Tonkean: Making Enterprise Procurement work with ProcurementWorks, Part 1

Tonkean was founded in 2015 to transform the enterprise back office. Tonkean leverages smart technology to bring people, process, and technology together in a manner that revolutionizes how businesses operate, allowing people to focus on high value work that gets results, and not redundant data processing, unnecessary application usage (which requires unnecessary training and unnecessary time), or unnecessary emails. One of the big problems Tonkean saw with traditional enterprise systems is that anyone who didn’t need to use the system daily was resistant to learning yet another system they saw as difficult or cumbersome (which applied to any system that didn’t use their terminology), adoption was a major problem, and employees would constantly look for ways to circumvent the system. Tonkean’s goal was to solve the adoption problem by providing users a superior intake experience, that could be as simple as a standard form-based or natural language interface like they’d find on the web, that didn’t require any training and that helped these employees make their requests through official channels instead of sneaking through back doors and dark hallways.

After a few custom projects, they found an initial niche in the Legal department and created Tonkean LegalWorks to help Legal Teams with legal mail routing, legal matter intake, matter lifecycle management, legal discipline and category classification, conflict waiver processing, law firm onboarding, contract routing and review, and even legal risk monitoring. It brought together the systems used by Legal (email, word processors, specialized Legal Billing Management solution, etc.), any risk and compliance applications they use to ensure their lawyers and firms dot all the ies and cross all the tees they need to take on every case and practice in every state they are taking legal matters in, and any other enterprise applications the team used to work and communicate internally (Slack, Teams, etc.).

And while we’re not here to discuss LegalWorks, it is through the development of LegalWorks that they learned how to bridge the gap between people, process, and technology in a in a way that empowered their clients to spend more time on strategic (legal) work instead of redundant data entry and system usage, get more value out of the tools they already purchased, and be more productive and satisfied with their technology. They learned how to enable a department to use the tools they have in ways that went beyond the original use cases, and learned they could do more and set out to identify where the biggest needs were and where they could do more. And once they found Procurement, and realized that Procurement had a lot of the same challenges as Legal, but considerably more amplified (with more systems, more complexity, and higher stakes), they knew they had found an area where they could provide their enterprise clients with the most value (and especially those that were using the major S2P suites but getting low utilization rates due to lack of intake support and a lack of integration with other internal systems).

When investigating Procurement in their enterprise customers, they found that while the major suites were reasonably suited for, and well used by, the Procurement team in strategic projects, they weren’t used much in tactical purchasing, especially in tail spend, as most of the organizational users found the system too complicated and bypassed it whenever possible (as the P2P tool lives on the long tail of enterprise applications of choice for the average enterprise employee).

So, as with some of the new breed of vendors who started specifically with the goal of Procurement intake and/or orchestration, one of their first goals was to help their Enterprise customers get more value out of their big S2P suites (and Ariba and Coupa in particular; for example, they have Intake Orchestration for Ariba and the Coupa Intake Experience to help the organization route all indirect spend, no matter how far down the tail, through Ariba or Coupa). While that’s where they are still focussed (given their current Enterprise customer base), they’ve expanded their ProcurementWorks to be a full Procurement lifecycle orchestration solution, from intake to resolution, regardless of what solutions the customers have or don’t have, what enterprise applications the teams use to communicate, what external catalogs or data sources they need to integrate with, and what policies and procedures need to be followed. In this way, ProcurementWorks is a system-agnostic solution that wraps around the customer’s existing process and applications to orchestrate and better coordinate that process.

However, one major difference is that, to Tonkean, full orchestration means creating a solution that solves all of the Procurement related problems an organization’s employees have, not just Procurement requisitions or catalog buying. That means answering all of their Procurement related questions in addition to taking their product and service requests, guiding them to the right systems if needed, or being the one interface of choice if Tonkean can be that. That means a much smarter intake process that can take any Procurement related natural language request, interface with all of the organizational data sources, and provide an appropriate answer.

For Tonkean, that starts with a smart AI interface, that they call the AI Front Door. The AI Front Door, unlike many other LLM-based products, is not just ChatGPT in a shiny wrapper, but a hybrid solution based on in-house engineering, the client organization’s preferred LLM, and knowledge systems owned by the client. It’s a very sophisticated “chatbot” compared to most offerings on the market, a technical definition would be very extensive (and lose non-PhDs), but we can illustrate much of the uniqueness of the capability with a high level overview and an example or four.

For example, when a user inputs a request, the general approach the system takes is:

  • use their AI to process the question for the type, intent, and goal
    and inform the user if they have no information (or are unable to process it) while simultaneously
    redirecting any unanswerable query to a human expert for review
  • use internal, trusted, knowledge bases to get initial information and potential answers
  • feed the question, processed clarification, and internally retrieved knowledge into the organization’s LLM to provide Natural Language feedback to the user, which could be the answer, or a refinement question if ambiguity existed in the question or potential answers from organizational data sources, which causes (an extension of) this 3-step loop to repeat
  • verify the response is sensible before presenting to the user (and, if not confident, route to a human for feedback for future internal Tonkean model training while informing the user no relevant information can be found)

Thus, if the user asks if there is an agreement with Vendor V:

  • their AI Front Door will process the query and determine that the user is asking if there is a signed contractual agreement with Vendor V that is currently active, and potentially what that agreement is
  • create the appropriate queries for each organizational system that stores contracts and agreements
  • take the responses and construct a carefully engineered prompt for the LLM that will return an answer indicating if there are agreements, and, if so, what they are and where they can be found (possibly including a direct link if the document can be accessed through the Tonkean platform)

If the user asks if she can purchase a license for SaaS app S:

  • their AI Front Door processes the request, determines that the user wants to make a purchase, it falls in the software category, and asks a few clarifying questions about the type and purpose of the product and, if it discovers the organization already has a license for a tool of that type, asks why the other tool won’t do
  • the system takes the responses and prompts the user with a link to launch a purchase request, where the system then pre-populates key fields of the organization’s software license purchase request form based on its learnings from the AI Front Door interaction and data attributes from other relevant systems (such as budget information in the ERP)
  • the system bundles the appropriate information and prompts the LLM to create grammatically correct responses that not only explain the request to the Procurement Buyer, but a Supplier if an RFP is required
  • the draft form is then presented to the user to verify, and one click puts it into the Procurement Request queue (where it can be accessed from the ProcurementWorks My Requests page at any time)

If the user asks for the procurement policy for SWAG for the marketing event she is attending:

  • their AI FrontDoor processes the requests and determines its a policy question
  • it creates the appropriate pattern match, DQL, or index query for each of the organization’s policy document data stores and collects the appropriate responses and documents
  • creates an appropriate prompt for the LLM that appropriately forms the question while asking the LLM to use only the inputs fed to it to create the response
  • ensures the response that comes back has a decent similarity to a subset of the text from the documents and then presents the natural language summary to the user

If the user asks the system for the results of the hockey game he missed working late:

  • the system processes the requests, realizes it doesn’t have that information (unless, of course, the enterprise is a sports news outfit), informs the user it doesn’t have that information and ends that interaction there

In other words, it’s built to be the central information source and jumping-off point for all types of inquiries and tasks a Procurement professional or employee with Procurement needs is likely to have, with the intent of cutting out 90% of unnecessary emails, texts, questions, and requests an augmented intelligence system can answer or guide a user through.

Moving on, the core of the Tonkean Intake Orchestration Platform that their Procurement solutions were built on is a workflow automation platform with extensive built in workflow customization, data integration, and form creation capability. In the platform, the customer can build forms (using a no-code form editor) they need to power any Procurement process (which can be created and modelled using a no-code process editor) they have, and customize them for requesters, buyers, risk & compliance, IT, or any other department as needed. They used this capability as the foundation not only for their Coupa Intake Experience and Intake Orchestration for SAP Ariba (as organizations never replace major investments, but innovative organizations look to improve and expand upon them), but their guided buying experience, supplier onboarding, and tail spend automation (among others).

One key differentiator is that any workflow can be updated at any time, something which is generally not possible in your traditional Procurement Suite such as Coupa, Ariba, and Jaggaer. For example, many of their customers now require an additional AI Review of any platform that uses AI to determine the nature of the AI and any direct and indirect risks in its proposed application to the business from a technical, legal, and brand perspective. For example, if the vendor is using Open Gen AI (such as ChatGPT), there are technical risks in that these platforms have been repeatedly demonstrated to have biased, harmful (and even murderous), hallucinatory, thieving, and sleeper behaviour. There are direct legal risks in that you could be sued (and on the hook) if the AI makes a recommendation that ends up causing personal or business harm, and indirect legal risks if the technology was trained on stolen data or data that contained copyrighted, illegal, or national secret material. There are brand risks if the Open Gen AI product you are using all of a sudden suffers extreme public backlash for its actions (or your software results in a decision that tanks shareholder value or increases environmental harm). However, they have found that most of the suites they work with do not yet have many of these new “standard” compliance checks in their relatively rigid product workflows (and telling their customers to just include it in the InfoSec review), which increases the likelihood a key check will be missed. [Considering the attention that AI is getting and the fact that legal frameworks will need to come soon, not the best idea for a large organization NOT to be assessing AI risks now.] However, with Tonkean, it takes minutes to add a compliance check and ensure it gets done by the right people before a decision is made on any Software purchase or use.

In our next article, we will dive deep into the major components of the Tonkean ProcurementWorks offering.