A standard is a decision about who gets the tool. Count the licenses and you learn how many people it actually reached. In most organizations that is a tenth to a fifth of the workforce, because a per-seat subscription stops being fundable long before it reaches everybody. The fair question follows: we standardized on Copilot, so why buy anything else?
Incumbent Suite & Local Client
Already Standardized on Copilot?
Where a Local Assistant Still Fits
The seats you already fund, the workforce they never reach, and the two-line-item case for a local AI client alongside the suite you standardized on.
Built from real buyer questions in our sales meetings
Direct Answer
Coexistence beats replacement. Standardizing on the suite you already own settles which assistant your knowledge workers open first. The decision leaves two gaps open: a per-seat price that keeps the license from reaching the whole workforce, and answers grounded in whatever the connected corpus holds rather than in a data set you curated and governed. The blend Iternal sees work keeps the seats that earn their keep on suite-integrated work, and puts a low-cost local client on everyone else.
The limit: nothing credits you for the seats you keep. Iternal is candid that its shipped commercial model assumes a custom standalone build and takes no account of licensing a customer already owns, so a coexistence plan earns no automatic credit for the subscription you keep. Run the two as separate line items, and justify the local client on its own merits: the employees it reaches, the material it may process, and the work it takes off the subscription.
Verify the arithmetic against your own rate card. Buyers named roughly $30 per seat per month most often as the price they resist, and Microsoft keeps reshaping tiers and metering agent workloads, so your renewal quote outranks any published figure. Iternal puts the amortized equivalent of its one-time per-device license at roughly $8 to $10 per user per month across a three-to-five-year device life, roughly a third of a per-seat subscription over three years.
Coverage and capability are separate questions. Whether an incumbent standard closes the case for a local client is a coverage question, and it is the one answered above. For more information on assembling the capability yourself, visit the build-or-buy page; for more information on commercial models, visit the licensing-models page.
The Coexistence Pattern: Which Seats Stay and Which Go Local
Replacement is the rare outcome. Iternal states that AirgapAI is not a one-to-one replacement for the suite assistant and that the two typically blend, and the deployments buyers described follow that shape: a few subscription seats for suite-integrated work, a one-time local license for the rest of the workforce. Some organizations kept as few as 10 or 15 licenses, describing splits of roughly 10/90 or 20/80. Draw the line by the work rather than the department:
| Work that keeps earning a subscription seat | Work that moves to a local client |
|---|---|
| Drafting and editing inside Word, Excel and Outlook, where the assistant sits in the document. | Answering from a curated, governed data set every employee can carry. |
| Call summaries written back into a file, plus bundled transcription minutes. | Material your policy will not allow into a cloud environment at all. |
| Agents and workflows built from the suite panel by the people who own those processes. | Work done on a plane, on a factory floor or at a disconnected site. |
| The funded fraction of staff whose output justifies a recurring charge. | Everyone the subscription was never funded to cover. |
For more information on the local client that covers the second column, visit AI Assist for the rest of the workforce.
The Cost Gap in Two Numbers
The number buyers resist is roughly $30 per seat per month. Lighter tiers were quoted lower, three-year totals including training and integration far higher, and Microsoft has since pushed a cheaper chat tier and metered agent workloads. Treat $30 as the working figure and your own renewal quote as decisive.
At that rate the arithmetic runs out before the workforce does. Organizations typically license 10 to 20% of staff, leaving the remainder with nothing. Agents add a second bill: building them requires an additional per-user license beyond the M365 entitlement, and agent workloads are metered per token.
The second number is a one-time per-device perpetual license. Iternal amortizes it across a three-to-five-year device life — the window in which the hardware gets replaced anyway — and puts the equivalent at roughly $8 to $10 per user per month, roughly a third of a per-seat subscription over three years. List price varies by edition, so the figure that matters is the one Iternal quotes for yours. Three questions close the distance, answered in writing:
Pin it down: questions for your evaluation
-
What is the current list price per device for the edition we would deploy, and what does it include?Which amortized monthly equivalent applies to your plan, rather than a range quoted from a different deal.
-
Which subscription seats does the local client take work away from, and which should stay?The seat count you keep funding, agreed before the renewal date rather than after it.
-
Which of our data classes can go into a local client that cannot go into a cloud assistant?The coverage gap in your own data classes, in writing, before a pilot starts.
What Buyers Say the Assistant They Already Own Does Not Do
The next reason comes from the people using the assistant. Buyers running an incumbent deployment described the same disappointments again and again, in their own words: the assistant did not accurately do the job asked and ignored repeated prompts to read all of the data; it could not handle more complex work such as editing emails; sessions ended without warning and forced starting over; transcripts had to be saved by hand or the work was lost; and memory turned out to mean personalization rather than a persistent memory that behaves like a brain.
Two qualifications bound that evidence. Those are user reports from a moment in time, and a suite assistant ships changes monthly, so treat each as a test to run in your own evaluation. More usefully, the cause often sits in the corpus rather than in the assistant: retrieval is garbage in, garbage out. Blockify cleans documents before they are uploaded into SharePoint, improving the answers your existing assistant returns, and the cleaned data set stays portable across tools.
The Microsoft-Shop Objection and the Gaps It Leaves Open
The objection arrives in a few reliable forms. We are a complete Microsoft shop and we are all about Copilot. We already use Copilot and chat, so why do we need something more. We are just going to go with Copilot as our AI approach. The affordable-alternative conversation gets harder once an organization has made its cloud and M365 decisions. Take all of that at face value, and three gaps still stand.
- Coverage. The standard reaches the funded fraction of staff. Everyone outside it works the way they worked before, which is where unapproved tools appear.
- Where processing happens. The suite assistant always performs some processing on servers in Microsoft’s environment, so material your policy bars from cloud processing stays barred. AirgapAI processes on the device and the data never leaves it, which is why buyers reach for it there.
- Connectivity. A cloud assistant goes quiet on a plane, in a plant with no outside access, or on a closed network. A local client keeps answering because the model and the data set already sit on the machine.
Give the Incumbent Its Due
A challenger that will not name what the incumbent does better has not earned the comparison. Three strengths are real, and they are why the recommendation keeps subscription seats rather than canceling them:
- Placement inside the work. The assistant lives in Outlook, Word and Excel, writes a call summary into a document on request, and coaches users toward better prompts while they type.
- Agents where the process owners already are. Agents and workflows can be built from the suite panel without opening a separate studio, and that capability has advanced quickly.
- Terms and tiering that hold up. Business plans carry intellectual property and confidentiality protection by default, and a US federal tier exists for content needing a more protective enclave. The disclaimer buyers sometimes quote applies to the free consumer plan, so no business buyer falls under it — an angle Iternal declines to use.
The counter is arithmetic rather than argument. They arrive per seat, per month, for the share of the workforce you can fund, and in a cloud environment. A one-time local license answers the other question: everyone else, and everything the cloud may not process. Iternal is equally candid about its own gap — the AirgapAI chat experience is very close to, and not yet equal to, the polish of the assistants people compare it against.
Answered elsewhere
- Whether to assemble the capability in-house from open-source parts — see the build-or-buy page.
- How to compare several products at once without adding tool sprawl — see the evaluation pillar.
- How subscription and one-time pricing compare as commercial models — see the licensing-models page.
- Whether a capable AI device pays for itself against a recurring plan — see the device economics page.
- What to do about staff using unapproved tools — see the workforce policy page.
FAQ
FAQ: An Incumbent Suite and a Local Assistant
Answer it with your license count. If the subscription covers everyone who needs an assistant and every class of material you work with, the standard has done its job. In most organizations it covers a tenth to a fifth of staff and stops at the cloud-processing boundary. A one-time per-device license covers the remainder at an amortized $8 to $10 per user per month across a three-to-five-year device life.
For drafting, summarizing and workflow inside Office it earns its seat, and business plans carry intellectual property and confidentiality protection by default. Two gaps stay open: the employees a per-seat budget never reaches, and material your policy bars from a cloud environment, since some processing always happens on servers you do not run. Enterprise coverage means both halves — the assistant inside the suite, and a local client for everyone the subscription cannot take.
Roughly $30 per seat per month is the figure buyers named most often as the one they resist, with lighter tiers quoted lower and three-year totals including training and integration far higher. Microsoft keeps reshaping tiers and has metered agent workloads, so price your own rate card. At $30 a seat, funding the entire workforce is what stalls: organizations typically license 10 to 20% of staff, and agents add a second bill on top.
Yes. Iternal states that AirgapAI runs at the same time as the suite assistant under standard Windows resource provisioning, and builds it as a complementary layer rather than a rip-and-replace. A sequence buyers use: prove the local client on one machine, then keep the subscription for email, documents and agents.
Often the corpus is the fix. Buyers describe an assistant ignoring repeated prompts to read all of the data and struggling with complex edits, and retrieval is garbage in, garbage out — source content has to be cleaned before any retrieval system receives it. Blockify cleans documents before they are uploaded into SharePoint, improving the answers your existing assistant returns. Add a local client when the gap is coverage or data class rather than answer quality.
Price the Second Line Item on Its Own
The strongest case for a local client never asks you to cancel anything. Count the employees your subscription reaches, count the material it may not process, and price the remainder as its own line. If that line stands up, the standard you set stays where it is, and everybody else finally gets an assistant.