How We Test AI Slide Tools: Our 10 Consulting Criteria

We score every AI slide tool and model on 10 criteria that describe a consulting-grade deck. A consulting-grade deck is one a McKinsey, BCG or Bain team would send to a client and is based on our experience at the those three consulting firms. The 10 criteria fall into three groups (the individual slide, the deck as a whole and the finish overall), and each is scored pass, partial or fail. Price, speed and exported file format are reported next to the score, along with a date for testing and model settings.

In this article, we explain why we have decided to use a fixed standard, what our 10 criteria are in more detail, what the method doesn’t cover, and examples of what we mean to make it concrete.

Why we use a fixed standard

With AI tools and models popping up and getting better week to week it can be difficult to get an overview of how good each new tool or model is, and especially to assess how well they function in a typical business or consulting workflow.

We’re interested in creating high-quality, consulting-level slides that ideally match the outputs from top-tier houses like McKinsey, BCG, and Bain. Therefore, we decided to set up a fixed standard and sets of tests to be able to objectively and consistently judge outputs of new models or specialized AI slide tools and thus compare them and ultimately recommend the best ones to use.

The standard is based on the training and techniques that were drilled into us during our years as consultants at Bain, BCG, and McKinsey. We’ve synthesized them here in 10 criteria that reflect both the little things that make a difference like consistent font sizes or well-formulated action titles, as well as the big things that make decks look and feel client-ready like clear messages, the right amount of information on slides, and compelling storylines.

Let’s dig in.

How we set up a test

A score only means something if every tool gets the same task. In our tests, we’ve narrowed it down to the following setup:

1. The same brief: Each tool receives the same prompt with the same source material, when relevant. We purposefully try to test mostly finished research and reports to limit the effects of how well the different models conduct research and only focus on how they turn data into slides. 
   Our standard source materials are a 52-page industry report Oil & Gas in Egypt: Market Report, a 28-page public industry report on the China Juice Market Report, and a made-up business case that we ask the tools to make better. Each test post describes the brief it used.
    
2. The same slide master: When relevant for the test, we use our own Slideworks PowerPoint master, which represents a neutral, corporate master and color palette. Where a tool cannot take a master, we note it and score criterion 10 as a fail.
    
3. The same starting point: We use the prompt as written, and typically follow the recommended options if a model or tool suggests something. We do not edit the output by hand before scoring.
    
4. Recorded and dated: We note the tool, the model or version, the plan tier and the test date. Where possible, we publish an unedited screen recording of the run.
    
5. Fixed criteria: We use the same 10 criteria to score each output. We take turns scoring output and often discuss it between us. When the call is close, we say so in the review.

As a rule of thumb, a test is only comparable with another test if the brief and the method version match. Each review states both in a method box near the top.

The 10 criteria we use to score AI slide tools

We have set up 10 criteria that fall into three main sets; the first set judges each slide individually, the second set judge the deck as a whole, and the third set judges the overall output look and feel:

The last criterion works a bit differently from the rest as it’s the ultimate yes/no pass and verdict. We’ll describe this in more detail later. Now let’s look into each of the criteria in more detail.

The slide

1. Layout fits the message

A slide's layout should emphasize the data that carries its message. It makes a big difference whether you use a table, a chart or a diagram to get a message across. Choosing the layout is a skill consulting firms train through years of seeing examples and getting feedback, and it’s where AI tools (so far) often default to the generic.

• Pass: Every slide uses a layout that matches its message e.g., a trend as a line or column chart, a comparison as a table or bar chart, a process as a flow etc.
• Partial: Most slides fit, but one or two use a generic layout, such as bullets where a chart would make the point.
• Fail: Layouts look chosen for variety or decoration, or most slides default to text boxes and bullets.

2. Only relevant data

Just like it's important to choose a layout that communicates the key message, it's also important to clean the slide for any details or info that make it more difficult to grasp the key message. It's fine adding detailed info (in fact, it's encouraged with techniques like the pyramid principle) but having data on the slide that is not directly connected to the main message makes it confusing.

• Pass: Every element on every slide supports the title's message.
• Partial: A few slides carry loosely related detail that should be cut or moved to the appendix.
• Fail: Slides are crowded with information that competes with the main message.

3. Action titles

An action title states the slide's main message as a complete sentence, for example "Revenue grew 12% a year, driven by enterprise segment" rather than "Revenue development". Action titles are a core part of any consulting deliverable and read in sequence, they should tell the whole story. See our post on how to write action titles like McKinsey for reference.

• Pass: Every content slide has a full-sentence title that states its takeaway, and the body proves it.
• Partial: Titles are sentences, but some describe the topic rather than the conclusion, or claim more than the body shows.
• Fail: Titles are topic labels.

4. Consulting tone and depth

Consulting slides are often quite detailed, and the text carries a point of view or conclusion on whatever the slide shows. AI text tends towards fluff; correct, general and safe.

• Pass: Text is specific (numbers, names, causes) and states a conclusion.
• Partial: The right level of detail, but neutral or descriptive in tone.
• Fail: Generic statements that could appear in any deck on the topic.

The deck

5. Visual balance

This is one of the most important aspects of a consulting deck and what helps lift the outputs from fine to great. The mix of layouts across the deck should be varied and support the arc of the storyline.

• Pass: Layouts vary with the content with a balanced mix of different types of layout that both suit the content and are interesting in plenum.
• Partial: Some variation, but a default to same layouts throughout.
• Fail: A handful of layouts repeated in the deck, especially in sequence.

6. Storyline structure

Storylining is a key part of what makes consulting decks good. Different types of deck follow different storylines, and the audience and setting change the structure again. A business case for a steering committee is not built like a market study for a strategy team.

• Pass: The structure fits the deck type and audience, and the executive summary matches the body.
• Partial: A sensible structure, but a generic one, or with a section out of order.
• Fail: A list of topics rather than an argument.

7. Coherent messages and numbers

Messages and numbers must agree across the whole deck. A figure on slide 3 should match the same figure on slide 18, and the conclusion should follow from the evidence.

• Pass: No contradictions in messages or numbers, and every number traces to the brief or a stated source.
• Partial: One or two minor inconsistencies, such as rounding or a label mismatch, that does not change a conclusion.
• Fail: Any number or message that contradicts another, or a number that cannot be traced.

The finish

8. Looks like real consulting work

Setting aside colors and fonts, do the slides look and feel like real McKinsey or BCG slides? This is arguably one of the most subjective criteria and so we try to anchor it both in our own experience and against real decks from McKinsey, Bain, and BCG.

• Pass: The slides look and feel like real consulting slides with complex layouts, chart-heavy slides, and clear arguments.
• Partial: The right style, but visibly or informationally lightweight.
• Fail: The slides read as generic presentation software output.

9. Follows the brand

An important part of any slide workflow is to be able to follow a certain slide master and brand guidelines. A deck that requires a recoloring and rebranding compared to the requirements won’t fly, however good its content.

• Pass: The output uses the master's layouts, fonts, colors and placeholders and the master carries over into the output to be easily changed.
• Partial: Colors and fonts are applied, but positions, sizes or placeholders drift from the master OR the master does not carry over into the output and each slide is linked to a generic slide master.
• Fail: The master is ignored or cannot be supplied.

10. Client-ready

Would we send this deck to a client? This is the verdict, explained in the next section.

How the scores become a verdict

Each of criteria 1 to 9 gets one of three marks: pass, partial or fail. We use three levels rather than a 1-to-10 scale on purpose so there is no false precision.

The verdict, criterion 10, follows three rules:

1. Client-ready means the deck could go to a client after a final proofread. That requires no fails and at most one or two partials.
2. Client-ready after editing means the structure and content are sound, but slides need rework. We state our estimate of the editing time, for example "about 3 hours".
3. Not client-ready means fixing the deck would take about as long as building it.

In every review, the verdict appears first, followed by the nine marks. In comparisons, we total the passes and partials (pass = 2, partial = 1, fail = 0, out of 18) so tools can be sorted.

What we report next to the score

Price and speed matter to anyone choosing a tool, but they are not measures of quality. A cheap tool that produces a mediocre deck is still a mediocre deck. We therefore report four facts next to every score and never fold them into it.

What this method does not measure

This method is one we developed internally to test and judge new AI slide tools. Like any other test method, it’s not able to cover every aspect. These are the limits we know about:

• It’s our own judgement: The pass, partial and fail lines make the judgement consistent and open to challenge, not objective. We publish the outputs and recordings so readers can check our calls.
• It's based on consulting standards: Linked to the previous bullet, then our judgement is based on consulting slides. Our tests are not relevant if you're evaluating slide tools for image-heavy sales pitches or graphical funding decks. We focus solely on the consulting/business sphere.
• It’s a snapshot: AI tools change every few weeks. Every score carries its test date, and we re-test when a tool releases a major update.
• It measures the output, not the experience: Ease of use, security and enterprise features matter, but they are outside this method. Where they matter for a decision, we cover them separately in the review.
• It tests our briefs: A tool may perform better or worse on decks unlike ours. We use common consulting briefs to keep the test relevant to our readers.

Version history