News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
Analysis of the present situation of artificial intelligence
Analysis of the present situation of artificial intelligence
The discussion on the commercial application of artificial intelligence (AI) is in full swing. From the 16 years of star fire and 17 years of prairie prairie, this year can be said to be all pervasive. Now, if there is no AI element on the occasion, the organizers are embarrassed to invite guests to attend.
One day, one day, someone said a joke, said the current artificial intelligence application, see only "artificial", no "intelligence", how many "artificial" has how many "intelligence". It was a bit of a joke, but all of a sudden it was really a good idea, and it was inadvertently breaking the current status of AI, that is, the voice is very high, but the landing is very few.
So on this topic, I will continue to chat with tea teacher, and this time mainly discusses how to generate "intelligence" from "data". What I am more concerned about is how to apply data to business analysis and decision making, and how to generate intelligent decision support from a large number of "data". Next is some of our practice and thinking.
640? Wx_fmt=png
"The essence of the data" by the author: Car sense
Big data expert, former vice president of Alibaba group and expert partner of Sequoia Capital, is known as "the first person to think about data in China".
Take the Palantir of Silicon Valley as an example
640? Wx_fmt=png
In the book "the nature of data", pin Yu mentioned the example of Palantir. As we all know, Palantir is a star company in the global AI field. It is considered to have solved many of the most complex social and business problems with the algorithm. But after communicating with Palantir, it is actually in their customer project that a lot of work is doing the integration of data ETL and data, and the AI based algorithm application, association analysis, is just the tip of the iceberg that floats out of the water.
640? Wx_fmt=jpeg
(Palantir example scene pictures originate from the network)
Two cases of local practice
640? Wx_fmt=png
At home, we also encounter many similar scenes. Today we share some examples.
Take an industry leading chain retail brand as an example, this customer has 1000 stores in the whole country, mainly has the timeliness fresh product. For the first time, we have discussed that the cut in scene is a store sales forecast and an intelligent order based on this, which is a typical AI and business application scenario. Cut from this scenario, because this is a significant pain point on the side of business, and the senior leaders attach great importance to it.
But after formal contact, we and our customer IT team quickly realized that the first and second floors ("data base") had not yet been covered, and it was unrealistic to go directly to the third floor of the horse ("the" AI application ").
Specifically, to predict store sales, we need to model the historical data and sales data of store promotion. Stores will promote different products and recommend different products according to seasonal factors. The promotion and recommendation of different products will undoubtedly affect the sales of single products and store sales. Furthermore, some stores will adjust sales according to some special reasons such as extreme weather or surrounding activities. These promotional data are crucial for discovering the relevance of product sales, and are key variables for building feature engineering.
We worked closely with our IT team to systematically sort out the relevant sales data and found that historical data were missing. The algorithm team spends a lot of time building and finding suitable model eigenvalues, but because of the missing key data, the effect of the model is very slow.
Further, why are these key promotional data not well preserved before? A key reason is that the application of BI is not yet in place.
The customer's previous decision to make a decision is to look at the report data in the POS and ERP systems, as well as to add a number of summary classes in OA or to fill in the data. And the report data in the business system is essentially different from the real BI application.
Various applications of BI help enterprises to get through all core business links, forcing enterprises to build a solid data base. For example, through the linkage analysis in BI, the overall correlation of a single product promotion and sales is analyzed, and then the BI drilling analysis is further carried out, and the specific store level is traced, and the associated performance of a single store is found.
640? Wx_fmt=gif
The analysis is a very classic BI analysis scene, and if you have these scenes, you can find it for the first time. In fact, some of the analysis is not going to go on, because the corresponding original data did not fall, such as the promotional data mentioned earlier. These findings will further optimize the IT system and improve the corresponding business processes, so as to replenish data integrity and consistency in time.
And this layer of logic is also surprisingly consistent among our 500 top customers.
This is a China business of FMCG fast moving consumer goods giant. Our goal is to increase the accuracy of sales forecasts by 5 percentage points.
As far as most domestic customers are concerned, this global customer has a more in-depth BI input and application basis, with relatively complete data precipitation, so it is easy to start the application of AI. But when we get to the bottom of the last 1%, we encounter many "landmines".
It is interesting that these "landmines" are not directly related to the AI algorithm, but they are basically inconsistent data. More advanced
One day, one day, someone said a joke, said the current artificial intelligence application, see only "artificial", no "intelligence", how many "artificial" has how many "intelligence". It was a bit of a joke, but all of a sudden it was really a good idea, and it was inadvertently breaking the current status of AI, that is, the voice is very high, but the landing is very few.
So on this topic, I will continue to chat with tea teacher, and this time mainly discusses how to generate "intelligence" from "data". What I am more concerned about is how to apply data to business analysis and decision making, and how to generate intelligent decision support from a large number of "data". Next is some of our practice and thinking.
640? Wx_fmt=png
"The essence of the data" by the author: Car sense
Big data expert, former vice president of Alibaba group and expert partner of Sequoia Capital, is known as "the first person to think about data in China".
Take the Palantir of Silicon Valley as an example
640? Wx_fmt=png
In the book "the nature of data", pin Yu mentioned the example of Palantir. As we all know, Palantir is a star company in the global AI field. It is considered to have solved many of the most complex social and business problems with the algorithm. But after communicating with Palantir, it is actually in their customer project that a lot of work is doing the integration of data ETL and data, and the AI based algorithm application, association analysis, is just the tip of the iceberg that floats out of the water.
640? Wx_fmt=jpeg
(Palantir example scene pictures originate from the network)
Two cases of local practice
640? Wx_fmt=png
At home, we also encounter many similar scenes. Today we share some examples.
Take an industry leading chain retail brand as an example, this customer has 1000 stores in the whole country, mainly has the timeliness fresh product. For the first time, we have discussed that the cut in scene is a store sales forecast and an intelligent order based on this, which is a typical AI and business application scenario. Cut from this scenario, because this is a significant pain point on the side of business, and the senior leaders attach great importance to it.
But after formal contact, we and our customer IT team quickly realized that the first and second floors ("data base") had not yet been covered, and it was unrealistic to go directly to the third floor of the horse ("the" AI application ").
Specifically, to predict store sales, we need to model the historical data and sales data of store promotion. Stores will promote different products and recommend different products according to seasonal factors. The promotion and recommendation of different products will undoubtedly affect the sales of single products and store sales. Furthermore, some stores will adjust sales according to some special reasons such as extreme weather or surrounding activities. These promotional data are crucial for discovering the relevance of product sales, and are key variables for building feature engineering.
We worked closely with our IT team to systematically sort out the relevant sales data and found that historical data were missing. The algorithm team spends a lot of time building and finding suitable model eigenvalues, but because of the missing key data, the effect of the model is very slow.
Further, why are these key promotional data not well preserved before? A key reason is that the application of BI is not yet in place.
The customer's previous decision to make a decision is to look at the report data in the POS and ERP systems, as well as to add a number of summary classes in OA or to fill in the data. And the report data in the business system is essentially different from the real BI application.
Various applications of BI help enterprises to get through all core business links, forcing enterprises to build a solid data base. For example, through the linkage analysis in BI, the overall correlation of a single product promotion and sales is analyzed, and then the BI drilling analysis is further carried out, and the specific store level is traced, and the associated performance of a single store is found.
640? Wx_fmt=gif
The analysis is a very classic BI analysis scene, and if you have these scenes, you can find it for the first time. In fact, some of the analysis is not going to go on, because the corresponding original data did not fall, such as the promotional data mentioned earlier. These findings will further optimize the IT system and improve the corresponding business processes, so as to replenish data integrity and consistency in time.
And this layer of logic is also surprisingly consistent among our 500 top customers.
This is a China business of FMCG fast moving consumer goods giant. Our goal is to increase the accuracy of sales forecasts by 5 percentage points.
As far as most domestic customers are concerned, this global customer has a more in-depth BI input and application basis, with relatively complete data precipitation, so it is easy to start the application of AI. But when we get to the bottom of the last 1%, we encounter many "landmines".
It is interesting that these "landmines" are not directly related to the AI algorithm, but they are basically inconsistent data. More advanced