If you are reading a conversational AI magic quadrant to choose a tool for a 60-person business, you are using a map drawn for someone else. Analyst quadrants rank vendors for large enterprises with contact centres, developer teams and multi-year procurement. This article explains what a quadrant can and cannot tell a company of 10 to 250 people, which criteria matter at your size, how to build a simpler grid of your own and how to test the winner on the levelpartner.ai platform in two weeks. The same reasoning applies if the chart in front of you is a chatbot magic quadrant.
What a conversational AI magic quadrant measures
A quadrant plots vendors on two axes, broadly how well they deliver today and how complete their vision is. Both are judged against enterprise buying criteria: global support, security certifications, the size of the partner network, the roadmap. The research is serious and useful for understanding where a market is heading. We do not summarise any particular edition here, because positions change and the reports are licensed.
What matters for you is the buyer the analysts have in mind. They assume in-house engineering, a conversation design team and a contact centre. A vendor can sit in the top right because it handles very large volumes across dozens of languages. You may need neither.
Chatbot charts and conversational AI charts
The chatbot magic quadrant and the conversational AI one overlap, but they cover different groups of vendors and buyers. Chatbot charts tend to focus on text assistants for websites and messaging. Conversational AI charts usually take in voice, contact-centre routing and developer platforms as well. For a firm your size the distinction matters less than it looks, because the questions you need answered are the same in both cases.
Why the top right may be the wrong corner for you
Leaders in an enterprise quadrant tend to be platforms you build on, not tools you switch on. They need integration work to reach your CRM and calendar, and they are contracted for large volumes. Their strength is breadth, and breadth costs set-up time.
A company with flat revenue and no spare headcount needs something different: an agent that answers enquiries, books meetings and updates the contact record next week, not after a six-month project.
The criteria that matter at your size
Score any option against these, in this order:
- Answer quality on your own questions, grounded in your own content.
- Actions: booking, contact updates, tagging and hand-off to a person.
- Where conversations land: one inbox with your email, SMS and calls, or a separate tool.
- Channels: website, WhatsApp, SMS and social messages from the same agent.
- Weekly running time once it is live.
- Total cost, including usage and the hours to set it up.
A tool in the top corner of an enterprise chart can still fail the third criterion and leave you with one more inbox to watch.
Before you start building your own grid
- Your 25 most common pre-sales and support questions, with the answer your best salesperson would give.
- The systems those conversations must touch: CRM, calendars, payments, phone.
- Who will run the evaluation for two weeks, and how many hours a week they have.
- A scoring sheet with the six criteria above and written definitions of right, partly right and wrong.
Then draw your own grid. On one axis, put how much of your required job a tool does without integration work. On the other, put how much weekly effort it takes to run. Place each shortlisted option using demos with your own questions and written answers, not marketing pages, and pick the option that does most of the job for the least effort.
How to set it up: a two-week test on the platform
If the platform is the option your grid favours, train a test agent on your real material and question it before any visitor sees it:
- Go to AI Agents > Conversation AI and open Agent List.
- Select an existing bot or create a new one.
- Open the Bot Training tab.
- Choose an existing Knowledge Base from the selector, if you have one.
- Select a crawling mode: Exact URL, All URLs with this Path or All URLs in this Domain.
- Enter the full URL, including https://, and click Get Data.
- Review the pages found and select only the relevant ones.
- Click Train Bot and wait until each URL shows Trained.
- Click + Add Q & A, enter a question and a complete answer.
- Save the Q&A and test variations of the question.
- Enter a question in Test your Bot and check the reply.
Ask all 25 questions in the test panel and score each reply. Fix the knowledge for every wrong answer and ask again. In the second week, put the agent on a low-traffic page in Suggestive mode so staff approve each reply, read every real transcript and score it on the same sheet. Record the hours spent on set-up and review, because weekly effort is one of your two axes.
Suppose the brokerage scores a test agent on 25 questions and the first pass gets 14 right, 6 partly right and 5 wrong. After two hours of knowledge fixes, a second pass might get 22 right and 3 partly right. If the live week then brought 31 real conversations, 9 booked adviser calls and about 40 minutes of review, the brokerage would hold its own evidence to set against any analyst chart.
Imagine the consultancy's shortlist plotted on an enterprise quadrant: a large contact-centre platform sits top right and the platform sits in a niche position. Re-score the same shortlist on the consultancy's own grid, by job coverage and weekly effort, and the order could flip. The contact-centre platform would need an integration project and a part-time administrator, while the platform agent would cover chat, booking and CRM updates with about an hour of weekly review.
What most articles about the magic quadrant get wrong
They treat the leaders as a shortlist. A quadrant is a ranking against a fixed set of criteria, not a recommendation. If the criteria do not match your job, the ranking does not apply. The useful thing to take from a report is its list of criteria, which you can then weight for yourself.
Reports also exclude tools that fall outside their definition. Products that combine CRM, messaging and AI agents often never appear, because the report is about a different category. Absence is not a verdict.
Comparing vendors on demo conversations is the usual error. Demos are written to show a product at its best. Use your own questions, including the awkward ones about fees, eligibility and complaints, and ask how many of your listed conversations a tool can finish without a developer.
Weekly effort once you choose
Whatever you choose, plan on one to two hours a week for the first month: reading conversations, adding answers and adjusting hand-offs. On the platform that work happens in one place, because conversations, contacts and calendars share a screen.
When an enterprise platform is the right call
If you run a contact centre with dozens of seats, operate in many languages, face strict procurement rules or need deep integration with legacy systems, an enterprise leader may be worth the project, and analyst research is a reasonable starting shortlist. Read enterprise conversational AI and conversational AI for contact centres for that situation. Test on your own content even then.
Troubleshooting your evaluation
Every vendor scores well in demos. Demos use prepared questions. Give each vendor ten real messages from your inbox and watch the answers live.
Scores swing between days. The questions were worded differently. Keep the wording fixed for the scored test and try variations separately.
The agent scores well in testing but poorly with real visitors. Real visitors write shorter, vaguer messages. Add those phrasings as Q&A entries as they appear in transcripts.
Your grid has no clear winner. The axes are too broad. Replace "does the job" with the three conversations that matter most and score each one.
Frequently asked questions
Is the magic quadrant for conversational AI useful for smaller companies?
It is useful for understanding the market and the criteria analysts use. It is weak as a shortlist, because the evaluated vendors are built for enterprise scale.
Why is the platform not ranked as a conversational AI vendor?
It combines CRM, messaging, calendars and AI agents. Analyst reports usually evaluate standalone products, so combined systems often fall outside the definition.
How many tools should I trial?
Two is usually enough: your strongest option and one alternative. More than three spreads the evaluator too thin to read transcripts properly. Two weeks of real conversations per tool shows whether an agent can carry your main job.
What if no tool answers my questions well?
The answers probably do not exist in writing yet. Write them and every tool improves; knowledge-based agents explains why. After choosing, move to your main site with the full website chat bot setup.