Excel to Automated Presentation: Weekly Slides From a Template, Checked and Versioned
An automated presentation from Excel is a weekly deck generated from a template, with every number taken from the source table instead of pasted by hand. Saving time is the obvious gain; the bigger one is that a wrong number never reaches a slide. In a hand-updated deck two things slip through easily: the headline total stops matching the sum of the bars in the chart, and the commentary contradicts the table. I build this as a web app around your company: it produces the deck from your own PowerPoint or Google Slides template, the deck can’t be sent until the checks pass, every sent file is a numbered version, and an AI drafts the summary from the table’s numbers for a person to approve.
Checking month-end source files, tying totals to the trial balance and locking closed months is a separate job, covered on the month-end management report page. If one person prepares the deck and product groups change once a year, an Excel–PowerPoint link is usually enough; if several people feed it or management asks which version it saw, you need an app. I haven’t compared ready-made Excel-to-slides tools or AI assistants on this page; whichever you use, the checks below are still yours to run.
A sample company: 14 slides, 4.5 hours a week
Sample data, fictional. Picture a 240-person pump and valve manufacturer. Before its Monday 9:30 sales and operations meeting, a 14-slide deck goes to 9 people, from the managing director to four regional managers. A sales operations analyst prepares it; the template belongs to the analyst’s manager, the sales operations manager. Amounts are in the company’s reporting currency.
| source | where it comes from | format |
|---|---|---|
| Order lines | ERP order report, pulled Friday 5 pm | Excel |
| Shipment list | ERP shipment report | Excel |
| Customer complaint log | spreadsheet kept by quality | Excel |
| Weekly targets | the director’s annual budget file | Excel |
Today the deck takes about 4.5 hours a week (sample data): 3 hours after the Friday 5 pm cut-off downloading files, refreshing tables and moving charts onto slides, 1.5 hours on Monday fixing numbers and writing the summary. Over 48 meeting weeks that is 216 hours a year.
The sample week: what did the checks catch, and who decided what?
Three checks fired and the deck stayed in draft. They are grouped here for the example. The analyst uploaded the files on Friday at 5:20 pm and the app reported this (sample data):
| check | what it found | what happened |
|---|---|---|
| Headline vs chart total | order intake on the summary slide 6,412,000; the four bars in the product group chart add up to 6,226,000: a gap of 186,000 (2.9%) | a new product group, “Service kit”, had been opened in the ERP that week and wasn’t in the template’s mapping table; 2 order lines never reached the chart |
| Target threshold | on-time delivery 89.6% against a 92.0% target, 2.4 points below (a point is the difference between two percentages); 92.1% the week before | slide marked “comment required” for the operations manager |
| Commentary draft | the AI draft said “on-time delivery rose 2.5 points” | right number, wrong direction: it fell 2.5 points; the draft was blocked and the analyst corrected the sentence |
The first row is the quietest error: a new group raises no error, it just isn’t in the chart. Previous week’s order intake was 6,044,000. The chart suggests a 3.0% rise (6,226,000); the real rise in the ERP is 6.1% (6,412,000).
The week closed in three steps:
- The sales operations manager decided to show “Service kit” under “Spare parts” this week and as its own bar from next week. As template owner, the manager approved the new mapping row on Friday evening. The spare parts bar went from 1,388,000 to 1,574,000 and the gap dropped to zero.
- On Monday morning the operations manager explained the on-time slide. Of the 43 late shipment lines, 26 came from a two-day press line stoppage; had they shipped on time, the rate would have been 95.9%. The explanation and action went onto the slide under the manager’s name.
- The sales and operations director approved the deck at 8:40 on Monday, and version 1 went to the distribution list.
At 10:15 the quality manager found one complaint logged twice and deleted the duplicate: 7 complaints became 6. It had been entered under two different numbers, so the number check didn’t catch it. Version 2 went to the same 9 people at 10:40 with the note “changes since version 1: slide 9 and summary, complaints 7 → 6”.
I expect Friday’s 3 hours to shrink most; commentary stays with people. That is an expectation, not a measurement.
Template and placeholders: where does each slide get its numbers?
The template, your current deck, says where every number and chart goes and which source feeds it; design, fonts and slide order stay. Only the hand-typed numbers become named placeholders: instead of “6.4 million” the summary slide holds {{orders_week}}.
Part of the sample template (the structure I propose):
| slide | placeholder or element | source | check |
|---|---|---|---|
| 2 Summary | {{orders_week}}, {{on_time_rate}}, commentary | order lines, shipments | text vs table; headline vs chart |
| 3 Order trend | line chart, last 12 weeks | archive of sent weeks | all 12 weeks present |
| 4 Product groups | column chart, 4 groups | order lines + mapping table | chart total = headline; no unmapped group |
| 8 On-time delivery | rate, target, chart | shipments + targets | comment required below target |
| 9 Complaints | count, breakdown | complaint log | no duplicate complaint number |
| 14 Control page | upload times, check results, approver | the app’s own log | always last |
The mapping table links source codes to deck names (ERP “PMP” → “Pumps”); an unknown code stops the deck.
With dashboards, the most common mistake I see is putting every piece of data on them until nobody can read them. For a deck my advice is the same: don’t add slides when you automate. For the Excel summary sheet itself, see the Excel dashboard setup page.
How do charts and tables update with the data?
In the structure I propose, the app opens a copy of the template each week, fills the placeholders and aims to write the chart data from the source table; the master is never touched. I aim for two forms; which one works with your template is written down in the first step’s output:
- PowerPoint: per Microsoft, a .pptx is a ZIP package, each slide is a separate XML part, and a slide part can relate to a chart part (Microsoft Learn, structure of a PresentationML document, read 16 October 2026). I aim to fill placeholder text in those parts and write chart data without turning charts into pictures; the latter depends on your template. A chart may also carry an embedded workbook; if only the chart is updated, Edit Data can show the old table; how to handle that, and whether a PDF is needed, goes in the first step’s output.
- Google Slides: Google’s developer guide describes filling
{{…}}tags with thereplaceAllTextrequest, on a copy rather than the master (Google for Developers, merge). A Sheets chart added asLINKEDcan be refreshed at any time withrefreshSheetsChart; it must exist in Sheets first, and a chart added as a static image can’t be refreshed (Google for Developers, add charts). I read these on 16 October 2026.
Either way, the program that writes a number onto the slide is the one that checks it.
Rules that keep wrong numbers off the slides: my starting proposal
A deck can be sent only when every source arrived on time, every slide total ties to its source and every metric below target is explained. If any fails, the deck stays in draft and the send button stays off. This is my starting proposal, not a measured truth; you set the thresholds.
| rule | what it tests | if it fails | my proposal |
|---|---|---|---|
| Source timing | file pulled before the cut-off (Friday 5 pm here); row count below half the 8-week average | draft; the source owner gets a task | one written cut-off for everyone |
| Unmapped code | a group, region or product missing from the mapping table | deck stops; the template owner decides | never silently dropped into “Other” |
| Headline vs chart | each breakdown chart adds up to its headline | draft | ±1 for rounding |
| Same history as sent | earlier weeks in the trend match the versions sent then | differences listed for approval | past weeks never change silently |
| Target threshold | a metric below target | “comment required”; no approval without it | comment under the metric owner’s name |
| Text vs table | every number, direction and unit in the text exists in the table | text can’t be sent | see the AI section |
Why “same history as sent”: if the trend re-reads 12 weeks from the ERP, an order booked late against an earlier week changes that bar, and what management saw three weeks ago no longer matches. I propose drawing the trend from an archive of sent weeks; a late record goes into the current week, labelled.
Staying in Excel: a headline vs chart total check
In a “Raw” sheet of ERP lines, column A is the product group code and B the amount. A “Mapping” sheet has the code in A and the deck name in B. On the “Deck” sheet, A2:A5 hold the four group names and the chart uses B2:B5.
- Raw!C2 (deck group):
=IFERROR(VLOOKUP(A2,Mapping!A:B,2,FALSE),"UNMAPPED") - Deck!B2 (group total):
=SUMIF(Raw!C:C,A2,Raw!B:B) - Check!B1 (headline):
=SUM(Raw!B:B) - Check!B2 (chart total):
=SUM(Deck!B2:B5) - Check!B3 (gap):
=B1-B2 - Check!B4 (unmapped lines):
=COUNTIF(Raw!C:C,"UNMAPPED") - Check!B5 (status):
=IF(ABS(B3)>1,"MISMATCH","ok")
The argument separator depends on your regional settings. Copy column C down the data rows. I recomputed this logic in a script on the sample week’s 14 fictional lines: gap 186,000, 2 unmapped lines, “MISMATCH”; a “Service kit” mapping row brings the gap to zero. I didn’t run it in Excel itself, so try it once on a copy of your file.
Excel runs the check but doesn’t stop you sending the deck while B5 says “MISMATCH”.
Versions and the distribution list: which number did management see?
Every send is a numbered version, old versions are kept, and the distribution list records who received which. If someone in the meeting says “my file says 7 complaints”, the record shows they read version 1.
- The distribution list belongs to the deck, not an email; the approver changes it, on record.
- A sent file is never edited; a correction is a new version, not “deck_final_v3_fixed.pptx”.
We choose the channel the deck goes out on together in the first step.
Who prepares, who comments, who approves?
The preparer never approves; in the app that is a rule. Sample roles (my proposal):
| role | what they do | what they see |
|---|---|---|
| Analyst (preparer) | uploads files, builds the deck, corrects the AI draft | whole deck, control page |
| Data owners (ERP, shipping, quality) | upload or fix their own source on time | own source, control page |
| Metric owners (operations, quality) | explain slides below target | own slides, whole deck |
| Sales operations manager (template owner) | changes template and mapping table | everything |
| Director (approver) | approves and sends, manages the distribution list | everything |
| Recipients | read the version they receive | sent versions |
In a small team too, if the director prepares the deck, the managing director approves; separate accounts make the rule real.
The audit trail is an undeletable row for every version and decision: uploads, check results, mapping changes, comments, the AI draft and edits to it, approval and recipients, each under a person’s name.
AI writes the commentary, not the numbers
In this app the AI only writes sentences: the summary and draft explanations for metrics below target. The app gives it each metric’s value, unit and direction, and it may use only those. For which report sections AI can prepare, see monthly report automation with AI.
Number, direction and unit checks. The app extracts every number in the draft and compares it with the table. The model wrote “on-time delivery rose 2.5 points”. 2.5 points was in the table, so a number-only check would have passed it, but the table’s direction was “fell” and the direction check blocked the draft. The method I propose: direction words (rose, fell, grew, dropped) are matched with the table’s direction for the metric in the same sentence. It may miss negations (“didn’t fall”) or wording without a direction verb (“stayed below target”). Units are checked too: the drop is 2.5 points but 2.7% in relative terms, so “fell 2.5%” would also have been blocked.
Where the check stops. A model can still mix up two metrics or invent a cause, so the preparer approves every draft. It couldn’t know the press line had stopped; the operations manager wrote that.
What goes to the model? No row reaches it without passing the permission filter. In the sample deck it receives only summary values; customer and sales rep names stay out, region codes go instead. A rep’s named performance is personal data; if the model runs abroad, check the transfer rules where you operate.
In Turkey, that transfer falls under Article 9 of KVKK (Law No. 6698), rewritten by Article 34 of Law No. 7499 (Official Gazette 12 March 2024, No. 32487) with effect from 1 June 2024. I read the amending text in the Official Gazette and the current wording of Article 9 on mevzuat.gov.tr on 15 October 2026. This isn’t legal advice; I list in writing which fields go to the model for your legal adviser.
The AI can’t override checks, change the mapping table, approve or send the deck.
When an Excel–PowerPoint link or a linked Slides chart is enough
In the simple case from the introduction a link is enough and I wouldn’t recommend an app. I read the following on Microsoft’s and Google’s help pages on 16 October 2026.
PowerPoint:
- A worksheet links through Insert > Object > Create from file > Link; Microsoft describes updating as right-clicking the data and choosing Update Link (Microsoft Support).
- If the source workbook is moved or deleted, the link breaks; File > Info > Related Documents > Edit Links to Files lets you change the source or break the link, after which the chart’s data no longer updates (Microsoft Support).
- PowerPoint for the web can link to an Excel chart only if it is on a cloud service such as OneDrive or a website; local files need the desktop app (Microsoft Support).
Google Slides (Google Docs Editors Help):
- Insert > Chart > From Sheets with “Link to spreadsheet” ticked keeps the chart linked. You refresh it with its Update button or through Tools > Linked objects > Update all, a manual step per the help page; an update overwrites edits made in the deck.
- Anyone with access to the deck sees linked charts and tables even without access to the source spreadsheet. If a regional manager’s deck contains another region’s chart, they see it whether it is linked or not; the difference is that with a
LINKEDchart, people with access to the deck also see a link to the source spreadsheet (Google for Developers). Splitting slides by recipient is a distribution decision.
A link refreshes the chart but doesn’t ask whether a new product group made it in, whether headline and chart tie, or whether the commentary contradicts the numbers; an unrefreshed chart just keeps last week’s numbers.
How to start, and who owns the data and the code
The first step isn’t software; it’s mapping the deck you have: which number comes from which file and calculation, which checks are done by hand, who receives it. The written output:
- the placeholder list, with the source of every number and chart,
- the mapping table and proposed thresholds,
- the role and distribution list,
- the scope and timeline of the first version.
In automation the most common mistake I see is starting with the most complex process. So the first version is one deck and two or three sources:
- Template, order and shipment sources, headline vs chart check; matched against the last hand-built deck.
- Target threshold, metric owners’ comments, approval, versions and distribution list.
- Trend charts from sent versions, and a second deck if there is one.
- AI commentary, only after several matching weeks.
Where does the data live? The app runs on a rented virtual private server (VPS) that can be opened in your company’s name. The data is yours, and you can take all of it as Excel or CSV whenever you want; how sent deck files are handed over is written down in the first step. In a small or mid-sized team there is no per-user licence; the cost is the server’s monthly fee plus maintenance.
Who owns the code? The full source code sits in a repository you have access to, with run instructions, a dependency list and a known-limitations list. Keys and accounts are opened in your name. Because decks carry sales and customer data, I assess access, security, hosting and maintenance separately at the start.
What if the ERP changes or the template is redesigned? Sent versions and the audit trail stay. Uploads and mapping are rebuilt for a new ERP, placeholders for a new template; scope is agreed in writing.
Send the deck you update every week, numbers masked, through the contact page. In the first meeting we work out which slide is fed by which file and which number reaches a slide unchecked today.
Frequently Asked Questions
Is our corporate template and design kept?
That is the aim; it shows once the first version is built on your template, and anything that can’t be kept goes in the first step’s output. Redesign only after the first version matches the hand-built deck, or you can’t tell whether a difference came from design or calculation.
If a number changes during the meeting, does the deck update live?
In the structure I propose, no, on purpose. A sent deck freezes the numbers so the meeting and everyone’s file agree. A correction becomes a new version with its change list.
Can the same app produce both a weekly and a monthly deck?
Yes; each deck has its own template, sources and distribution list, and monthly numbers can come from the weekly archive.
Can the source data come straight from the ERP instead of Excel files?
With an integration interface, yes; otherwise file uploads work with the same checks. Either way the pull time shows on the control page.
