Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, information, and safety leaders. Subscribe Now
The promise sounds virtually too good to be true: drop a messy comma separated values (CSV) file into an AI agent, wait two minutes, and get again a cultured, interactive chart prepared on your subsequent board presentation.
However that’s precisely what Chinese language startup Manus.im is delivering with its newest information visualization function, launched this month.
Sadly, my preliminary hands-on testing with corrupted datasets reveals a basic enterprise downside: spectacular capabilities paired with inadequate transparency about information transformations. Whereas Manus handles messy information higher than ChatGPT, neither device is but prepared for boardroom-ready slides.
The spreadsheet downside plaguing enterprise analytics
Rossums’ survey of 470 finance leaders discovered 58% nonetheless rely totally on Excel for month-to-month KPIs, regardless of proudly owning BI licenses. One other TechRadar examine estimates that general spreadsheet dependence impacts roughly 90% of organizations — making a “last-mile information downside” between ruled warehouses and hasty CSV exports that land in analysts’ inboxes hours earlier than vital conferences.
The AI Impression Sequence Returns to San Francisco – August 5
The following part of AI is right here – are you prepared? Be part of leaders from Block, GSK, and SAP for an unique have a look at how autonomous brokers are reshaping enterprise workflows – from real-time decision-making to end-to-end automation.
Safe your spot now – house is restricted: https://bit.ly/3GuuPLF
Manus targets this precise hole. Add your CSV, describe what you need in pure language, and the agent routinely cleans the information, selects the suitable Vega-Lite grammar and returns a PNG chart prepared for export—no pivot tables required.
The place Manus beats ChatGPT: 4x slower however extra correct with messy information
I examined each Manus and ChatGPT’s Superior Knowledge Evaluation utilizing three datasets (113k-row ecommerce orders, 200k-row advertising and marketing funnel 10k-row SaaS MRR), first clear, then corrupted with 5% error injection together with nulls, mixed-format dates and duplicates.
For instance, testing the identical immediate — "Present me a month-by-month income development for the previous 12 months and spotlight any uncommon spikes or dips" — throughout clear and corrupted 113k-row e-commerce information revealed some stark variations.| Instrument | Knowledge High quality | Time | Cleans Nulls | Parses Dates | Handles Duplicates | Feedback |
| Manus | Clear | 1:46 | N/A | ✓ | N/A | Right development, customary presentation, however incorrect numbers |
| Manus | Messy | 3:53 | ✓ | ✓ | ✗ | Right development regardless of inaccurate information |
| ChatGPT | Clear | 0:57 | N/A | ✓ | N/A | Quick, however incorrect visualisation |
| ChatGPT | Messy | 0:59 | ✗ | ✗ | ✗ | Incorrect development from unclean information |
For context: DeepSeek may solely deal with 1% of the file measurement, whereas Claude and Grok took over 5 minutes every however produced interactive charts with out PNG export choices.
Outputs:
Determine 1-2: Chart outputs from the identical income development immediate on messy e-commerce information. Manus (backside) produces a coherent development regardless of information corruption, whereas ChatGPT (prime) reveals distorted patterns from unclean date formatting.
Manus behaves like a cautious junior analyst — routinely tidying information earlier than charting, efficiently parsing date inconsistencies and dealing with nulls with out express directions. Once I requested the identical income development evaluation on corrupted information, Manus took almost 4 minutes however produced a coherent visualization regardless of the information high quality points.
ChatGPT operates like a velocity coder — prioritizing quick output over information hygiene. The identical request took simply 59 seconds however produced deceptive visualizations as a result of it didn’t routinely clear formatting inconsistencies.
Nevertheless, each instruments failed by way of “government readiness.” Neither produced board-ready axis scaling or readable labels with out follow-up prompts. Knowledge labels have been often overlapping or too small, bar charts lacked correct gridlines and quantity formatting was inconsistent.
The transparency disaster enterprises can’t ignore
Right here’s the place Manus turns into problematic for enterprise adoption: the agent by no means surfaces cleansing steps it applies. An auditor reviewing the ultimate chart has no approach to affirm whether or not outliers have been dropped, imputed or reworked.
When a CFO presents quarterly outcomes based mostly on a Manus-generated chart, what occurs when somebody asks, “How did you deal with the duplicate transactions from the Q2 system integration?” The reply is silence.
ChatGPT, Claude and Grok all present their Python code, although transparency by code evaluate isn’t scalable for enterprise customers missing programming expertise. What enterprises want is an easier audit path, which builds belief.
Warehouse-native AI is racing forward
Whereas Manus focuses on CSV uploads, main platforms are constructing chart era immediately into enterprise information infrastructure:
Google’s Gemini in BigQuery grew to become usually out there in August 2024, enabling the era of SQL queries and inline visualizations on dwell tables whereas respecting row-level safety.
Microsoft’s Copilot in Cloth reached GA within the Energy BI expertise in Could 2024, creating visuals inside Cloth notebooks whereas working immediately with Lakehouse datasets.
GoodData’s AI Assistant, launched in June 2025, operates inside buyer environments and respects current semantic fashions, permitting customers to ask questions in plain language whereas receiving solutions that align with predefined metrics and enterprise phrases.
These warehouse-native options remove CSV exports solely, protect full information lineage and leverage current safety fashions — benefits file-upload instruments like Manus battle to match.
Crucial gaps for enterprise adoption
My testing revealed a number of blockers:
Reside information connectivity stays absent — Manus helps file uploads solely, with no Snowflake, BigQuery or S3 connectors. Manus.im says connectors are “on the roadmap” however affords no timeline.
Audit path transparency is totally lacking. Enterprise information groups want transformation logs exhibiting precisely how AI cleaned their information and whether or not its interpretation of the fields are right.
Export flexibility is restricted to PNG outputs. Whereas satisfactory for fast slide decks, enterprises want customizable, interactive export choices.
The decision: spectacular tech, untimely for enterprise use instances
For SMB executives drowning in ad-hoc CSV evaluation, Manus’s drag-and-drop visualisation appears to be doing the job.
The autonomous information cleansing handles real-world messiness that will in any other case require handbook preprocessing, reducing turnaround from hours to minutes when you’ve fairly full information.
Moreover, it affords a big runtime benefit over Excel or Google Sheets, which require handbook pivots and incur substantial load instances as a result of native compute energy limitations.
However regulated enterprises with ruled information lakes ought to watch for warehouse-native brokers like Gemini or Cloth Copilot, which preserve information inside safety perimeters and preserve full lineage monitoring.
Backside line: Manus proves one-prompt charting works and handles messy information impressively. However for enterprises, the query isn’t whether or not the charts look good — it’s whether or not you possibly can stake your profession on information transformations you possibly can’t audit or confirm. Till AI brokers can plug immediately into ruled tables with rigorous audit trails, Excel will proceed to carry its starring position in quarterly shows.
