The Kenya Revenue Authority (KRA) is planning to use artificial intelligence (AI) to screen shipments entering the country, in a move that could uncover a web of tax evaders.
Already, the taxman says it has made “some serious interceptions” based on AI cargo analysis in a week of trial use, hinting at the scale of interceptions it could unravel once fully deployed.
The KRA began testing various aspects of the AI to be used in its operations three months ago, but deployed it as an experiment to run alongside its system a week ago. A Chinese firm, Nuctech Hong Kong Company Limited, is behind the AI project currently being tested.
“We have been training it with our KRA images. It’s now three months old but we have already made some very serious interceptions because it’s working real-time,” said Albert Atambo, Chief Manager, KRA Scanner Management Unit.
The official said KRA plans to introduce AI for image analysis and other cargo screening by June next year to boost efficiency at the port and borders, describing it as a major game changer.
KRA has scanners at various border points, airports and the Mombasa port that scrutinise cargo entering the country using x-ray images taken while consignments are in containers or in traveller’s luggage.
The screening helps KRA detect concealment, misdeclaration of goods and other tactics used by tax evaders or those attempting to bring in dangerous items.
While the authority has been relying on its employees located at three centres in Nairobi and Mombasa to analyse images from the scanners, it notes that while human analysts take about two minutes to analyse a consignment, AI takes one second and has proven to be more accurate.
“Our vision is not just to have an AI for analysing an image, we want an AI that is all-round. It should be able to look at a consignment and provide the risk profile. We want it to give us details such as whether the importer has mis-declared before, the value of goods based on what the specific goods are and other details,” Mr Atambo said.
Over the past three months, the KRA has been feeding the AI with data relevant to customs operations, which, once fully deployed, will help it analyse consignments by assessing them based on the details.
For instance, KRA projects that the AI to be able to analyse a consignment of kitchenware and provide its assessment of their value based on details of previous consignments of similar goods in the database.
“For image analysis, we have managed to demonstrate and we are briefing top managers to demonstrate to them what it can do,” Mr Atambo said.
The KRA official said, however, that news of the plan to introduce AI into its processes has rattled several players suspected of smuggling and abusing existing loopholes to evade taxes.
Some of the players have reportedly threatened to block the process in court because KRA has not conducted public participation.
“We made some interceptions and word has gone out that we are now going AI and that means that if you had your people who have been helping you in any way you will not be helped by AI. We expect a lot of lobbying against the move but it will happen,” Mr Atambo said.
“In the next financial year, I am 100 percent sure that KRA is going to procure AI for analysis. We will make sure we are compliant with the law to prevent someone from blocking us in the court,” he said.
Deploying AI will, however, be a costly venture for the authority since a single graphic processing unit (GPU), which is required to analyse images from the scanners, costs about Sh100 million.
The Chinese firm deployed two GPUs at the KRA premises for use during the proof of concept (POC) period, which ends in July, but should the KRA decide to go all in on AI, it will have to buy the equipment.
Mr Atambo said KRA has completed the first phase of the POC and that the second phase would involve training the AI on aspects of risk management, valuation capabilities, and other functions relevant to customs operations.
According to the taxman, the AI has proved to have a high level of accuracy in identifying homogeneous cargo such as rice and wheat, which have been quite complicated for human analysts.
“It’s going to take the bulk of what is obvious, do it, and give us a final result as we concentrate on a small fraction where we can do a very thorough job and make sure we have the maximum revenue from there. It will be a very serious game changer,” Mr Atambo said.
During the first week of testing it alongside normal operations, KRA received parallel reports from human analysts and AI analysis on consignments and compared results.
