AI use case prioritisation is the process of scoring AI ideas from every department against the same criteria to decide which to build first. Each use case is rated 1 to 5 on business value, data readiness, task fit, ease of integration and risk; the weighted total and individual scores lead to a decision: pilot now, queue, prepare first or redesign.
Key takeaways
- AI use cases are scored 1–5 on five criteria: business value (×3), data readiness, task fit, risk (×2) and integration (×1); the maximum total is 50.
- Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025 (July 2024).
- A high-value use case whose data is not ready is not dropped; it becomes a data preparation project and is rescored a few months later.
- The first wave pairs one quick win with one strategic use case, and every pilot starts with a measurable success threshold written in advance.
- The prioritisation matrix is reviewed quarterly; gains measured in pilots replace the estimated value scores.
Why should AI ideas be prioritised?
Once AI is on the agenda, the wish list grows quickly. Finance wants invoice capture, sales wants proposal drafts, HR wants CV screening, operations wants smarter scheduling and the board wants to "ask the reports questions". Each department sees its own idea as the most urgent, and the winner is often the one with the loudest sponsor. If you do not have a list yet, our round-up of 30 AI use cases in business makes a good starting pool.
This prioritisation guide is for companies that have already taken the first step with AI. If you have not started at all, read where to start with AI in business first. If what you need to rank is rule-based process work, our process automation prioritisation guide is the better fit. Here we assume you have AI ideas from several departments and want a single, defensible ranking.
What does picking the wrong AI use case cost?
Many AI projects fail not on technology but on the choice of use case, and the warning signs are often visible before work begins:
Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, due to poor data quality, inadequate risk controls, escalating costs or unclear business value. (Gartner press release, July 2024)
A proof of concept (PoC) is the small-scale trial that shows an idea can work. All four of the causes Gartner lists, however, can be scored before the project starts: data, risk, cost and value. RAND's 2024 report on the root causes of AI project failure, based on interviews with 65 experienced data scientists and engineers, notes that by some estimates more than 80% of AI projects fail, twice the rate of IT projects without AI. We look at the other lessons from studies and real cases in why AI projects fail.
Where the value sits matters too. BCG's 2024 AI adoption research found that surveyed companies derive 62% of AI value from core functions: operations (23%), sales and marketing (20%) and R&D (13%). Support functions contribute 38%, led by customer service (12%), IT (7%) and procurement (7%). A portfolio made up only of easy back-office wins may leave most of the value on the table. Examples from two of those support functions are in AIOps for IT operations and AI in procurement.
Which five criteria are used in AI use case prioritisation?
In an AI prioritisation matrix, rate each use case from 1 to 5 on the criteria below and multiply by the weight. The maximum total is 50.
- Business value (×3): What does the use case save or earn in time, errors, revenue or risk? Estimate it from what the task costs today; our guide to measuring AI project ROI covers the calculation in detail.
- Data readiness (×2): What data will the model use, and can you get to it today? Scattered, duplicated or incomplete records pull the score down, and data quality and duplicate records are usually the first obstacle.
- Task fit (×2): Is this the kind of work AI does well, such as reading text, summarising, classifying, drafting or spotting patterns? Where the job needs exact calculation or a single correct rule, rule-based automation usually wins.
- Ease of integration (×1): Can the output land in the ERP, CRM or inbox your staff already use? Score high if there is an API, low if the only way in is through the screen.
- Risk (×2, reverse-scored): Does an error reach customers, employees or regulators? Personal data, recruitment or credit decisions score 1–2; internal drafts reviewed by a person score 4–5.
The 2023 Harvard–BCG experiment shows why task fit deserves its own line. On 18 tasks inside AI's capability frontier, consultants using GPT-4 finished 25.1% more quickly. On a task deliberately chosen to sit outside that frontier, those using AI were 19 percentage points less likely to reach the correct answer. The same tool, pointed at the wrong job, did harm.
A sample AI use case matrix across departments
Below is an illustrative matrix scoring use cases common in a mid-sized manufacturing and trading business. Your scores will differ; what matters is one scale for every department.
| Department | Use case | Value | Data | Fit | Integration | Risk | Total (50) | Decision |
|---|---|---|---|---|---|---|---|---|
| Finance | Extracting and matching supplier invoice data | 4 | 5 | 5 | 3 | 4 | 43 | Pilot now |
| Customer service | Reply suggestions and case summaries for agents | 4 | 4 | 4 | 3 | 4 | 39 | Pilot now |
| Sales | Drafting proposals and follow-up emails | 3 | 4 | 5 | 4 | 4 | 39 | Pilot now |
| Marketing | Product copy and content drafts | 2 | 4 | 5 | 5 | 4 | 37 | Queue |
| Procurement | Comparing supplier quotes side by side | 3 | 3 | 4 | 3 | 4 | 34 | Queue |
| Production | Order priority and scheduling suggestions | 5 | 2 | 3 | 2 | 3 | 33 | Prepare data first |
| Management | Asking reports questions in plain language | 3 | 3 | 3 | 2 | 3 | 29 | After data warehouse |
| HR | CV pre-screening | 3 | 3 | 3 | 3 | 1 | 26 | Redesign |
When reading the prioritisation matrix, look at the individual scores, not just the total. Production scheduling has the highest business value but ranks low because its data is not ready. It is not dropped; it becomes a data preparation project first, as we explain in AI in production scheduling.
CV pre-screening falls to the bottom because of risk. AI used to screen job applicants is classed as high-risk under the EU AI Act, and the Act can reach Turkish companies that place such a system on the EU market or whose output is used in the EU. In Türkiye, processing candidate data also needs to be assessed under KVKK, the country's personal data protection law. Reframing it as something lower-risk, such as drafting replies to candidate questions, is a safer entry point for AI in human resources.
We have separate guides for the top rows of the matrix (invoice data extraction and agent support): AI in finance and accounting covers invoice and reconciliation work, while agent assist for customer service details tools that help human agents rather than replace them.
How do you prioritise AI use cases? Six steps from scores to portfolio
- Collect ideas with the people who own them. Write each use case as one sentence: who, with what input, producing what output. Not "AI in sales" but "drafting a first proposal from an incoming quote request".
- Score as a mixed team. The department head knows the value, the person doing the work knows the exceptions, and IT knows the data and integration. One person's score reflects one person's bias.
- Record the reason behind each score. A note such as "Data: 2, because order history sits in two spreadsheets with mismatched product codes" is invaluable when you rescore six months later.
- Balance the portfolio. In the first wave, pick one quick win (high total) alongside one strategic use case (high value, low readiness). The quick win builds confidence while data preparation for the strategic one runs in the background.
- Set a success threshold for every pilot. Before the pilot starts, write down a measurable result that justifies rolling it out, for example a target level for handling time or error rate. A pilot that misses the threshold stops, and resources move to the next use case.
- Rescore every quarter. A new data warehouse or system changes some scores, and use cases that were parked can move up the list.
Pegasus Airlines offers a public example of a large use case portfolio. According to the company, as described in a Microsoft customer story, at the time of publication it had almost 100 AI use cases in its portfolio, 30 of which were active. Even at that scale only a share runs at once, so prioritisation is an ongoing decision, not a one-off exercise.
Which method suits each AI use case?
Prioritisation scoring answers "what"; "how" is a separate decision that drives cost. A method column makes the budget conversation easier:
| Use case type | Suitable method | When is a custom model needed? |
|---|---|---|
| Reading documents and invoices | Document OCR plus a language model | When layouts are highly specific and numerous |
| Q&A over company knowledge | Assistant grounded in your own documents (RAG, retrieval-augmented generation) | Rarely; improve search quality first |
| Drafting text | Off-the-shelf language model, templates and an approval step | When house style and terminology are very distinctive |
| Forecasting and planning | Model trained on historical data | Usually; the data is unique to you |
| Visual inspection | Computer vision model | Almost always; your products and lines are unique |
The method column can hold one more option: an existing product. Checking it before building RAG or a custom model protects the budget. Common use cases such as extracting fields from incoming invoices or drafting follow-up emails already run without a project in SmartMail business email, part of our Smart360 product family, through the attachment analysis and writing assistance features developed by Digital Bridge. Before scoring such a use case as a custom project, test how much of it the product covers on your own data; that brings the cost assumption in the matrix closer to reality. For businesses starting small with ready-made tools, AI for small businesses covers this option too.
How does Digital Bridge run AI use case prioritisation?
- Use case workshop and scoring. Through our digital transformation consultancy we gather use cases with department heads and the staff who do the work, score them on value, data, task fit, integration and risk, and produce a written matrix with the reasoning behind each score.
- Data and systems discovery. For each candidate we check where the data lives, which systems offer an API and whether personal data is involved. Our guide to AI and personal data under KVKK summarises the compliance points.
- Pilot and integration. Our AI integration service embeds the chosen use case in the ERP, CRM or email flow you already use. Document-heavy use cases rely on document OCR; knowledge Q&A use cases rely on an enterprise LLM assistant.
- Custom models where they pay off. For forecasting or vision work that depends on your own data, we use custom AI model training to train the model on it.
- A written proposal. We do not sell off-the-shelf packages; after a needs analysis we prepare a written proposal setting out scope, phases and fees.
Next step to choose your first AI pilot
Gather up to ten use cases from your departments and write each one as a single sentence. Then get in touch: we will score the list with you and agree the first pilot and its success threshold in writing. All our AI guides are collected on the Artificial Intelligence topic page.