Start with AI already included in your software when the approved capability can handle the actual task, with checking and support your team can sustain. Before buying another tool, establish what the employee really has access to. A subscription name, an AI button and a successful demonstration can describe three different experiences.
For an IT or department lead, the first useful question is specific: can this person, in this account and application, produce the result the next person needs? If the answer is unclear, diagnose the gap before comparing purchases. If the existing service is unsuitable for the information or required operation, move directly to an approved alternative.
Establish what the employee actually has
Ask the software owner to confirm the entitlement assigned to the employee, rather than starting from a product-family name. Then open the application and surface where the work would happen. A general chat window, an assistant inside a document and a service searching organizational records should not be assumed to have the same context.
Current vendor guidance shows why that distinction matters. Microsoft's Copilot overview, updated September 10, 2026, describes experiences that vary by license, application and tenant. Some use provided content or supported application context; broader organizational grounding is a separate capability. Outlook has its own included-context provisions, so “no paid add-on means no work data” is too broad a rule.
Microsoft also explains how to identify the named Copilot experience. Check the employee's actual label and assignment before reproducing a demonstration delivered with a different license.
The same care applies beyond Microsoft. Google's work-account guidance for Gemini Apps distinguishes access by Workspace license and notes that Gemini Apps and Gemini in Workspace applications have separate limits. Access to one surface does not establish identical availability elsewhere. Recheck current guidance with your administrator before designing around a feature or allowance.
Keep a short record of the test conditions:
- Account and assignment. The approved work account, relevant edition or add-on, and who confirmed it.
- Application and feature. Where the employee starts and which capability they use.
- Source context. The files, messages or records the capability is expected to use, and how they become available.
- Required result. What the recipient must be able to rely on, including important omissions and exceptions.
This is a record of what you are evaluating. Permission to process the information still comes from the organization's rules. The employee AI policy guide explains how to define approved accounts, inputs and tasks.
Test the missing detail before replacing the tool
Suppose Jamie coordinates speakers for a professional conference. He needs a briefing that gives a speaker the correct session time, room and arrival instructions. The approved program contains the original room; a later email confirms a room change. His colleague checks the briefing before it goes to the speaker.
If Jamie pastes only the program into a chat and receives the old room, that result cannot establish whether the service can find the later email. It has tested drafting from the program. The missing email may reflect the chosen surface, unavailable context or the way Jamie supplied the sources.
First establish which task matters. If Jamie can reasonably supply the two approved sources, test that drafting arrangement. If the requirement is to find all relevant changes across a mailbox without manually collecting them, test that retrieval explicitly in the supported, approved experience. Do not quietly replace one requirement with the other to obtain a better result.
For the briefing test, Jamie and his reviewer should:
- Prepare the evidence. Keep the approved program and confirmed room-change message, with their dates, available for checking.
- Check source use. Ask for the briefing and supporting source references where available. Open the references; their presence alone does not establish that the right details were used.
- Verify the handoff. Confirm the latest room, unchanged session time and supported arrival instructions. Leave unconfirmed details unresolved rather than letting the draft invent them.
- Record the remaining work. Note source collection, corrections and review before the colleague accepts the briefing. Sending the message remains a separate action.
Use other representative briefings and meaningful exceptions before expanding the arrangement. One correct result does not establish reliable handling of a late change, conflicting instruction or unavailable source. The human-review guide explains how to make the recipient's check specific.
Give different findings different next steps
A disappointing demonstration is evidence about that test. Make its conditions clear enough that someone can decide what to change.
| Finding | What remains unresolved | Next step |
|---|---|---|
The employee cannot open the feature | Assignment, availability or configuration | Ask the software owner to confirm access before judging output quality. |
The draft omits the later room-change email | Whether the correct source was available and used | Check the supported context, then repeat the same briefing test. |
The draft gives the wrong room despite using the current message | Accuracy on an important requirement | Keep the manual check and investigate whether the failure can be corrected consistently. |
The briefing is correct but source gathering and review remain burdensome | Whether the whole task improves | Compare the complete effort with the existing process and alternatives. |
The briefing is useful under the agreed conditions | Whether the result holds across representative work | Retain the bounded use, owner and retest conditions before expanding it. |
Do not broaden access to make a demonstration look better. Ask the information and technical owners to resolve a legitimate access problem through the normal process. Equally, do not keep changing prompts indefinitely when the required operation is unavailable or the checking burden defeats the purpose.
Follow the work after the useful output
In an August 2026 Reddit discussion, a Reddit user in a customer-facing project role described preparing a time report from project codes, meetings, email and chat. They still entered the result into their company's business-management system manually. The useful output and the remaining transfer were separate parts of the job.
Another Reddit user in the same discussion reported that a timesheet attempt worked initially but later struggled to retrieve and summarize a week's emails. These are individual accounts, with unknown configurations and no independently verified results. They do not establish a product ranking. They show why “it prepares my report” needs questions about source completeness and what happens afterward.
Name the requirement behind another purchase
For Jamie, the equivalent question is whether assembling and checking the briefing leaves a worthwhile improvement once the correct room reaches the speaker. Count that work even when the license adds no new charge. Have an owner for changed templates, access problems and recurring errors; an included feature does not remove those responsibilities.
When another tool deserves evaluation, state the unmet requirement precisely. “Find confirmed speaker changes across the approved correspondence and preserve source references” is more useful than “better AI”. Compare options against that requirement and the same acceptance conditions. The buy, build or combine guide covers the wider integration, support and exit commitments.
Before making a purchase request, leave a short conclusion with the software owner: what was tested, what worked, what remains unmet and which change would justify another test. An existing subscription earns its place through usable work. Its presence on the renewal invoice cannot decide the question.
Return to the tools and workflows guide for the broader operating arrangement.
Questions about AI included in existing software
Does included AI mean the task is free?
An included AI entitlement may avoid an additional license charge for a particular capability. It does not remove source preparation, configuration, checking, support or manual transfer. Confirm the actual contract and assignment with the software owner, then compare the complete effort needed to produce accepted work. Do not infer that another capability or higher allowance is included because the product name is familiar.
Does a missing file prove the AI tool cannot do the task?
No. A missing detail may reflect unavailable source context, an unsupported retrieval method or an incorrect answer despite suitable sources. Establish which account, application and information the employee tested before drawing a conclusion. In the conference briefing example, drafting from a pasted program is different from finding a later email. Both are testable, but they answer different questions.
Should every employee receive a paid AI add-on?
Decide from the capability each role needs and the evidence from its work. A successful demonstration under a paid license does not establish that the current entitlement provides the same function, or that every employee needs the upgrade. Confirm the required application and source context, test representative tasks with appropriate review, and compare the full commitment before expanding assignments.
When should we stop testing the included feature?
Stop when the existing service is unsuitable for the approved input or required operation, or when repeated representative tests leave unacceptable errors or impractical checking. Record the unmet requirement and what evidence led to that conclusion. Keep work moving through the established process while evaluating an approved alternative. The buying and building comparison explains how to compare options on the same finished task.
