when usage is low but the account owns projects, agents or prompts somebody still needs.
Export before anything is cancelled, then open the file and confirm the projects and attachments are in it. A Claude organisation export has to be started by the Primary Owner and its link expires after 24 hours, and anything already deleted is missing from it35. Give every export a named owner and a place to live.
Use real work with the same source material and acceptance criteria, such as cited research, long-document analysis, spreadsheet review, campaign images, code review, or a task that needs the company instructions. Include one workflow that moves between models and one that a second teammate has to continue.
Track time to a useful answer, manual edits, repeated uploads, successful model changes, active against assigned users, and how long onboarding took. Then confirm training terms, processing regions, permissions, retention and pre-spend limits, and test what happens to a person's chats and files when they are removed from the team.
Consolidation works when it takes away duplicated access without taking away a capability, a context or an owner somebody still needs. For most teams the right answer is one main workspace plus a small exception list. The exceptions are usually a native integration such as Gemini inside Gmail and Docs, a specialist model, or an account whose content cannot be reproduced anywhere else.
Four limits apply. Plans with similar names include very different usage, tools and controls, so a comparison by plan name alone is worthless. Shared memory helps only when its saving, retrieval and permissions are clear. Cancelling and deleting are separate actions, and only one of them is reversible. And prices, model catalogues and plan caps all move faster than any audit document.
So the decision comes from the inventory and a pilot on your own work, rather than from a feature count. Run the audit first, test the replacement on the workflows that actually pay for themselves, and keep every export until the new setup has survived a full month of real use.
Upgrade as needed, and only pay for what you actually use
Save ~17% with the annual plan
Pro
Perfect for small and medium teams
Unlimited users & infinite memory
Multi-LLM chats
Granular access control to models
EU data residency
Ultra
Best for large, growing teams
Unlimited users & infinite memory
Multi-LLM chats
Unlimited use of DeepSeek V4 Flash
Granular access control to models
Choose US or EU data residency
Enterprise
Get in touch
For organizations with advanced needs
Unlimited users & SSO
Priority Support
Unlimited use of DeepSeek V4 Flash
Granular access control to models
Choose US or EU data residency
30-days money back guarantee
Playgram will automatically choose the most cost-efficient model suitable for the task. It will be chosen by users in approximately 80% of requests. Your models for the remaining 20%:
If you bought each separately: