Sellable, but only as a KQL-and-BYOC wedge — as a generic fast database the field is already full
Asked (summary):
If a company has built this, will it be hard to sell, and what is the best ICP? The product: a cloud-agnostic analytics engine for AI/observability by the ex-Microsoft Kusto team — KQL plus Postgres SQL on customer object storage with NVMe caching, benchmarked >2× faster than ClickHouse, on the thesis that AI agents fan out parallel queries so per-query speed compounds at scale.
Evidence: 20 market-signal rows (surveys, cost reports, analyst data through 2026), 8 normalized competitor platforms, and 20 company and customer evidence rows across many independent domains. Verdict from the data: roughly 8/10 sales difficulty as a horizontal engine, 5–6/10 with a KQL/BYOC/high-concurrency wedge. Vendor claims are labeled as claims; survey findings are independent unless noted.
Where to sell: attractiveness against difficulty, ten segments
Pursue first — urgency plus a KQL/BYOC wedgeAvoid initially — weak urgency or crowded positioningHover or tap a dot for the reasoning and buyer
Positions are the assessment built from the evidence rows; axes are judgment scales, not survey values.
What the map cannot say alone
The pain is independent and quantified: 74% put cost first in tool selection, observability averages 17% of compute infrastructure spend, and 37% say it costs too much — Grafana's 2025 survey of 1,255 responses. grafana.com
Openness to switching is high but self-interested: an Imply-sponsored report says 87% report slow queries, 87% are exploring alternatives, and 98% would adopt a fully compatible option — vendor-sponsored, so treat as directional. businesswire.com
The AI fan-out thesis has independent support: an AI SRE chasing 10–12 hypotheses generates 10–100× normal query load, while telemetry volume grows 28–40% a year against flat budgets. platformweekly.com
The incumbent is compounding: ClickHouse claims $250M run-rate (tripled year over year), 4,000 customers, and a $400M Series D — a >2× single-query benchmark claim will not outrun that alone. lifestyle.ucconnection.org
BYOC is a proven wedge right now: Tsuga went from stealth to millions in contracted ARR in six months and a $35M Series A on bring-your-own-cloud economics and AI-agent telemetry. beri.net
The crowded part: eight platforms already claim the same architecture
Every cell is read from each vendor's own positioning source, so most claims are vendor claims. Object storage, decoupled compute and a cost claim are table stakes; KQL is the near-empty column — only Hydrolix states KQL support.
StatedPartialNo / not statedHover a row for the vendor's claim in its own words
Sales motion, disqualifiers, and the 90-day validation plan
Sell it this way
Wedge message: keep data in your object store, keep KQL, get materially faster investigations and agent query fan-out at lower total cost.
Lead with one workload — incident investigation, security hunting, long-retention logs, or AI-agent telemetry — and an auditable before/after proof.
Enter as an accelerator or coexistence tier behind Grafana, Splunk or ADX APIs; expand toward system-of-record later.
Founder pedigree opens the door; switching safety, operational simplicity and economics close the deal.
Economic buyer: VP engineering, head of platform/SRE, or CISO for security workloads. Champion: observability platform lead or staff data-infrastructure engineer.
Disqualify early
Teams below roughly 100 GB/day — spend too small to justify migration.
Greenfield teams happy with managed ClickHouse — no urgency.
Generic BI and data warehousing — no KQL wedge, entrenched rivals.
Buyers who want a full out-of-the-box observability UI or APM suite — the engine alone loses to suites.
90-day validation
Weeks 1–4: recruit 3–5 design partners matching the top-left segments (1–10+ TB/day, KQL or mixed KQL+SQL skills, retention or residency pain).
Weeks 3–8: run 2–4 week proofs of value on the customer's real data and concurrency — measure p50/p95/p99 latency, cold vs warm, concurrency, cloud compute cost, ingestion, and migration effort.
Weeks 6–12: publish one auditable customer before/after under concurrent real-world load. That, not another single-query benchmark, is the missing proof.
All 48 evidence rows
Type
Who
Key point
Source
Method: 48 evidence rows across three result sets — 20 market signals (surveys, analyst and cost reports, published 2025–2026), 8 normalized competitor platform profiles, and 20 company/customer evidence rows — each linked to its source domain. The lead chart plots an assessed attractiveness (1–10) against sales difficulty (1–10) per segment, derived from the brief's commercial assessment and the rows; the matrix reads capabilities from each vendor's own stated positioning, so those are claims. Long claim texts are truncated for space; full context sits at each link.