# AI Report Builder: Describe It, Get a Dashboard | Mihu AI Product update · Reports · September 2026 # Custom Reports: Describe the Report in One Sentence, Get the Dashboard Type what you want to see. The AI reads your workspace data model, builds each widget, checks its own numbers, and lays them out on a canvas you rearrange by hand. Or open one of eighty sector templates and watch it fill with your data. Either way, nothing a report does can write anything. Mihu AI Team September 25, 2026 9 min read v1.1.0 New · Custom Reports One sentence · Verified widgets · Your canvas You describe It builds & checks You arrange The AI does the part that used to need an analyst. You keep the part that needs you: deciding what matters and where it goes. In this article 1 One sentence in (https://mihu.ai/blog/custom-reports#sentence) 2 Built, run, then checked (https://mihu.ai/blog/custom-reports#verified) 3 Refining with the report assistant (https://mihu.ai/blog/custom-reports#assistant) 4 The canvas: your layout, your filters (https://mihu.ai/blog/custom-reports#canvas) 5 Eighty templates that open with your data (https://mihu.ai/blog/custom-reports#templates) 6 When the data is heavy: Excel by email (https://mihu.ai/blog/custom-reports#heavy) 7 Read-only by construction (https://mihu.ai/blog/custom-reports#safety) 8 Sharing, and reports on your home page (https://mihu.ai/blog/custom-reports#share) 9 Frequently asked questions (https://mihu.ai/blog/custom-reports#faq) The built-in analytics pages answer the questions everybody asks. Custom Reports is for the question only you ask: satisfaction by channel for the two agents you changed last week, missed calls between five and seven on weekdays, which campaigns produced appointments that were actually kept. Until now that question went to somebody with database access and came back in a few days. Now it goes into a text box. ## One Sentence In Describe the report. The AI builds the widgets, checks the results and lays them out. The Reports page opens on a single box headed What do you want to see? Type a sentence — CSAT report for the last 30 days, by channel and by agent — and press Create report. The steps that follow are shown as they happen: Understanding the instruction. The sentence is split into the widgets a person would expect: a headline number, a trend, a breakdown by channel, a breakdown by agent. Reading the data model. The AI is shown the tables a report may use and what their columns mean. Nothing else. Building and testing each widget. Each one is written, run against your data, and repaired if the run fails. Designing the report. Widgets get a chart type that suits their shape — a KPI tile for one number, a line for a trend, a donut for a share — and a place on the canvas. Saving. The report opens with every widget filled. It is already saved; everything you do from here is saved automatically too. Reports · CSAT by channel and agent · last 30 days Satisfied customers 87% Conversations 4,312 Handled by AI 71% Handed to your team 12% Satisfaction trend · per day By channel WhatsApp 91% Calls 85% Web chat 88% Email 79% Filters: Last 30 days All agents Saved automatically One sentence, six widgets, about a minute. The numbers are illustrative; yours come from your workspace. ## Built, Run, Then Checked A number nobody verified is a rumour with a chart. The thing that makes generated reports dangerous is not that they fail. It is that they succeed with the wrong number, confidently, in a nice font. So every widget here goes through a second pass after it runs: the AI reads the result and writes, in plain language, what it shows and whether it answers the instruction. Average satisfaction per channel, weighted by conversation count, last 30 days; e-mail is lower because 40% of its conversations have no rating. That explanation is saved with the widget and shown on it. If the check finds a problem — the wrong grain, a period that does not match, several metrics squeezed into one table — the widget is repaired and re-run before it is added. If it cannot be repaired it is left out and the report says so, rather than showing you something plausible. And every widget can be opened as a table, so the chart is never the only thing between you and the rows. ## Refining With the Report Assistant Tell it what to change. It edits the widgets on this report. The first version is rarely the last. A panel beside the report takes instructions about the report you are looking at: You say What happens Split the CSAT chart by channel The widget is rebuilt with one series per channel, re-verified, and swapped in place Last 7 days instead of 30 Every widget on the report moves to the new period Add average call duration A new widget is built, checked and placed in the first free spot Make the first chart a pie Chart type changes; the data does not Remove the table Gone, and the report is saved Each change is listed as it is applied — Updated “Satisfaction trend” , Added “Average call duration” — so a report edited by three people over a month still reads as a sequence of decisions rather than a mystery. ## The Canvas: Your Layout, Your Filters The AI proposes a layout. You own it. The report is a twelve-column canvas. Drag a widget to move it, pull its corner to resize it, and it stays exactly where you put it — the canvas does not snap everything upward the moment you remove something. Chart type, colours, number format and the mapping of columns to axes are on each widget's controls. Ten chart types are available: bar, line, area, pie, donut, radar, scatter, funnel, table, and the single-number KPI tile. Widgets can carry filters . A date range, an agent, a channel, a queue: define them once and they appear as a filter bar across the top of the report. The AI uses a date range by default when you ask for a period, so last 30 days is a filter you can move, not a number baked into the widget. Filter sets can be saved as presets, and a filtered view has its own link you can send to somebody. ## Eighty Templates That Open With Your Data Pick a report. It fills instantly. No AI wait. Not every report needs describing. The template gallery holds eighty ready-made reports across eight sectors — automotive, hospitality, tourism, healthcare, e-commerce, real estate, financial services, professional services — built from more than a hundred prepared widgets, grouped by what they track: conversations, calls, satisfaction, appointments, tickets, campaigns, contacts, agents, team, usage and cost. Report templates · Automotive Automotive Hospitality Tourism Healthcare E-commerce Real estate Financial Professional services Service desk overview Volume, channels and how much of the service desk the AI agent handles alone. 8 widgets Missed & answered calls Every call to the dealership: answered, missed, when they come in and how they end. 7 widgets Service bookings Workshop slots booked, completed and cancelled, week by week. 6 widgets Test drives Requested, scheduled and completed test drives, by model. Open with my data Every widget in a template is already built. Opening one creates the report and fills it from your workspace in a few seconds. A template is a starting point like any other report: rearrange it, filter it, and ask the assistant to change what it does not get right for you. ## When the Data Is Heavy: Excel by Email A screen is for a few thousand rows. A spreadsheet is for the rest. Before an interactive run, the platform estimates what it is about to do. If a widget would return more rows than fit on a screen, or would have to scan more than is sensible in the time a page can wait, you are told — This widget's data looks heavy: it would return about 240,000 rows — and offered a choice: run a capped preview anyway, or prepare the full result as an Excel file in the background and get an e-mail when it is ready. The e-mail carries a link that requires signing in to your workspace and expires after a set number of days. The Exports tab lists every file: status, rows, when it was requested, how long it is available. The same offer appears on a widget that timed out or was capped, so a heavy report is never simply blank. ## Read-Only by Construction There is no path from a report to a write. Several layers make sure. Only allowed tables. The AI is shown a list of tables a report may use and nothing outside it. Columns holding personal or secret values are hidden from that list and cannot be selected, by the AI or by you. Only reads. Every widget is a single read query. Anything else is refused before it runs. Inside a read-only transaction, with a time limit. Even a read that tried to do something clever would be stopped, and a slow one is cut off rather than allowed to weigh on the workspace. Optionally, a separate read-only database identity. Workspaces that want it can point reports at a database user that is physically unable to write. Widgets are computed by background workers, not by the page you are looking at, and results are cached for a few minutes, so a report opened by ten people at nine in the morning runs its queries once. ## Sharing, and Reports on Your Home Page Yours until you share it. On your home page if you want it there. Reports and report widgets belong to whoever made them until they press Share with your team ; then the team can open them and only the owner can edit. And a report widget does not have to live on a report: it can be placed on one of your Boards (https://mihu.ai/blog/boards) , the customisable home page every team member gets, next to the platform's own tiles. The number you care about most can be the first thing you see when you sign in. Try this: write the sentence you would have sent to an analyst last month. Read the explanation on each widget before you read the chart. If the explanation is right, the chart is right. ## Frequently Asked Questions What data can a report use? Your workspace data: conversations, calls, messages, sessions, appointments, tickets, contacts, campaigns, agents, team members, evaluations and usage. The AI is shown the data model and builds widgets on it. Columns that hold personal or secret values are hidden from reports entirely. Can a report change or delete anything? No. Every widget runs as a read-only query, inside a read-only transaction, against a list of allowed tables, with a time limit. There is no path from a report to a write. If a request is not a read, it is refused before it runs. How do I know the numbers are right? Each widget is built, run, and then checked by a second pass that reads the result and explains what it shows. The explanation is saved with the widget. If the check finds a problem, the widget is repaired before it is added. You can also open any widget’s data as a table. What if I do not want to describe anything? Open a template. There are eighty ready-made reports across eight sectors, built from more than a hundred prepared widgets, and they open with your own data instantly, no AI wait. Change anything afterwards with the assistant. What happens with a very large result? Before an interactive run the platform estimates the size. If a widget would return more rows than fit on a screen or take too long, it offers to prepare the full result as an Excel file in the background and e-mail you a signed-in download link. The Exports tab lists every file and when it expires. Can I filter a report by date or agent? Yes. Any widget can carry named parameters such as a date range or an agent, and they appear as a filter bar on the report. Filter sets can be saved as presets, and a filtered view has its own link. Can I share a report with my team? Yes. Reports and report widgets are yours until you share them; shared items are visible to the team and editable only by the owner. Report widgets can also be placed on a board, your customisable home page. Which chart types are available? Bar, line, area, pie, donut, radar, scatter, funnel, table, and a single-number KPI tile. Change the type from the widget’s controls or by asking the assistant. ## Ask your data a question today Open Reports, type the sentence, and read the report a minute later. Or open a template for your sector and see it fill with your data. Custom Reports is included on every plan. Register free → (https://app.mihu.ai/register) Product update Reports Analytics Dashboards Templates Excel export ## Keep exploring More ways to see what your contact center is doing 🏠 Product update #### Boards: Get Your Custom Boards, Built From the Whole Platform Sep 2026 (https://mihu.ai/blog/boards) 📊 Guide #### Conversation Intelligence Blog (https://mihu.ai/blog/conversation-intelligence) 📈 Product #### Analytics Product page (https://mihu.ai/analytics)