GPT for MS Excel — Twistly

🔍 Tool Overview & Architecture

GPT for MS Excel by Twistly introduces a suite of native spreadsheet functions prefixed with AI. directly into the Excel formula engine. These functions execute asynchronous API calls to LLM endpoints, returning scalar text, categorized choices, or dynamically spilled arrays directly into worksheet cells.

flowchart LR
    Cell["Cell Formula<br/>=AI.EXTRACT(A2, 'Phone')"] --> AddinEngine["Twistly Office.js Add-in Engine"]
    AddinEngine --> API["OpenAI / Twistly Cloud API Endpoint"]
    API --> AddinEngine
    AddinEngine --> Result["Spilled / Scalar Result in Grid"]

    style Cell fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff
    style AddinEngine fill:#1e293b,stroke:#10b981,stroke-width:2px,color:#fff
    style API fill:#1e293b,stroke:#f59e0b,stroke-width:2px,color:#fff
    style Result fill:#1e293b,stroke:#06b6d4,stroke-width:2px,color:#fff

Supported Platforms & Requirements

• Excel Platforms: Excel for Microsoft 365 (Windows & Mac), Excel on the Web, Excel 2016/2019 (with Office.js requirement set 1.1+).
• Authentication: Free trial tier with starter credits; Pro/Team subscription via Twistly or Bring-Your-Own-Key (BYOK) OpenAI API connection for enterprise privacy and unlimited volume.
• Network: Constant internet connectivity required. Not functional in offline or strictly air-gapped environments.

📚 Core AI Function Reference

Below is the verified reference manual for Twistly custom functions:

1. AI.ASK

Purpose

Prompts the language model with an open-ended natural language question, returning a text or numeric response directly into the cell.

Syntax

=AI.ASK(prompt, [value])

Inputs

• prompt (Required, String): The instruction or question to send to the model.
• [value] (Optional, Reference/String): Optional cell reference providing context data for the prompt.

Outputs

• (String): The scalar text completion generated by the AI model.

Example

=AI.ASK("Summarize the customer complaint in under 10 words:", B2)

Data Analyst Use Case

Generating rapid executive bullet summaries of customer feedback logs or call transcript notes directly beside raw survey rows.

Common Mistakes

• Omitting quotation marks around the string prompt.
• Expecting numerical calculation results that require exact mathematical precision (AI may hallucinate arithmetic; use native Excel SUM/AVERAGE instead).

Limitations

• Re-evaluates upon workbook recalculation if automatic calculation is enabled, consuming API tokens.

Verification Steps

1. Verify the summary against the original row text.
2. Confirm the word limit was strictly respected.

2. AI.TABLE

Purpose

Generates a dynamic 2D table complete with headers and rows based on a descriptive prompt.

Syntax

=AI.TABLE(prompt, [headers])

Inputs

• prompt (Required, String): Description of the dataset or template to generate.
• [headers] (Optional, Array/String): Explicit column headers to constrain table structure.

Outputs

• (Dynamic Array): A multi-row, multi-column spilled range.

Example

=AI.TABLE("Generate 5 synthetic retail transactions with Date, Product, Region, and SalesAmount")

Data Analyst Use Case

Synthesizing realistic dummy test data to stress-test complex formula logic, pivot tables, and dashboard layouts before production data arrives.

Common Mistakes

• Placing AI.TABLE in a cell where adjacent cells are occupied, triggering an Excel #SPILL! error.

Limitations

• Synthetic data cannot replace real statistical distributions for econometric modeling.

Verification Steps

1. Ensure the destination range has adequate empty space to prevent #SPILL!.
2. Inspect data types across columns to verify dates and numbers are properly parsed.

3. AI.TRANSLATE

Purpose

Translates cell text from a source language to a specified target language.

Syntax

=AI.TRANSLATE(text, target_language, [source_language])

Inputs

• text (Required, Reference/String): Cell containing the text to translate.
• target_language (Required, String): Desired output language (e.g., "English", "French", "Arabic").
• [source_language] (Optional, String): Source language if known, otherwise auto-detected.

Outputs

• (String): Translated text.

Example

=AI.TRANSLATE(C2, "English")

Data Analyst Use Case

Standardizing international customer support survey feedback or multi-regional product descriptions into a common corporate language.

Common Mistakes

• Passing empty cells without wrapping in an IF(C2="","", ...) check, incurring wasted API calls.

Limitations

• Nuances, idiomatic industry jargon, and regional colloquialisms may be mistranslated.

Verification Steps

1. Spot-check 10% of translated records with native speakers or reference dictionaries.

4. AI.FORMAT

Purpose

Reformats unstructured, messy, or inconsistently formatted cell text into a standardized target format.

Syntax

=AI.FORMAT(text, target_format)

Inputs

• text (Required, Reference/String): The unformatted cell value.
• target_format (Required, String): Description or mask of the desired pattern (e.g., "(###) ###-####", "YYYY-MM-DD", "Proper Case").

Outputs

• (String): Formatted text adhering to the target pattern.

Example

=AI.FORMAT(D2, "Standard International Phone: +1-XXX-XXX-XXXX")

Data Analyst Use Case

Standardizing phone numbers, postal codes, and employee ID codes imported from disparate legacy ERP/CRM dumps.

Common Mistakes

• Assuming the output is a native Excel numeric serial date; it is often returned as text.

Limitations

• Inconsistent source data with missing digits may be hallucinated or truncated unpredictably.

Verification Steps

1. Test with LEN() to verify uniform character counts across cleaned outputs.
2. Confirm no arbitrary filler digits were introduced.

5. AI.EXTRACT

Purpose

Extracts specific named entities, substrings, or attributes from freeform unstructured text.

Syntax

=AI.EXTRACT(text, entity_to_extract)

Inputs

• text (Required, Reference/String): Source cell containing unstructured narrative or strings.
• entity_to_extract (Required, String): The specific attribute to extract (e.g., "Email Address", "Invoice Number", "Order Date").

Outputs

• (String): The isolated attribute value.

Example

=AI.EXTRACT(A2, "Invoice Number")

Data Analyst Use Case

Parsing unstructured bank statement transaction descriptions or automated billing emails to pull invoice IDs and tax numbers into structured columns.

Common Mistakes

• Requesting multiple disparate entities in a single call without specifying delimiter formatting.

Limitations

• If the entity is absent, the function may return "N/A" or blank depending on the prompt; handle with IFERROR.

Verification Steps

1. Filter results for blank or error outputs.
2. Verify extracted strings match source substring coordinates.

6. AI.FILL

Purpose

Infers a pattern from provided few-shot examples and auto-completes target rows based on context.

Syntax

=AI.FILL(example_range, target_range)

Inputs

• example_range (Required, Range): Input/output pair cells demonstrating the pattern.
• target_range (Required, Range): Input cells to evaluate against the pattern.

Outputs

• (Dynamic Array / String): Completed values adhering to the inferred logic.

Example

=AI.FILL(A2:B4, A5:A20)

Data Analyst Use Case

Handling complex string splitting or parsing where standard Excel Flash Fill (Ctrl+E) fails due to irregular delimiters or contextual semantic variations.

Common Mistakes

• Providing ambiguous or contradictory example rows that confuse pattern inference.

Limitations

• Slower than native Excel Flash Fill; should only be used when Flash Fill pattern detection fails.

Verification Steps

1. Review the first 5 generated outputs against the input logic.
2. Validate edge cases with non-standard lengths.

7. AI.CHOICE

Purpose

Classifies input text into one of a strictly defined list of categories.

Syntax

=AI.CHOICE(text, valid_choices_range_or_array)

Inputs

• text (Required, Reference/String): The text to classify.
• valid_choices_range_or_array (Required, Range/Array): Allowed category labels (e.g., {"Positive", "Neutral", "Negative"} or reference to $F$2:$F$4).

Outputs

• (String): Exactly one value selected from the provided valid choices.

Example

=AI.CHOICE(E2, $K$2:$K$5)

Data Analyst Use Case

Automating sentiment categorization, ticket prioritization ("Urgent", "High", "Medium", "Low"), or customer department routing across inbound support tickets.

Common Mistakes

• Not locking category references ($K$2:$K$5), causing lookup range drift when dragging down.

Limitations

• Multi-intent text (e.g., a review with both positive and negative comments) will be forced into a single category.

Verification Steps

1. Build a quick Pivot Table of AI.CHOICE outputs to confirm 100% of rows map strictly to valid categories.
2. Manually audit border-line cases.

8. AI.LIST

Purpose

Generates a one-dimensional array (vertical list) of items matching a specific prompt.

Syntax

=AI.LIST(prompt)

Inputs

• prompt (Required, String): Description of items to enumerate.

Outputs

• (Dynamic Array): Vertical column of spilled items.

Example

=AI.LIST("List the 10 most common customer churn reasons in retail banking")

Data Analyst Use Case

Brainstorming dimension categories, data validation picklist values, or survey answer options during analytical planning phases.

Common Mistakes

• Failing to specify quantity limits in the prompt, leading to inconsistent list lengths.

Limitations

• Cannot be directly edited without breaking array spill integrity; copy and paste as values (Ctrl+Shift+V) for manual editing.

Verification Steps

1. Verify list uniqueness with =COUNTA(range) = COUNTA(UNIQUE(range)).

⚠️ Analyst Best Practices & Operational Hazards

1. Recalculation Cost: Never leave hundreds of AI. formulas running on Automatic Calculation in massive workbooks. Every edit triggers recalculation, exhausting API quotas and causing severe UI latency.
2. Convert to Static Values: Once verified, copy AI formula columns and use Paste as Values (Ctrl+Alt+V ➔ Values or Ctrl+Shift+V).
3. Never Trust Math to AI: Never use AI.ASK for sums, percentages, or weighted averages. Always use native Excel functions (SUMIFS, XLOOKUP, CALCULATE).