# Where to start with AI: a method to rank use cases

Start where the work is heavy and the data is ready. Most decision-makers already hold a long list of AI ideas. Four questions help you choose the first two or three, and you can run a first pass with your own team this week.

Octopus research team · 16 November 2026 · AI strategy & value

Most decision-makers in the Kingdom now have a long list of AI ideas. Some came from vendors, some from a workshop, and some from a board member who saw a good demo. Finding ideas is not the hard part. Choosing the first two or three is.

The choice matters more than it looks. On average, only 48% of AI projects reach production, and the move from prototype to production takes 8 months [1]. Much of that loss is decided on the day the first use case is picked.

The pattern holds closer to home. McKinsey surveyed 139 senior executives and board directors in GCC organisations. Of every 100 GCC organisations, 84 use AI, 31 are scaling it or have deployed it across the organisation, and only 11 see a clear impact on profit (5% or more of earnings) [3]. In the Kingdom, 33.1% of establishments used AI technologies in 2025, up from 27.6% in 2024, according to GASTAT [4] [5].

Gartner's survey also found that the primary obstacle to AI adoption, cited by 49% of respondents, is the difficulty of estimating and demonstrating the value of AI projects [1]. Gartner later predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, because of poor data quality, inadequate risk controls, escalating costs or unclear business value [6]. Most of those causes map to one of the four questions below.

Ideas usually arrive ranked by who asked loudest or which demo looked best. That ranking produces pilots. A better one starts from the work itself. We score each candidate on four questions: value, data, fit and risk.

## Value: how much work does it touch?

Count volume and hours: requests per month, calls per day, minutes per case. A modest gain on a process that runs thousands of times a month is worth more than a large gain on one that runs twice a quarter. Use your own numbers, even rough ones, before anyone's estimate. If nobody can say how often a process runs, that is a finding too. Measure it before you rank it.

## Data: is it there, and can you use it?

Check where the inputs live today, in what form, and who owns them. Paper and scanned files do not rule an idea out. They add a document step to the plan. Personal or classified data can be used once hosting and controls are settled up front, not after the design is fixed.

Poor data quality is one of the four reasons Gartner gives for generative AI projects being abandoned after proof of concept [6]. Test it on a sample before you rank: pull a hundred real records, check how many are complete and current, and note who would fix the rest. A use case whose data needs months of cleaning can still be a good one, but its first months belong to the data work, and the plan should say so.

## Fit: does it go into a system people already use?

Opportunities that live inside an existing system get used. That means the ERP, the case management tool or the internal portal. Ideas that need a new screen and a new habit need more change management, so discount their value until adoption is planned. A simple test: ask who would open it on an ordinary working day, and from which screen.

## Risk: what happens when it is wrong?

A draft that a person reviews carries little risk. A decision that affects someone's rights or money carries more, and under SDAIA's AI ethics framework it needs more testing and oversight. Start with use cases where a person stays in the loop and errors are cheap to catch. Classify each idea early, because its risk class sets how much testing, documentation and oversight it will need.

SDAIA's AI Ethics Principles sort systems into four tiers: little or no risk, limited, high and unacceptable. High-risk systems that affect basic rights need conformity assessments before and after deployment [8]. That is a strong reason to keep the first pick in the lower tiers. A high-risk use case can be the right one later, once the testing and oversight it needs are in place.

## Put it on one page

Plot the results on two axes, value against ease. The first candidates usually stand out: high-volume work, data you can reach, an existing system to build into, and a person reviewing the output. Look for them in contact centres, finance and procurement operations, HR services and internal knowledge search.

They are rarely the most exciting ideas on the list. They are the ones most likely to reach operations and pay for what comes next. Park the rest. Parked is not rejected; it means the idea waits until its data, its system or its owner is ready.

## Fewer, deeper bets

BCG's 2024 survey of 1,000 executives in 59 countries found that 74% of companies had yet to show tangible value from AI. The leaders pursued, on average, only about half as many opportunities as their less advanced peers, and generated 62% of their AI value in core business processes rather than support functions [7].

They also put their effort where the difficulty is. BCG's leaders follow a 10-20-70 rule: 10% of resources on algorithms, 20% on technology and data, and 70% on people and processes [7]. For a ranking exercise, that argues for two or three use cases in core operations, each with a budget for the process and adoption work, over ten small pilots spread across functions.

## A worked example

Take three ideas from a typical list (an illustration, not a client case). The first is an assistant that drafts replies to customer emails in Arabic and English, for a team that handles several thousand messages a month inside an existing case management tool. The second is a demand forecast for procurement that needs data from three systems that do not yet share a product code. The third is a public chatbot on the website, with no named person accountable and answers that would reach customers directly.

The email assistant scores well on value, data and fit, and its risk is low because an agent reviews every draft. The forecast scores high on value but low on data until the codes are aligned, so it is parked behind a data task with a date. The chatbot scores lowest: it needs a new channel, someone accountable and a higher risk class before it can be judged. The first pick is the email assistant, and the data work for the forecast starts in parallel.

## Match the first picks to the framework's horizons

SDAIA's AI Adoption Framework sets three time horizons. The short term, one to two years, focuses on quick wins by automating routine processes and activating predictive analytics to support decisions. The medium term, three to five years, moves to more advanced applications such as smart products and services. The long term, beyond five years, covers intelligent systems capable of self-learning and adaptation [2].

The ranking above fills the first horizon. High-volume, reviewable work inside an existing system is what the framework calls a quick win. The more ambitious ideas on your list usually belong to the second or third horizon, and the framework's own timing is a good reason to park them without dropping them.

> A pilot is not progress. A system in operations is.
> (Octopus)

## Why the first pick sets the pace

SDAIA's AI Adoption Framework treats the move from pilot to production as a maturity signal, and asks organisations to track it as a KPI [2]. A first pick that reaches operations moves that number and earns the budget for the second. A first pick that stalls teaches the organisation that AI is a demo, and the next proposal starts from doubt.

The framework also asks for each initiative to carry measurable outputs, such as work hours saved, beneficiary satisfaction or model accuracy, and for initiatives to be reviewed every quarter [2]. Write those outputs into the ranking sheet now: a use case whose output nobody can name is not ready to be ranked first.

This ranking is the core of the Octopus 4-week diagnostic: interviews with your teams in Arabic and English, a review of the processes and data behind each idea, and a ranked list of opportunities with a roadmap.

**Run a first pass this week**

- List ten candidate use cases, one line each.
- Next to each, write its monthly volume and the hours it takes today.
- Mark where its data lives, and whether it is digital, scanned or on paper.
- Name the system it would live in, and the person who would own it.
- Rate the risk low, medium or high, based on what a wrong answer would cost.
- Keep the top three and park the rest for now.

Start where the work is heavy and the data is ready. The exciting ideas will still be on the list next quarter, and by then a live system will have made the case for them.

Ranking and funding AI ideas this way is the work of our practice in AI strategy, investment and value realisation.

## Sources

Figures are quoted as published. Forecasts and drafts are labelled.

1. [Gartner, survey on generative AI deployment in organisations, 7 May 2024](https://www.gartner.com/en/newsroom/press-releases/2024-05-07-gartner-survey-finds-generative-ai-is-now-the-most-frequently-deployed-ai-solution-in-organizations)
2. [SDAIA, Artificial Intelligence Adoption Framework, 2025](https://sdaia.gov.sa/en/SDAIA/about/Files/AIAdoptionFramework.pdf)
3. [McKinsey and GCC Board Directors Institute, The state of AI in GCC countries: In pursuit of scale and value, November 2025 (survey of 139 senior executives and board directors)](https://www.mckinsey.com/middle-east/our-insights/the-state-of-ai-in-gcc-countries-in-pursuit-of-scale-and-value)
4. [GASTAT, Establishments' ICT Access and Usage Statistics 2025](https://www.stats.gov.sa/documents/d/guest/establishments-ict-access-and-usage-statistics-2025-en-pdf)
5. [GASTAT, Establishments' ICT Access and Usage Statistics 2024](https://www.stats.gov.sa/documents/20117/2435267/Establishments+ICT+Access+and+Usage+Statistics+2024-EN.pdf/2b4e45e8-ccbe-9767-1615-2f7d21236a9c)
6. [Gartner, prediction that 30% of generative AI projects will be abandoned after proof of concept by end of 2025, press release, 29 July 2024](https://www.gartner.com/en/newsroom/press-releases/2024-07-29-gartner-predicts-30-percent-of-generative-ai-projects-will-be-abandoned-after-proof-of-concept-by-end-of-2025)
7. [BCG, AI adoption in 2024: 74% of companies struggle to achieve and scale value, press release, 24 October 2024](https://www.bcg.com/press/24october2024-ai-adoption-in-2024-74-of-companies-struggle-to-achieve-and-scale-value)
8. [SDAIA, AI Ethics Principles, September 2023](https://dgp.sdaia.gov.sa/wps/wcm/connect/4c56ed1c-1b82-447d-ac29-638f5f99c12e/ai-principles-EN.pdf)

## Related practices

- [AI strategy, investment and value realisation](https://pioctopus.ai/en/services/ai-strategy-value)

## Put it to work.

The tool that goes with what you just read.

[90-Day Board Brief](https://pioctopus.ai/en/toolkit/90-day-board-brief): A board pack for the next 90 days of AI: three bets, budget, metric, kill criteria and owner.

## Keep reading

- [How to measure the ROI of AI projects in Saudi Arabia](https://pioctopus.ai/en/insights/measure-ai-roi-saudi-arabia): Set the baseline before the build, use the numbers finance already tracks, count the full cost and report value every month. A practical method for Saudi companies and government entities.
- [SDAIA's AI Adoption Framework, explained](https://pioctopus.ai/en/insights/sdaia-ai-adoption-framework): The national reference for adopting AI in the Kingdom, read from the source: what it contains, how binding it is, and where it meets SDAIA's other AI documents.
- [How to choose an AI consultancy in Saudi Arabia](https://pioctopus.ai/en/insights/choosing-an-ai-consultancy-saudi-arabia): Five kinds of firm sell AI work in the Kingdom. They differ most on who stays accountable after the roadmap. Here is how to compare them, and what to ask.

## Want to talk about your own operations?

- [Book a session](https://pioctopus.ai/en/contact)
- [Get the brief](https://pioctopus.ai/en/toolkit/90-day-board-brief)

---
Web page: https://pioctopus.ai/en/insights/where-to-start-with-ai
Arabic version: https://pioctopus.ai/insights/where-to-start-with-ai.md
Index for AI agents: https://pioctopus.ai/llms.txt
