What we're solving
Most analytics tools were designed for one channel and one question: "are we ranking?" They produce three things — a chart, a delta, an alert — and they produce them in three separate dashboards nobody opens in the morning.
Meanwhile, the consequential question changed. A retail brand director now needs to know not just whether their category page ranks on Google, but whether ChatGPT recommends them when a customer asks for the best product under €500. Whether Claude's training data still has 2023 prices. Whether Perplexity cited a Reddit thread that misrepresents their return policy. Whether Gemini's shopping graph has the right SKUs.
Combot was built to answer those questions in the same place you answer the old ones.
What we believe
AI visibility is layered. A brand can be known but not retrieved, retrieved but not cited, cited but not recommended, recommended but with wrong facts. Each layer fails differently and each requires different work to fix. Pretending it's a single ranking position misses everything that matters.
Data lakes beat dashboards. The interesting questions are compound: "Did the new product page launch correlate with the spend spike on Ads and the ranking drop on the parent category?" You can't answer that with one chart. You answer it with SQL against everything at once. So we put everything in BigQuery and let Claude reason across it.
Chat is the operating system. The dashboards that get opened are the ones that come to you. We don't ask teams to learn another UI. We ask Combot in the channel where the work already happens — Slack, Teams, Mattermost, or any platform connected via API integration.
Multi-tenant where it counts; isolated where it must be. Every client gets their own GCP project, their own datasets, their own service account, their own audit trail. Shared infrastructure is fine. Shared data is not.
How we work
Combot runs continuously. Pipelines hit GA4, Search Console, Ads, Merchant Center, Cloudflare, HubSpot, WooCommerce, and others on a 15-min / hourly / daily cadence. Anomaly detection compares each metric against multiple baselines (week-over-week, year-over-year, learned seasonality). Correlations across channels surface compounds humans miss.
The Claude tool-use loop is the front door. It picks the right tool — sometimes a BigQuery query, sometimes a live API call, sometimes a compound audit — and assembles the answer with sources. When you ask "what changed last week," it doesn't return a chart of one metric; it returns a paragraph that names the cause.
On top of all that, we now ship a dedicated AI visibility module: nightly prompt runs across Claude, GPT, Gemini, and Perplexity; citation extraction; fact accuracy scoring; competitor benchmarking; recommendations ranked by commercial impact.
Who's behind this
Combot is built by the Accolade team, a small group of engineers and SEO operators who have been running multi-channel analytics for venture-backed retail, B2B SaaS, and Nordic telecoms for the better part of a decade. We use Combot to run our own client work — it's not a speculative product, it's the tool we needed.
Where we are
Currently in private beta with selected client teams. If you want to evaluate Combot for your organisation, write to [email protected] with a short note on your stack and the questions you'd want a competent analyst to be able to answer at 9 a.m. without you asking.