Independent, methodology-published research for enterprise AI decisions. Four formats — market maps, buying guides, benchmarks, and vendor comparisons — each answering one specific step of the buying question. No rankings, no paid inclusion, no vendor-optimized categories.
Why research
AI procurement fails at selection, not deployment. Not because buyers are incompetent — because the categories that would let a buyer decide what to shortlist don't exist in a usable form yet. So buyers default to feature comparison, feature comparison defaults to the best demo, and the best demo defaults to whichever vendor spent the most on producing it.
The existing research products don't close this gap. Analyst quadrants rank markets against each other — useful if you already know which market you're in, which you usually don't. G2-style grids rank vendors against each other — useful if you already know which criteria matter, which is usually the thing you most need help with. Vendor sites describe what the vendor wants to sell. None of them answer the actual question: what category should you be buying in, and how do you evaluate within it?
Four formats, each answering one step. Market maps categorize the market — the categories, buyers, failure modes, and selection criteria per category. Buying guides decide within a category — weighted criteria, shortlist process, reference script, TCO model, contract clauses. Benchmarks show what the market collectively believes — annual structured surveys with published methodology and sample. Vendor comparisons — the format we haven't published yet, and explain below — compare specific products against stated criteria, not against each other.
The discipline behind these is the point. Every format publishes methodology. Every format rules out paid inclusion. Every format logs corrections in public. Those aren't marketing claims; they're the terms of the product. A research product that requires you to trust the researcher is a research product that can't be audited. The public record is what closes that gap.
The formats
Each format answers one question a buyer will ask in order. Read the one that matches the step you're on — not the one that sounds most interesting.
What does the market look like?
Categories, buyers, failure modes, and per-category selection criteria.
1 published · 2 in research Browse market mapsHow do we decide within a category?
Weighted criteria, shortlist process, reference script, TCO model, contract clauses.
1 published · 2 in research Browse buying guidesWhat does the market collectively believe?
Annual structured surveys, with published methodology and sample.
1 published · 1 next up Browse benchmarksHow do specific products compare, against criteria?
Format not yet published. Position stated below — read it before assuming this is live.
Not currently published Our positionRead the market map if you don't yet know which category you're buying in — that's the starting question. Read the benchmark to see what the market collectively believes about the category, and where it might be wrong. Read the buying guide once you know your category — it's the decision framework. Read vendor comparisons when you're ready to compare specific products against criteria, not against each other. The order isn't a sequence to complete; it's a way to find the format your question requires.
Not currently published — and it's a deliberate choice. A vendor comparison names specific products against stated criteria. That's a different editorial posture than a market map (which names vendors as category examples, not as comparisons) or a buying guide (which names no vendors at all). Every claim in a vendor comparison is specific and challengeable, and the disclosure standard it demands is higher than any other format. Publishing one before we've built that standard would be the wrong order.
What we're building first: a disclosure protocol that names every relationship with every named vendor, an evaluation methodology that isolates criteria cleanly, and a corrections process that can update a vendor's status without pulling the whole comparison. When that exists, vendor comparisons publish on the same terms as the other three formats — no paid inclusion, methodology on the page, corrections logged. Until then, this format is a stated position, not a live product. If you were referred here expecting a comparison page and found this instead: the reason is deliberate, and the standard above is what "we're working on it" actually means.
What's live
Three items across three formats. The collection is small on purpose — each item takes a quarter of research, and a report you can't audit is worse than no report.
What to shortlist, what to verify, and what to walk away from — before the demo. Data access and integration fit as eliminative criteria, in that order, before feature comparison starts.
Published · 12 min Read
Customer service AI isn't one market. Five categories — agent assist, self-service, voice, quality, orchestration — with different buyers, deal sizes, and failure modes.
Published · 10 min Read
340 transformation leaders across 12 sectors. Where organizations actually sit in the five-stage model, what changed in real workflows, and the four mistakes that repeated across sectors.
Published · Report Read
In research
Honest state, not marketing state. "In research" means a named analyst is on it with a draft timeline. Items here don't have URLs yet — they're not published, and we don't pretend they are.
How it's made
Categories are derived from observed buying behavior — talking to enterprise buyers who have evaluated or purchased in a category — not from analyst quadrants. Market maps require a minimum of twelve enterprise buyers per category interviewed before publishing. Buying guides use criteria derived from observed failure modes — the specific reasons enterprises walked away from shortlists — not from vendor-supplied comparison grids. Benchmarks require the full sample and methodology published with the report, so the analysis can be recomputed by an independent analyst.
Every format is reviewed quarterly. Quarterly means category validation, not just a re-date on the page: before an update publishes, the source interviews are re-run, and if the categories have moved, the update reflects that.
Questions about the research product, not the individual reports. For report-level questions, each collection page has its own FAQ.
Ranking is a claim about relative quality, and it ages faster than the market. A vendor that's #1 in Q4 is often #3 in Q1, and the ranking that named them is now wrong — through no fault of the reader's, and often not through any fault of the vendor's either. The three live research formats answer different questions: market maps categorize, buying guides evaluate criteria, benchmarks show what the market collectively believes. None requires a ranking, and each is more durable without one. Vendor comparisons, when we publish them, will compare specific products against stated criteria rather than rank them against each other — the difference is that a comparison is auditable against its own methodology, while a ranking requires you to trust the researcher's judgment.
Three differences. First, methodology is published — sample sizes, review standards, and criteria weighting are in the report, not behind a paywall, so the analysis can be audited. Second, categories are workflow-grounded — derived from observed enterprise buying behavior, not analyst abstractions that may not match your organization's actual workflows. Third, no vendor can pay for inclusion, prioritization, or editorial input — the terms are the terms. G2 is genuinely better at peer-review breadth; if that's what you need, use it. This research is a category-and-decision framework: the thing that helps you decide what to buy in, not just who to buy from.
When we can publish them on the same terms as the other formats: no paid inclusion, methodology on the page, comparisons against stated criteria rather than vendor-vs-vendor rankings. Named vendors in a comparison are a different editorial risk than named vendors in a market map — every claim is specific and challengeable. We're building the standard first; the format publishes after. The full position is stated above.
Through the same channels as everything else on the hub: newsletter subscriptions and our own tools and playbooks. No vendor pays for inclusion, prioritization, or coverage. Any commercial relationship that affects content is disclosed on the page it affects. The disclosures page names every active relationship — currently none, so you can always check what has changed.
Quarterly review per format. Benchmarks update annually — the 2027 baseline is scheduled for Q4 2027. When a report is materially out of date — because the market has moved, a key vendor has changed, or the criteria no longer hold — the report is re-verified, re-dated, and the previous version is logged in the corrections page. A report whose masthead says Q4 2026 and hasn't been updated by Q4 2027 should be viewed with caution — that's a signal, not a mistake.
Yes — send a note naming the market, the category, or the specific question. We prioritize by how many readers ask the same question, how much buying behavior we can observe in the market, and whether the format (map, guide, benchmark, comparison) is the right one for the question. Categories that are still settling — agent orchestration being the obvious example — will be waited out until they've stabilized enough to research honestly.
The operational side
Once a research decision becomes a workflow change, the playbooks cover implementation — mapping the workflow, setting the baseline, measuring ROI, governing autonomy. Research and playbooks are the two halves of a transformation decision: what to buy, and what to do once you have it.