Tower Engineering

Tower tendering process from 2 weeks to 2 minutes with AI/ML.

Using our PriceBook™ application it is possible to generate detailed tower designs including all accessories and transportation with prices. The process takes about 2 minutes, compared to traditional approaches ranging from days to weeks. #MNO #towerco #telecom #towerdesign #pricebook

Balázs Kisfali
Balázs Kisfali
Dec 15, 2020
min read
Tower tendering process from 2 weeks to 2 minutes with AI/ML.

If you are a towerco or MNO deploying passive assets,  you might search out for tower suppliers time to time to get quotations for tower or masts supplies. But you might not know what process it takes on the suppliers' side to prepare a quotation including all details like designs, accessories, foundation, transportation and so on. This process  can take some days or even weeks. In this article I will discuss how we solved this problem and speed up the tendering process significantly, so eventually support towercos with automated processes and instant pricing functions. (asd...)

The PriceBook™ is an online application which serves as a tower configurator to speed up tender processes.

The challenge

The idea was born when I had been working as tower designer and making tower designs in a daily basis, and completing 200+ tower optimisations a month, at least. This huge need was originated from different factors, like a well-known tower supplier brand with highly optimised tower designs and the many running tenders in Sub-Sahara region in Africa. Our customers demanded quotations for their tower tenders asap and we had to keep up. A typical tender included maybe 8 different tower designs with 3 typical foundations sets, meaning 24 foundation designs. If you are in a competitive market you cannot just pull-out a tower design from the shelf and send it over to your client. Well, you can do it, but then you likely loose the tower bid. So, designing and optimising 8 lattice towers + 24 foundations takes couple of days if you provide quality designs and reports. Then usually supply chain will run the products through their steel suppliers to get updated pricing and reach out other vendors for accessories. Someone else can work on the logistic challenges and figure out the best route and prices.  Someone also needs to collect all accessories, like AWL, fences, grounding, painting, etc. to make a complete package.  Finally, an upper management member needs to give the green light on the pricing and eventually deliver the tender package to the client. At best this takes about 3-4 days but if you have suppliers in other continents, slow internal corporate mechanism, outsourced or overseas engineering you can end up to spend two weeks from the moment you get the RFQ (request for quotation) until you hit the send button in your email to the customer attached your quotation.

Beside all of these factors we knew that there is a need for a better solution which can speed up and simplify this process. I took the challenge and started to work on the solution at nights, and the goal was to make this happen less than 2 minutes. It was clear that the solution should be web-based, access only for authenticated users and intuitive enough for new comers to click through the buttons within the 2 minutes even in a Sunday morning.



How do we know what type of towers our clients will require? This we cannot know, but we can predict it quite well. I have analyzed more than 4.000 tower designs from historical data of the company and I've got some patterns. Then making statistical models, probability functions I found some well fitting distribution functions to get good-enough predictions about what are the most demanded tower parameters:

  • tower type
  • tower height
  • basic wind speed
  • antenna EPA area
  • topography category
  • exposure category
  • deflection criteria
Tower key plans based on user inputs.

Then we only needed a matrix to display all combinations and come up with a plan to design hundreds of towers which will fit to the input choices by the users. The PriceBook™, or PB for short, was designed to cover the most frequent tower tenders and it passed the exam many times. The prediction model was fairly good and the input selections covered 80% of a typical tower design request.

Since artificial intelligence has evolved in recent years, more tools are available in machine learning (ML). We have recently replaced our statistical model with ML models to get better and more flexible predictions. This helped a lot to create different combinations as design inputs to support more efficient use of tower designs. The ML model is un-supervised and it learns with every new tender request. Then we have the opportunity to add new designs or remove unused ones, and give the most optimal solution for the user. The ML-based prediction model, the hearth of the system, helps to reduce the need to create thousands of tower designs to only a hundred to cover the 80% of the market needs.

The PriceBook now is running in multi-vendor mode as well, and finds the most optimal designs among different tower suppliers. This function is price independent as the objective is the lowest tower weight. However, the user can override this default as not always the lightest is the cheapest.


The pricing of telecom structures and equipments are very dynamic, similar to other goods. However, some rules need to be applied and regular updates are needed. This is not different as getting updated prices otherwise. The only difference is that someone takes the responsibility to establish and update the database in every quarter. In this way we can assure that the user/customer will get the latest updated prices. There is no need to send many emails, back-and-fourth, the user always get an updated info.  Thanks to our RBAC (role-based-access system) there is the possibility to give access to vendors who makes this price updates on an agreed schedule. Then an AWL (aviation light) supplier, or steel supplier will be responsible to update the DB with their latest prices.

Product database extract.


Similarly to the pricing of goods, transportation costs are even more dynamic. For this reason we suggest to our clients to have update on transportation costs in monthly basis. Shipping companies can also login and update their prices with their promised timeline. The PriceBook will display the predicted lead-time of all structures and accessories for most used ports around the globe. The database consists of hundreds of ports of destinations and several departure ports.

One record of port. @50%


Based on the tower designs, foundations are also dimensioned and volumes will be presented for each tower types. These are the volume of concrete and excavation, and reinforcement weight. The program also provides the number of anchor bolts necessary to deploy the tower or monopole on a single footing.

Results for foundation. @80%

Database functionalities

The main datasource storage is a SQL database with highly customisable and powerful search functions. The user can gain an insight about frequently used tower types in tenders or use tower specifications as templates.

The multi-vendor version can support complex tower tenders and basically serve as collaboration platform between the participants. Then the purchasing task is then simplified and the first "filter" and benchmarking is done within the app. Then the promising candidates can supply more detailed information and drawings for their products.

Fences and other accessories

With the program the user can also customise the accessories like:

  • fall arrest system type
  • antenna brackets
  • bolts + nuts
  • fence around the site
  • tower paint
  • grounding system
Accessories steup. @60%

Based on several types of fence designs, a decent amount of combination can be generated by the program. Finally, the fance and other accessories can be saved along with the tower main parameters.

One typical fence layout from the PriceBook™. @60%

With these features the tower supply can turn into a full delivery plan for the passive parts of a telecom site. Furthermore, our database structures would allow us to integrate even active components, like GSM antennas or MW dishes with cables.

Limitations and post-processing

Even by using ML models to predict the most commonly used parameters and doing other maths to reduce possible combinations, still lots of details has to be agreed before hand to be hard-coded in the process. Otherwise the number of combinations would been in the range of tens of millions. For this reason we have specified some factors which should be project specific and some of them listed below.

  • Design code of steel (Eurocode, TIA, etc.)
  • Design code of concrete (ACI-318, EC2, etc.)
  • Foundation code (EC7, etc.)
  • Feeders' width
  • Number of standard platforms
  • Access ladder type
  • Basic exposure category (like "C" or "urban")
  • Basic topography/orography category (1,2, etc.)
  • Antenna load distribution on top of the tower (i.e top 10m)

Our prediction model can also be fine-tuned and make automatic reduction based on number of inputs decided by the user. However, it is recommended to limit the possible scenarios and input types, which is usually doesn't require huge compromises. The generated data and designs then can be further processed by exporting them in XLS, CSV or other formats. For integration purposes, it is possible to utilise our API endpoints, so your IT team can fetch data from the PriceBook system.

If you want to speed up your tower purchasing process, or want to be sure that your suppliers are tendering with the same content and same specification, we recommend to use a central tendering tool to be able to be on the same page for all stakeholders. If you want to implement our PriceBook™ solution for your needs, please contact us today and we can provide you a customised platform based on your requirements.

More articles

Go to blog