Migration
Moving from Intercom to Creobot?
Keep Intercom live while Creobot proves the gap. We benchmark both assistants on your own public pages, using your own questions, and hand you the scorecard with the raw answers and the source each one cited. Switch when the scorecard makes the case.
- Your current bot stays liveNothing is switched off
- Public pages firstNo sensitive data required
- Done-for-you benchmarkWe run it, not you
- Source-grounded scorecardEvery citation opened
Straight up
Where Intercom earns its place
A benchmark is only worth reading if it starts honest. This is what Intercom does well, and none of it is in dispute.
A real inbox behind it
Intercom is a support desk first. If your team lives in that inbox, routing, assignment, SLAs and reporting are already solved. Creobot answers questions and hands off; it is not a helpdesk and does not try to be.
Fin resolves genuinely hard tickets
Fin is a strong product and on mature help centres it resolves a lot. If your documentation is already excellent and the resolution rate is where you want it, Fin is doing its job and the benchmark will say so.
Deep product integrations
Intercom connects to most of your stack. Creobot does not have that surface area and will not for some time.
The gap
Where Intercom leaves answer quality on the table
Not faults. Consequences of what Intercom is built to be — a full customer messaging platform with Fin, its AI agent, layered on top.
Resolution pricing is hard to forecast
Fin bills per resolution. That aligns incentives but makes a monthly figure hard to predict before you run it, which is the complaint most finance teams raise.
You are buying the platform to get the assistant
If all you need is answers on your public site, you are paying for an inbox, seats and workflow you may not use.
Answers come from the help centre you already maintain
Which is fine when it is good, and invisible when it is not. Nothing tells you which questions your content fails to answer.
How it works
What a benchmark actually involves
You send four things
The page your assistant lives on, three to ten questions you already know get asked, your current provider, and an email. No account, nothing to install.
We ask both, by hand
The same questions to Intercom and to Creobot, on your own public pages, in a fresh session each time.
We check the citations
Every answer that cites a source gets that source opened and read. A citation that does not support the claim is recorded as a failure, not a pass.
You get the transcripts
Raw answers from both, side by side, with our scoring and the reasoning for it. You can disagree with the scoring; you cannot disagree with the transcript.
Nothing changes
What stays exactly as it is
Your Intercom inbox, workflows and reporting stay exactly as they are. Creobot answers on the public site and hands to a person; it does not replace the desk.
What we test
Nine things the scorecard measures
The same nine on both assistants, on your pages, in a fresh session each time.
Answer correctness
Does the answer match what the page actually says, judged against the page, not against a plausible-sounding paraphrase.
Source grounding
Whether the answer came from your content or from the model's general knowledge. The second one is where confident wrong answers come from.
Citation accuracy
Every cited source gets opened and read. A citation that does not support the claim is recorded as a failure, not a pass.
Refusal quality
What happens on a question your site does not answer. Silence and invention are both failures; a clean 'that is not on our site' is a pass.
Hallucination risk
Deliberate probes on pricing, certifications and policy -- the three topics where an invented answer costs the most.
Lead signal capture
Whether a buying-intent question turns into something your team can act on, or disappears into a transcript nobody reads.
Handoff payload
What a person receives when the assistant gives up. A transcript with context, or a cold start that makes the visitor repeat themselves.
Missed source pages
Pages that should have been answerable and were not. This is usually the most actionable page in the report.
Widget placement and conversion usefulness
Where the launcher appears, when it interrupts, and whether the answers move anyone closer to a decision.
What we do
You send four things, we do the rest
This is a done-for-you benchmark. The work sits with us.
We map your public sources
Sitemap, docs, help centre and pricing. You do not assemble a corpus for us.
We write the question set
Yours first, then the ones your traffic implies and the ones that break assistants.
We set Creobot up
Configuration, grounding rules, refusal policy and handoff. Nothing for you to install and no account to create.
We run both sides by hand
Same questions, same pages, fresh session each time, one at a time.
We document the gaps
Including ours. A comparison that only ever produces one answer is an advertisement.
We show the migration effort
What moving would actually involve, in hours, before you commit to anything.
We make a recommendation
With the reasoning attached, so you can disagree with the conclusion and still use the evidence.
Scope
Why the first benchmark needs no sensitive data
Creobot is not SOC 2 certified today. That is exactly why the first benchmark stays public-content-only: public website pages, docs, help content and sources you approve. You can measure answer quality, source grounding and handoff without sending customer data anywhere.
Public content only
Marketing pages, docs, help centre and anything else you explicitly approve. Nothing behind a login, nothing from a customer record.
Your stack does not move
Intercom keeps every integration, workflow and report it has today. Creobot answers on the public site and hands to a person.
Security review pack on request
Questionnaire answers, data-flow summary, pilot rules, deletion process and current subprocessor status, before any paid rollout.
Teams that need a report first
If your process requires a SOC 2 report before any pilot at all, Creobot is a later-stage fit and we will say so on the call.
Questions
No. Intercom stays live and untouched for the whole benchmark. Both assistants answer the same questions on your own public pages and you read both transcripts before anything changes.
A scorecard with the raw answers from both assistants, the source each one cited, and whether that source actually contains the claim. Working shown, not a score out of ten.
Send four things: the page your assistant lives on, three to ten questions you already know get asked, your current provider, and an email. We map the sources, write the rest of the question set, configure Creobot, run both sides and document the gaps.
A few working days. We run your questions by hand rather than automating a scan, which is slower and produces a result you can defend in a room.
Not for the benchmark. It runs on public website pages, docs, help content and sources you approve. That is deliberate -- you can measure answer quality and handoff before any security review is involved.
It depends entirely on your volume, and we would rather you ran the numbers than took a claim from either vendor. The cost estimator uses your own traffic.
It is on this page in plain language, above. The benchmark measures answer quality, grounding and handoff -- if you need the rest of Intercom's platform, that is a real reason to keep it and the scorecard will not argue with you.
When the scorecard makes the case on your own pages and your own questions. Not before, and not on either vendor's say-so.
See the gap on your own pages
Your questions, both assistants, every citation opened and checked. Intercom stays live throughout. Free, nothing to install, and you keep the transcripts either way.