How to create an in-house bot
– Gods! My wife warned me that at 8:00 on Monday I would need a scan of my certificate from work!
An unpleasant series of thoughts flashed before the young man’s eyes: he clearly saw his wife’s disappointed face, her attempts not to show that she was terribly upset, her eyes “wet.” He himself, realizing that a joint visit to the bank to obtain a mortgage loan was postponed indefinitely, chilled inside. The thought began to creep in that his forgetfulness might cost him a peaceful family life.
– Something must be done! I will try to find acquaintances, I would give anything to get someone who can make such a reference to the office on a day off!
After an hour of frantically calling his colleagues, he still got through to the one who answered the right question:
– Why do you need a personnel officer in the office? Do you just need a certificate from the place of work? Well, write to the bot, do it for a few minutes.
Contents
Given:
-
Holding National Media Group is Russia’s largest private media holding that manages a diversified portfolio of assets in the field of production, distribution and monetization of high-quality media content in all consumption environments. In short, there are many different specifics.
-
Target audience users = Employees – different levels of positions, areas of activity, location and format of work (remote, hybrid, in the office). Number – ~ 4000 people.
-
A single window — a service that includes all services that can be used by employees of the holding (personnel, administrative and economic, legal, IT, etc.),
-
the result is provided by the performer (person),
-
regulated SLA regarding the performance of each service,
-
On average, the SLA of one service is 24 working hours (we understand that the SLA and the actual execution can differ, both for worse and for better – here we depend on the download of the artist at the other end of the wire).
Question: what needs to be done so that the employee receives an answer to his question as quickly and conveniently as possible, without being tied to the workplace, and without burdening the performers even more?
Answer: you need to create a 24/7 self-service service
Work progress
1. Determination of the self-service service format
For this, what is in the data was enough for us: for all categories of employees, we need to make a service that they could use without corporate access, at any time of the day or night. Reduce the response time as much as possible and relieve the executors. The opportunity to be creative – we work in a media holding after all!
That’s why we settled on a robotic virtual assistant to get answers to questions and chat outside of business hours.
2. Formation of the team
It was quite easy to assemble a team – they offered the best. They, in turn, supported the initiative. We did not hire new people for the project – our colleagues learned the best practices of implementing and using robots themselves.
The current team consists of a project manager, a business analyst, a systems analyst and two developers. The team’s task is to develop new services and improve + support already implemented ones.
3. Preparation for development
First, we analyzed the most popular requests in the Single Window and determined those that we can fully automate and which we can partially automate. Formulated business goals and effects of robotization. They drew up a backlog, thereby defining the boundaries of the project for the near future
Analyzed critical aspects of bots – from performance and functionality to user interface.
Next, we began a step-by-step description of the AS-IS and TO-BE processes. We note that for some services during the formation of TO-BE, we carried out a complete reengineering (we also improved the accounting systems). In addition to the standard description of the BPMN notation, target branches of dialogs were drawn for each service. Since the portfolio of the holding is diversified, the branches of dialogues take into account the specifics of each asset (in general, we stopped hoping to fit the branches on at least 2 sheets of A3).
We determined the location of the bot – placed the pilot services on the corporate portal, and then on Telegram (don’t panic, this has been agreed with the cyber security of the holding). By placing the bot in Telegram, they confirmed their hypothesis – people willingly communicate with the bot there: the numbers show an 8-fold increase in requests in Telegram compared to the corporate portal.
4. Development
For the chatbot, we created a separate working configuration taking into account 1C. They named it 1C: Connector. The configuration uses the typical functionality of building block diagrams, which has been modernized for our purposes.
Flowcharts allow you to duplicate the TO-BE business processes that the business analyst has displayed in Visio and transfer them to 1C. Each block of the 1C block diagram is filled with code that contains a certain logic and determines which database the request will go to or which answer must be returned to the user, etc.
We wrote the code for the bot branches. Here they resorted to third-party help from the vendor’s specialists for the code review. The chatbot has a mix of JS with the vendor’s self-written language, so at first we resorted to checking the structure and possible errors in the code.
The following scheme of work came out:
-
Processes are performed mainly in 1C systems.
-
The bot, having received a request from the user, processes it, then turns to 1C: Connector, which provides end-to-end communication for all databases.
-
Depending on the type of 1C request: The connector understands which of the bases you need to contact.
-
Then from 1C: Connector, the request is sent to the necessary system.
Now we have five information systems connected to 1C: Connector, three internal and two external (Huntflow for automating recruitment and NVIS for legally significant electronic document flow).
At the stage of developing the architecture, we analyzed our business processes in sufficient detail and took a lot into account. But if you suddenly need to make changes, it is implemented in a few clicks without updating the configuration.
If we talk about information security, working in Telegram with access to corporate resources is already a precedent for cyber security. We implemented two important solutions to reduce threats of unauthorized access and information leakage:
-
2FA – two-factor authentication is mandatory for the robot. The employee’s current phone number must be registered in internal databases and must match the number from which the request was received. Otherwise, the bot simply will not perform such requests.
-
We do not send any confidential information to Telegram, if it is necessary. It is sent to the user’s corporate email.
Together
Currently, more than 20 services are implemented. And here is a description of how some of them work:
Service |
Brief description of the process |
Time of receiving the result by the user |
Remaining vacation days |
The bot displays information about the remaining days of the user |
5 p |
Calculation sheet |
The user specifies the period during which he wants to receive a settlement letter. The bot sends an invoice to the company. user mail in PDF format |
10 s |
Reference 2-PTFO |
The user enters the required data. Depending on the selected help format, – if the original – the bot generates a 2-PTFO certificate and creates an Application in the Single Window for the performer; the performer prints and signs the certificate and informs the user that the certificate is ready. |
Within 1 day |
– if EP – the bot forms a certificate and sends it to 1C:DO for signing, and then to the corp. user mail |
Within 1 day |
|
– if the scan is done, the bot forms a reference and sends the scan to e-mail. post office user mail |
7 p |
|
Order a pass |
The user enters data about the guest, car, etc., and the bot sends the data to e-mail. mail security/office reception |
20 p |
Install 1C:DO on the mobile device |
The bot displays instructions for connecting 1C:DO to mobile. device individually for each user, substituting the user’s login, server address, etc. with the possibility of copying to the program settings. |
5 p |
Office addresses |
The user chooses the name of the office. The bot gives the full address (with the possibility of going to Maps/Navigator) + offers its detailed text description |
5 p |
And this is not the limit. Not even from the point of view of the scope of services, but also functionality. We constantly monitor requests to the chatbot and, together with users, determine new milestones in the development of the functionality.
Few numbers
-
28 services, the list is constantly updated
-
More than 5000 requests from users per year
-
18 services are fully automated – no executors are involved in the solution; answer “here and now”
-
10 services are semi-automated – we involve performers, but all the necessary data is already collected by the bot (we exclude time for additional clarifications)
-
Requesting and receiving the result of fully automated services takes the user an average of 26 seconds
-
The labor costs of dispatchers for routing applications in the Single Window are completely excluded (semi-automated services involve the transfer of data to the Single Window) – the bot itself classifies such a request and sends it to the “correct” performers.
-
Having reduced the labor costs of the performers, they got savings on labor costs. Let us give the example of the 2-PIT Tax References:
Service |
It was |
It became |
Comment |
||
Document format |
Labor intensity of execution, min. |
Document format |
Labor intensity of execution, min. |
||
Reference 2-PTFO |
Original |
21 |
Original |
13 |
In addition to the implementation of the service in the chatbot and setting up the integration with accounting systems, the process of generating the 2-PTFO certificate was reorganized internally. |
Document signed by EP |
8 |
||||
Document scan |
0 |
Thus, saving the accountant’s time is from 8 to 21 minutes per reference. Imagine the volume of such references per year.
What’s next
Despite the main focus of the chatbot (the provision of HR services), in the near future we plan to add the possibility of ordering IT services and expand AHO services. Our task is stable and correct work of our virtual assistant.
The robot already knows how to joke and generally entertain the user. He was recently taught how to play Cities. Now you can simply have a dialogue with the bot without sending it an exact request – it will be able to find keywords and offer the necessary service itself.
The most important thing is that NMGhosha (that’s how we called him) copes with his main duties perfectly and fulfills the set business goals, reducing the time for processing requests, taking routine work upon himself. We actively collect feedback from users, which is almost always positive, and most importantly, ideas (the bot has a service of the same name for this).