Signal Brief · the newsletter this call came out of
Five new AI tools every weekday morning, for B2B SaaS product managers.
One line on what each tool does, one line on why it matters for a PM, and the link. Give us your email and this morning’s issue — all five tools — lands in your inbox. No card, no password, nothing to install.
$10/month or $96/year after 7 free days, cancel in one click
Do-not-adopt call · 9 August 2026· free to read, no signup
Should we buy Glean? Not until the bundled assistants have failed
Enterprise AI search is the easiest six-figure line item to justify on a slide and the hardest to prove afterwards. The uncomfortable part of the business case is that vendors you already pay for shipped most of the job at no extra cost.
~7 min read · reviewed 2026-08-10· no sponsorships, no affiliate links
The call, up front
Do not adopt yet. Score fifty real questions against the AI already bundled into your suite — that is the baseline Glean has to beat, and almost nobody measures it.
Why not. Glean publishes no list price — its own documentation describes a credit-based Enterprise Flex model, so the real number arrives after a procurement cycle and usually carries a seat minimum. Meanwhile the specific job being bought (ask a question in plain English, get an answer grounded in company documents, with a citation) is already sold to you: Google stopped selling Gemini as a separate Workspace add-on in early 2025 and folded it into Business and Enterprise plans, Microsoft 365 Copilot does the equivalent inside its own suite, and Slack's June 2025 packaging change made conversation and thread summaries available across paid plans. Buying before scoring those means paying a second time for retrieval over the same corpus, and inheriting a second permissions model that has to stay in sync with the first.
Instead. Take fifty questions your team actually asks — where is the current pricing deck, what did we tell this customer in March, which runbook covers this alert — and score them against the assistant bundled with whichever suite holds most of your documents. Fix what turns out to be an information-architecture problem (stale duplicates nobody archived, three versions of the same wiki, permissions nobody rationalised), then re-score. What still fails after that is the gap Glean has to be worth, and now you can price it against a measurement instead of a demo.
Spend avoided: Roughly $50k–$250k/yr depending on headcount and the seat minimum you are quoted — Glean publishes no list price, so treat any figure before a quote as unverified
Written against this stack: Google Workspace, Slack on a paid plan, Jira and Confluence, Notion for product docs, Zendesk; ~400 people across four time zones.
A library that says no to everything is as useless as one that says yes to everything. These are the conditions under which this purchase is right — and if one of them describes you, stop reading and go and buy it.
Your documents are genuinely spread across more than eight systems, several of which the bundled assistants cannot index. Cross-suite fragmentation is the one problem a dedicated search vendor solves that a suite assistant structurally cannot.
You have already run the fifty-question test, fixed the obvious information-architecture rot, and the bundled assistant still fails a third of them. That is a measured gap, and a measured gap is a defensible purchase.
Search is being bought as the platform for internal agents rather than as a search box, and you have named the first two agents plus who owns them. Buying a platform for unnamed future agents is how this becomes shelfware with an auto-renewal.
01— Detail
The baseline nobody measures
Almost every AI search business case is written against the wrong comparator. It compares the vendor's demo to keyword search in Confluence circa 2019, when the honest comparator is the assistant already included in the suite where your documents actually live. That comparison is unflattering to the incumbent bundle in some orgs and devastating to the six-figure proposal in others — and the only way to know which you are is to score it.
Fifty questions is enough. Write them down as real questions with a known correct answer and a known source document, weighted toward the ones people ask repeatedly in Slack because they cannot find the answer themselves. Run them through the bundled assistant, unassisted, and record three outcomes per question: correct with a usable citation, plausible but wrong, or refused. The middle bucket is the one that matters, because a confidently wrong answer with a citation to a superseded document costs more than no answer at all.
Most orgs that run this test discover the failures cluster: they are not distributed evenly across the corpus, they concentrate in places where three versions of a document exist and nothing marks which one is current. That is not a retrieval problem any vendor fixes for you.
02— Detail
What a search vendor cannot fix
Every product in this category, Glean included, is honest about depending on your permissions model — and that dependency is where deployments stall. The index has to respect who can see what, which means reconciling access-control decisions made independently in Workspace, Jira, Confluence, Notion and your support desk. Any inconsistency between those systems becomes either a leak or a false negative, and you will find out which during a security review that was not on the project plan.
The second thing no vendor fixes is currency. Retrieval ranks by relevance, not by whether a document is still true. If your wiki keeps the 2024 onboarding runbook alongside the 2026 one, a very good search engine will confidently return either. Archiving hygiene is unglamorous, uncharged-for, and the highest-leverage work available in the fortnight after the test — which is also why it never wins against a purchase order in a planning meeting.
None of this makes the category worthless. It makes the sequence non-negotiable: measure the bundle, fix the corpus, re-measure, then buy against the residual. Done in that order a search purchase is defensible at renewal. Done in the reverse order you have bought a very fast way to retrieve documents nobody has curated.
Score the bundled assistant on fifty real questions before you take a single vendor call.
Count how many failures are stale-duplicate problems rather than retrieval problems. That ratio is your business case, in either direction.
Ask any vendor to run their trial against your real permissions, not a demo tenant. Permissions are where the deployment schedule dies.
The cheaper path
What to do with the money instead.
Every entry is a real product or a real capability with a link to the vendor’s own page. Where a vendor publishes a price we say so and date it; where they do not, we describe the posture rather than invent a number.
Google's assistant across Gmail, Docs, Drive and Meet, grounded in the Workspace content a user already has access to.
Why it covers the job. If Workspace holds most of your documents, this covers the largest share of the questions in your test set with no new vendor, no new permissions model and no procurement cycle.
Cost: Included in Workspace Business and Enterprise plans — Google stopped selling Gemini as a separate add-on in early 2025. Verify your plan's entitlements in the admin console. · workspace.google.com
Conversation and thread summaries, recaps and search inside Slack itself.
Why it covers the job. A large fraction of “where is that” questions are really “what was decided in that channel”, which is a Slack question, not a document question.
Cost: Summaries and huddle notes are available on paid Slack plans following the June 2025 packaging change; the richer AI features sit on higher tiers. Check Slack's pricing page for your plan. · slack.com
Microsoft's assistant across Teams, Outlook, SharePoint and Office, grounded in Microsoft Graph content.
Why it covers the job. The equivalent baseline for a Microsoft-centric org. If you are already paying for Copilot seats, you have bought a search product and should score it before buying a second.
Cost: Per-seat add-on to Microsoft 365, published on Microsoft's site; the relevant question is whether you already own seats, not what they list at. · microsoft.com
Keeping this honest
What would change our mind.
If one of these has happened since 2026-08-10, this page is out of date — tell us and we will change it. Pricing and plan contents rot fastest, so check the vendor’s current page before you budget either way.
Published list pricing from Glean. A quoted-only price is not a reason to reject a product, but it does mean nobody can compare it honestly against a bundle they already own, and that opacity is doing work for the vendor.
A scored, reproducible comparison of a dedicated search vendor against suite assistants on the same corpus and the same permissions. We have not seen one that survives scrutiny, in either direction.
Evidence that suite assistants systematically fail on cross-suite questions in a way that document hygiene cannot mitigate. That is the strongest argument for the category, and it is a measurable claim rather than a rhetorical one.
Straight answers
The questions people ask next.
How much does Glean cost per user?
Glean does not publish list pricing; its documentation describes a credit-based Enterprise Flex model, and reported deals typically involve per-seat rates with a seat minimum. Any specific figure you read from a third party is unverified — get a written quote, including the year-two number, before you build a business case. Checked 10 August 2026.
Is Glean better than Microsoft 365 Copilot or Gemini in Workspace?
For questions spanning many systems outside a single suite, a dedicated search product has a structural advantage. For questions whose answers live inside the suite you already pay for, the bundled assistant is usually good enough — and free at the margin. The only way to know your split is to score fifty real questions against the bundle first.
What is the fastest way to decide?
Two weeks. Week one: write fifty real questions with known answers and score the bundled assistant, recording correct, confidently-wrong and refused. Week two: archive or merge the stale duplicates behind the failures and re-score. Buy against what still fails, and you will have a business case that survives renewal.
How this was written
Written at the Signal Brief desk against the same hard rules the teardown generator follows: real products only, real vendor links, no invented prices. The bundling claims were verified on 10 August 2026 against the vendors' own material — Google Workspace's AI announcement, Slack's June 2025 pricing and packaging note, Microsoft's 365 Copilot page — and the pricing posture against Glean's own Enterprise Flex documentation. Because Glean publishes no list price, spend avoided is a deliberately wide range rather than a figure.
No sponsorships, no affiliate links and no paid placements anywhere on this site. That is a business-model decision rather than a virtue: a page that names a product and says don’t buy it cannot exist if a vendor is funding it. Corrections to [email protected].
Tomorrow morning, five tools you haven’t heard of yet.
Do-not-adopt calls like this one come out of reading every new AI tool that ships. Signal Brief is that reading, five tools a morning: what each does, why it matters for a PM, and the link. Start with this morning’s issue, free.