The ACE Framework: A Periodic Table for Business AI

Meet Sarah. She runs a 180-person B2B SaaS company. Business is good. Last quarter was the best in the company's history.

But something about AI is quietly breaking inside her team, and she can feel it before she can name it.

It came to a head last Wednesday. Her Sales Director dropped a vendor contract on her desk: $180K annual, signature required by Friday, "AI-powered sales intelligence." Sarah flipped through the slides. She couldn't tell what the product actually did. Analyze her sales calls? Score her leads? Draft her reps' emails? The demo showed all three. The vendor's deck said "transform your sales motion" and "AI-powered insights." Her Sales Director was ready to sign.

She asked him to walk her through it. Twenty minutes later, they both realized neither of them could explain what they were buying.

Sarah didn't need another AI strategy framework. She needed a vocabulary. A way to tell, in five minutes, what a vendor's AI actually does, and whether her team needs it.

This framework is for Sarah. And for every founder, owner, and business leader building something ambitious enough that AI matters — who needs to evaluate vendors without a translator.

The insight: a finite alphabet, infinite expressiveness

A small number of fundamental primitives, combined, can describe anything. Carbon, hydrogen, oxygen, and nitrogen build most of organic chemistry. Zero and one build every program ever written. Twenty-six letters build every English-language work. Chemistry works this way because it has to. The universe is compositional.

Business AI works the same way. Every AI use case in your business, whether it's Salesforce Einstein scoring leads, Intercom Fin answering support tickets, or Gong transcribing sales calls, can be described as a recipe of five core capabilities operating on data. Understand the five, and you can read any AI product pitch, tag any internal initiative, or audit any tool in your stack.

That's what this framework does. It gives you a vocabulary. Not a strategy. Not a roadmap. A vocabulary.

The 5 Capabilities: Ingest, Analyze, Predict, Generate, Execute

Ingest takes in information. OCR on a receipt. Transcription of a sales call. Parsing a PDF invoice. Pulling CRM records via API. Anything that converts a raw signal (image, audio, document, stream) into something the AI can operate on.

Analyze makes sense of what was ingested. Classifies an email as urgent or FYI. Extracts a vendor name from a contract. Summarizes a 60-page report. Detects sentiment in a customer review. Analyze answers the question: what is this?

Predict forecasts what comes next. Scores a lead at 87% likely to close. Forecasts $4.2M in closed-won revenue for Q2. Flags a transaction as 99.5% anomalous. Predict answers: what's likely?

Generate produces something new. Drafts an email. Writes code. Creates an image. Composes a project plan. Generate produces an artifact, a thing that sits in draft form until something else (a human or another system) pushes it out.

Execute changes state outside the AI. Sends the email. Commits the code. Updates the CRM. Charges the card. Places the order. Execute has consequences that are often irreversible.

That's it. Five capabilities. Every piece of business AI does one or more of these.

Why these five are exhaustive

Walk through any AI product you use this week. It will use one or more of these capabilities. Not five of the "ten pillars of transformation" from a consulting deck. Not "seven drivers of AI-readiness." Five verbs, applied to data, composed into workflows.

Every piece of information processing that AI does falls into one of these categories. Taking data in (Ingest). Understanding it (Analyze). Forecasting with it (Predict). Producing something new from it (Generate). Using it to change the world (Execute). There is no sixth action hiding somewhere.

Could this set change over time? Yes. In three years, "Remember" (persistent AI memory across sessions) or "Coordinate" (multi-agent orchestration) might deserve promotion to first-class capabilities. That's fine. Frameworks should evolve. Today, these five cover the field.

What this framework doesn't do is blur them. Most consulting-firm AI frameworks lump everything into "AI transformation" or "AI strategy," fuzzy categories that feel important but don't help you pick a tool next week. Precise vocabulary is the precondition for clear thinking.

The six-layer stack

Capabilities don't exist in isolation. They stack. Each level builds on the one below it.

Everything starts at Foundation. Before any capability works, you need data readiness: data that's accessible, structured, fresh, and permitted. Clean data is usually the difference between AI that works and AI that embarrasses. Gartner reports that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. Boring work, decisive outcome.

The stack is a model, not a sequence. You don't have to master Level 1 before touching Level 3. A real business adopts AI tools (Level 3) while its data (Foundation) is still messy and its strategy (Level 5) is half-formed. The stack gives you a map, not a route.

Read any AI use case in five minutes

Here's how to apply the framework. Pick an AI product you use. Walk through these five questions:

1. What data does it consume? (Text, structured, image, audio, video, code, time-series)
2. Which capabilities does it use? (Ingest, Analyze, Predict, Generate, Execute)
3. What's the dominant pattern? (RAG, Scoring+Routing, Vision Extract, etc.)
4. What's the output: artifact or state change? (Generate-only, or does it Execute?)
5. Where does the human fit? (Review gate, monitoring, absent)

Worked example: Gong, the sales call analysis platform.

• Data: Audio (recorded calls) + Text (transcripts) + Structured (CRM records)
• Capabilities: Ingest (call audio → transcript), Analyze (topics, objections, sentiment), Generate (summary + CRM notes), Execute (writes back to Salesforce)
• Pattern: Meeting Intelligence
• Output: Both. Generates human-readable summaries. Executes by updating CRM records.
• Human: Rep reviews summary. Manager reviews coaching insights. Nothing in Gong auto-acts on the customer.

In five bullet points, you've fully described what Gong does and how it fits in your stack. Now try it on ChatGPT (Analyze + Generate, pure artifact, human must Execute). Try it on Stripe Radar (Ingest + Analyze + Predict + Execute, auto-flags fraud). Try it on Salesforce Einstein (Ingest + Analyze + Predict, then human reviews scores, optionally Execute via auto-routing).

In a week of this practice, you'll read vendor pitches differently. You'll stop being impressed by "AI-powered" and start asking which capabilities are actually active and where the human stays in the loop.

Predictive AI vs. Generative AI: a mapping

The tech industry currently divides AI into two camps: Predictive AI (scoring, forecasting, classification) and Generative AI (text, image, code creation). This split became popular after 2022 when ChatGPT made the Generative side newly visible.

The ACE Framework maps to both:

• Predictive AI uses the Predict capability (primarily) with supporting Analyze. This is Salesforce Einstein, HubSpot Predictive Lead Scoring, Stripe Radar, recommendation engines.

• Generative AI uses the Generate capability (primarily) with supporting Analyze. This is ChatGPT, GitHub Copilot, Midjourney, Jasper, Writer.

Most modern products combine both. A customer support agent predicts intent (Predict) and drafts a response (Generate). An autonomous coding agent analyzes a bug (Analyze), predicts a fix approach (Predict), writes code (Generate), and commits it (Execute). The "Predictive vs. Generative" binary is useful shorthand, but it misses Ingest and Execute entirely.

The generative AI market alone is projected at $121 billion in 2026, growing at 33.2% CAGR. Enterprise AI spending has grown from $1.7 billion to $37 billion since 2023. That's why every software vendor is racing to claim a position in one camp or both.

What this framework is NOT

Honest limits, up front:

It's not a prescription. The framework gives you vocabulary and structure. It doesn't tell you which AI to adopt Monday, which vendor to buy, or how to run the change management. For that, you need playbooks built on top of this.

It's not a maturity model. The six layers aren't stages you advance through. A business can have sophisticated Patterns (Level 2) without a Transformation Strategy (Level 5), or vice versa. The stack is structural, not sequential.

It's not static. AI evolves quickly. The framework will need revisions, probably quarterly. Capabilities might split. New ones might emerge. We commit to keeping it current, not to being right forever.

It's not technology-specific. No dependency on GPT-5, LangChain, or any specific tool. Those change every six months. Capabilities outlast products.

It's not enough on its own. Citing the framework isn't analysis. Real articles built on it must add real examples, failure modes, and honest ROI data.

It's new. This framework was published in 2026. It will need to prove itself over time. Some pieces will hold up. Some will need revision. We'll update this article as we learn.

If those limits still leave room for a framework you can use this week, keep reading.

How to use this framework

1. Audit your current AI stack. List every AI tool your company uses. For each, tag the ACE capabilities it covers. You'll find redundancy (three tools doing Generate+Text), gaps (zero Predict capabilities despite paying for "predictive analytics"), and surprises (the tool you thought was modern is just Analyze with a chat UI).

2. Read vendor pitches skeptically. Most AI vendor pitches use words like "intelligent," "automated," "transformative." Replace those with ACE capabilities and suddenly the pitch becomes precise. A tool that "transforms your sales workflow" is, concretely, probably Analyze + Generate + maybe Execute. If the vendor can't tell you exactly which capabilities their product performs, that's information.

3. Evaluate your own AI initiatives. Are you investing in balanced capabilities, or all on Generate? Do your riskiest workflows have the right human-in-loop at the Execute boundary? Are you doing the unglamorous Foundation work, or chasing the latest LLM?

Per the OECD, 61% of SMBs cite cost as the primary barrier to AI adoption, followed by lack of expertise (54%) and data quality (41%). The ACE Framework doesn't solve cost. It solves the expertise gap: the vocabulary you need to make informed decisions without paying McKinsey $2M. That's the narrow claim, and we stand by it.

What's next

The rest of this collection builds out each layer:

• Foundation: the seven data types and the data-readiness work that makes AI possible
• The 5 Capabilities: one deep-dive per capability with real examples and failure modes: Ingest, Analyze, Predict, Generate, Execute
• Beyond the Capabilities: boundaries, practical use, evolution, and tagging AI initiatives

After this collection, the Patterns collection (Level 2) covers the ten recurring capability combinations that solve 90% of real business problems. Then AI for [your role] (Level 3) for role-specific depth. Then industry and strategy content for the higher levels.

The ACE Framework is a tool. Use it when it serves the work. Set it aside when it doesn't. The work is the work.

---

The ACE Framework was built by Rework's content and research team, April 2026. This document is a living reference and will be revised as the AI landscape evolves. If you spot an error or have a suggestion, we want to hear it.