What an AI quick scan delivers
Most businesses that want to 'do something with AI' start without knowing exactly which problem they are solving. They read about use cases at other companies, hear that competitors are getting involved, and decide they need to start too. Without first checking whether their own processes are ready for it.
An AI quick scan reverses that. I start with your current way of working, not with the technology. I look at how work flows now, where repetition occurs, where decisions are made based on patterns, and where manual steps create the most friction. Based on that picture I deliver a report with 3 to 5 concrete candidate cases.
For each case the report includes: what the process is, why AI is promising or unsuitable here, an estimate of the required effort, and what the expected impact is if you move forward. The report also includes cases where I explicitly advise against it, with the reason. An audit that only gives green checkmarks is not an audit.
My discovery process
A quick scan is not a questionnaire I send you. It is a hands-on observation of how work flows at your company. I use four steps:
- ▸Intake conversation with the team: I talk to the people who do the daily work, not just the director. They know where the friction is.
- ▸Observation of manual steps: I have employees walk through their own work while I look on. Many bottlenecks are not visible in a description, but they are in practice.
- ▸Data flow analysis: I map out where data comes from, where it goes, and how often it is manually retyped or copied.
- ▸Identification of automation candidates: I flag processes with high repetition, low exception rate or a lot of pattern recognition as potentially interesting AI cases.
I always do this custom. No standard checklists. The outcome depends entirely on what I found at your company, not on what I expected to find beforehand.
Cases I typically flag as promising
After multiple audits at diverse companies I see recurring patterns. These are the categories I most often flag as potentially interesting:
- ▸Document extraction: invoices, quotes or contracts where the same fields are manually transferred to a system every time. AI can take over that extraction with high accuracy when the document structure is stable.
- ▸Data classification: incoming messages, tickets or reports that someone manually categorizes before forwarding. When enough historical material is available, this is a strong candidate.
- ▸Content routing: emails or form submissions that need to be sent to the right department or person based on content. Easy to automate when the categories are clear.
- ▸Repetitive decisions with patterns: approvals, prioritizations or assessments that are always based on the same factors. Not to replace the human decision, but to prepare it.
- ▸Summarization of long sources: meeting notes, customer conversations or technical documentation that are manually summarized. AI can deliver that as a first draft that someone briefly reviews.
These are categories, not guarantees. Whether a specific case is promising depends on the quality of your data, the consistency of the process and the team's willingness to work with an AI component.
Cases I typically advise against
Just as important as the opportunities is knowing what is better not to tackle with AI. I advise against it when:
- ▸Volume is too low: a task that occurs twice a month does not generate enough return to justify it. A good automation needs scale advantage.
- ▸The process is deterministic: if every step can be described exactly in rules without exceptions, a rule-based solution is faster, cheaper and more reliable than AI.
- ▸Brand sensitivity is too high: for text or images that go directly outward and are strongly tied to your voice or style, AI generation without thorough human review is too risky.
- ▸The data is not there: AI applications based on pattern recognition need historical material. When that is missing or fragmented, the foundation is absent.
- ▸The exception is the rule: when the process has so many exceptions that AI output always needs correction, you are building extra work instead of less.
I mention this explicitly in the report. An honest 'this makes no sense' is worth more than an enthusiastic 'let's build this' that stalls after six months.
Example output: cases in a priority matrix
A report from me does not look like a presentation full of bullet points. It is a working document with the following columns per identified case:
This is a simplified example. The actual matrix contains more context per row and an explanation of why something scores high or low. The goal is that after reading the report you can make a considered choice, not that you follow the recommendations blindly.
-- Example case
A logistics company: 3 cases identified, 1 built, 2 advised against
A transport company asked me to look at where AI could contribute to their administrative process. They had the idea that processing of freight documents could be automated, and were also thinking about a chatbot for customer communication.
After the discovery three cases emerged. The freight document processing turned out to be a good candidate: high volume, stable document format, the same 8 fields every time. I built that as an extraction pipeline based on a document model. I advised against the chatbot: customer questions varied too much, and the reputational damage from a wrong answer was too great for an unsupervised system. For the third case, routing internal reports, I advised replacing it with simple categorization in their ticket system, without AI.
The report delivered a clear decision-making document. The company chose the extraction pipeline and deliberately left the other two cases aside. That was the right choice based on the facts, not based on enthusiasm.
What I do not do
A quick scan is a discovery, not a sales pitch for an implementation project. I make that distinction deliberately:
- ▸I do not use the audit as a funnel toward the largest possible follow-up project. The report is the product. Whether you build something afterward, with me or someone else, is your choice.
- ▸I do not write generic AI strategy reports. Everything in the report is based on what I observed at your company, not on sector-wide trends.
- ▸I do not make recommendations for technology I have not analyzed in your context. No 'you could buy ChatGPT Plus' without knowing what you would do with it.
- ▸I make no promises about implementation results. The audit gives an estimate, not a guarantee. The reality during the build may differ from what the discovery showed.
- ▸I do not pretend there are always AI cases. If I conclude after the discovery that there are no good candidates, that is in the report.
Cost of the quick scan
The price of an AI quick scan depends on the size of the company, the number of processes I analyze and the depth of the report. I always work on a fixed price per audit, not an hourly rate.
On request
Get in touch for a no-obligation intake. I will then give an indication of scope and cost based on your situation.