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 · 10 August 2026· free to read, no signup
Should we buy ThoughtSpot Spotter? Not before the semantic model exists
A conversational analytics quote lands, the demo answered every question perfectly, and the number is per-seat and annual. This is the call our teardown generator made on that exact purchase for a 340-person B2B SaaS org — and the three-week test that settles it either way.
~7 min read · reviewed 2026-08-10· no sponsorships, no affiliate links
The call, up front
Do not adopt yet. You are being quoted per seat for the interface, while the semantic model that decides whether any answer is correct stays your job.
Why not. Both products sell a natural-language interface on top of a curated semantic model, and the semantic model is the part you have to build yourself. On a Snowflake + dbt stack you would be paying per seat, annually, for the layer that sits in front of work nobody has done yet — and it is that work, not the interface, that decides whether a question about churn returns the number your finance lead recognises. ThoughtSpot publishes $25/user/month (Essentials) and $50/user/month (Pro, where Spotter carries a per-user monthly query allowance), billed annually, with enterprise quoted rather than published. Across a 100–300 seat go-to-market population that is a five- to six-figure annual commitment signed before the first correct answer.
Instead. Spend three weeks authoring a semantic model for the twenty questions your team re-asks every month, then put Snowflake Cortex Analyst in front of it — billed per message against credits you already buy, with no new vendor, no procurement cycle and no per-seat licence. Score it on those twenty questions with no analyst in the loop. Clear fifteen and you have your answer engine; clear five and a BI vendor was never going to fix it either, because the problem is upstream of the interface.
Spend avoided: ~$60k–$180k/yr (100–300 seats at ThoughtSpot's published $25–$50/user/month, before enterprise uplift)
Written against this stack: Snowflake + dbt warehouse, Jira, Amplitude, Zendesk; already paying for GitHub Copilot and ChatGPT Team; ~340 people, no dedicated analytics engineer.
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 warehouse has no native natural-language query service. Cortex Analyst is a Snowflake feature; on a platform without an equivalent, a BI vendor really is the only route to conversational access and this page collapses into a pricing negotiation.
You already maintain a real semantic layer — dbt semantic models, a metrics store, a governed Looker or Omni model — and your problem is reaching people who will never open a Slack app or a notebook. Then you are buying distribution, which is a legitimate purchase with a countable before and after.
You can state this week's ad-hoc analytics queue in tickets and days. A tool bought against a measured queue is a decision; a tool bought against “the team is swamped” is a hope.
01— Detail
What you are actually being sold
Every product in this category — Spotter, Zenlytic, the warehouse vendors' own text-to-SQL services — resolves to the same three-part machine: a semantic layer that says what a metric means, a model that turns a sentence into SQL against that layer, and a surface the business types into. The demo makes the middle part look like the product. It is not. The middle part is table stakes now and improves monthly across every vendor, including the ones bundled into platforms you already pay for.
The expensive part is the semantic layer, and it is expensive in engineering attention rather than licence fees. Somebody has to decide whether “active customer” excludes trials, whether revenue is booked or recognised, which of your four date columns means “signed”, and what happens to accounts that churned and came back. None of that arrives with a purchase order. It arrives when an analytics engineer sits with a finance lead for a fortnight, and it is the single determinant of whether natural-language answers are trustworthy or merely fast.
So sequencing matters more than the shortlist. Buy the interface first and you spend the first two quarters of a per-seat contract doing the semantic work anyway, while a business audience learns that the shiny new tool disagrees with the board deck. Do the semantic work first and the interface becomes a genuinely competitive purchase — including against the option already inside your warehouse bill.
02— Detail
Where the pilot goes wrong
Two failure modes recur, and neither shows up in a vendor-run demo, because a vendor-run demo uses a model a solutions engineer curated for your data the week before.
The first is silent wrongness. These tools rarely say “I don't know”; they return a plausible number computed off the wrong join or an unfiltered test account. A business user cannot tell, and the first time the number contradicts the weekly report, trust in the tool evaporates and takes some trust in the warehouse with it. The mitigation is a scored question set — fixed questions with known-correct answers, re-run after every model change — and that instrument is worth building whichever vendor you pick.
The second is that per-seat pricing bills population, not usage. Conversational analytics has a long tail: a handful of people ask something weekly, most of the org asks twice a quarter. Seats charge you identically for both, so pilot-cohort enthusiasm gets extrapolated across people who will never log in twice. That is the largest source of overspend in this category, and the reason message-metered pricing is worth arguing for in any negotiation you do run.
Write twenty questions with known-correct answers before you see any vendor. That list is your benchmark, your acceptance test and your negotiating position.
Insist any trial runs against YOUR semantic model, not one built for the demo.
Price the renewal, not the pilot: assume the seat count doubles, and get the year-two number in writing.
03— Detail
The three-week test that settles it
Week one: pick the twenty recurring questions and write down the correct answer to each, agreed with whoever owns that number today. Week two: author a semantic model covering only the entities those questions touch — not your whole warehouse. Week three: point Cortex Analyst at it and score all twenty with no analyst in the loop.
The score is the decision. Fifteen or more and you have a working answer engine for the price of messages against credits you already buy; put the leftover budget into the next twenty questions. Below ten and the constraint is your data model, which no vendor on your shortlist fixes — buying one relocates the disappointment into a contract with an auto-renewal clause.
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.
Snowflake's managed text-to-SQL service, driven by a semantic model you author and callable from an app, a Slack bot or the standalone Analyst API.
Why it covers the job. It answers the same class of question as a conversational BI product, against the warehouse you already run — so the semantic work becomes the whole project instead of an unbudgeted prerequisite.
Cost: Message-based billing against your existing Snowflake credits; the generated SQL costs warehouse compute as usual. No per-seat licence, no new vendor. Check Snowflake's AI pricing page for the current rate. · docs.snowflake.com
Metric and entity definitions living in the dbt project you already maintain, version-controlled alongside the transformations.
Why it covers the job. This is the asset the whole category depends on and the one thing you cannot buy. Built in dbt it stays portable, outliving any BI vendor you later pick or drop.
Cost: Included in the dbt tooling you already run; the real cost is analytics-engineering time — roughly a fortnight for a first useful slice. · docs.getdbt.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 evidence that Spotter scores materially better than warehouse-native text-to-SQL when both are pointed at the SAME semantic model. That is the only comparison that isolates what you would be paying for, and nobody publishes it.
Consumption pricing from ThoughtSpot for Spotter instead of seats plus a query allowance. The category's long-tail usage is a bad fit for population billing, and metered pricing would change this arithmetic immediately.
You are not on Snowflake, BigQuery or Databricks — then the cheaper path does not exist, and the verdict becomes a straight negotiation about seats.
Straight answers
The questions people ask next.
Is ThoughtSpot Spotter worth the money?
Not before you have a semantic model, and if you are on Snowflake, not before you have scored Cortex Analyst against twenty known-answer questions. Spotter is a good interface to a curated model; it does not create the model, and the model is the expensive part. Once one exists, Spotter becomes a distribution decision worth pricing on its merits.
Can Snowflake Cortex Analyst replace ThoughtSpot?
For the natural-language question-answering job on a Snowflake stack, usually yes — it is billed per message against credits you already buy and needs a semantic model you had to author anyway. It does not replace a full BI platform's dashboarding, embedding and governance surfaces, so compare it against the specific job you were buying Spotter for.
How much does ThoughtSpot Spotter cost?
ThoughtSpot publishes $25/user/month for Essentials and $50/user/month for Pro, both billed annually, with Spotter access on the paid tiers carrying a per-user monthly query allowance; enterprise pricing is quoted rather than published. Checked 10 August 2026 — verify on thoughtspot.com/pricing before you budget.
What about Zenlytic instead?
Same verdict for the same reason: Zenlytic is a conversational BI product that depends on a curated semantic layer. Choosing between it and Spotter is the second decision. The first is whether your semantic model can support correct answers at all, and that question is free to test.
How this was written
This call-out is verbatim output from the generator that writes the paid $199 AI Tooling Teardown — the same buildCalloutPrompt() → platform agent with web_search → parseCalloutResult() path a free call-out request runs — from a probe run on 10 August 2026 against the stack named above. Nothing in it was rewritten to read better. Every published price it quotes was then re-checked by hand against the vendor's own pricing page on 10 August 2026; spend avoided is a range because ThoughtSpot's enterprise tier is quoted, not published.
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.