Market analysis:
An accompanied, monitored and paid beta programme was set up for a range of different restaurant types. 17 interviews were carried out to identify which aspects of the meal planning process where most inefficient in restaurants. The findings support the significance of automated allergen detection, a key feature of the software Menutech.
The most frequently quoted reasons justifying the need for automation in meal planning and the preparation of food menus were the following: the shortage of qualified staff, non-aligned working hours between receptionists and chefs, a lack of understanding on how allergen declaration should be done in practice, the fear of potentially omitting an allergen contained in a dish, the repetitive nature of the work, and the difficulty in identifying what dish arrangement is suitable for clients with a combination of intolerances.
88% of the interviewees recognised the need for technology to improve the efficiency and quality of meal planning in restaurants. Amongst this group, the most frequently quoted impediments to the adoption of technology were the lack of user friendliness, high costs involved and unawareness about appropriate solutions.
Innovation strategy:
Two key technologies were identified to complement the innovation strategy of Menutech: i) artificial intelligence (AI) and ii) optical character recognition (OCR). They address directly the identified issues of time efficiency by minimising the process time and ease of use by automating the initial data input.
i) AI algorithms were identified as a means to improve the automated allergen detection by maximising the value of the existing curated allergen data of Menutech. Multiple prototypes of AI algorithms using neural networks trained with the Menutech database of dishes and allergens were built. Preliminary results were positive and promising, increasing the flexibility of the allergen detection. The use of the Keras AI framework together with SciKit Learn to complement the existing syntax analysis algorithms lead to a 20% increase of positively matched allergens in a random sample of 100 dishes (mixed modern European cuisines). AI algorithms performed stronger for long dish names (6+ words), whereas the existing syntax analysis algorithm performed stronger on short dish names (>6 words). A combination of both algorithms thus provides the greatest accuracy and is able to cover a larger range of dishes.
ii) Using cloud-based OCR computer vision services, information on existing food menus can be scanned and directly imported to Menutech. This is achieved by taking pictures of the existing menu with regular smartphone cameras. On Menutech, allergens can be automatically identified and translations can be proposed for the existing menu. The user can therefore digitise any existing menu within minutes, easing the transition from manual tools, word processors and external agencies. Following this digitisation, the menu can be modified and enhanced using Menutech’s meal planning features. Menutech thereby enables restaurants to update dishes and prices on their restaurant menus with ease, digitally share and synchronise their menus with third party providers, maintain an allergen-friendly and diverse food offering as well as fully respect European and national declaration requirements.
Economic strategy
Resulting from an analysis of internal resources, the need for the hiring of an additional data scientist or artificial intelligence engineer was ascertained. The addition of human resources is necessary to carry out appropriate efficacy analyses and run split tests with different sets of algorithms. As part of this project, an appropriate candidate was identified and hired, due to start on 4 February 2019.
IPR protection strategy
The combined syntax-based and neural network based algorithms for allergen detection in freely typed names and descriptions of dishes was further identified as a potential key technology developed entirely in-house and potentially qualifying for IPR protection.
