How do you get Max Value using GENAI in Procurement Cycle at Zero Marginal Cost

See if your procurement journey is similar and how to get "MAX AI VALUE and SUSTAINABLE procurement at Zero Marginal cost in one go. Got a similar use case , read on

A few months back one of a Fortune 500 client , asked me to review their proposal to " REVAMP their Procurement ". Here's what they wanted

1. their procurement on Ariba and dashboards should be digitilized .

2. They did that on Ariba,even after that their overall turnaround cycle only reduced by 2 -3 days , They wondered how to take it to the next level with GENAI. They are saving on procurement but now MAKE it More Intuitive on the GO

3. Ensure that procurement was sustainable so that it fed into compliance and NETZERO COP28 agenda.

4. Save money , optimize vendor supplier agreements and ensure best products procurement contract life cycle management but keeping in mind human rights and sustainable agenda

5. Give me a command center baseline , where procurement agents can get their answers and best case contract design by a single Prompt .

6. Help me to "Match contract terms" with payment terms and call out deviations ' before they are paid.

7. Provide a market view industry view of best supplier vendor contract agreement terms and how we can apply them.

8. Intelligent chatbots : not virtual assistants that have reloaded responses :for example : Example

Prompt : User: “What’s the return policy for my recent purchase?”

• Chatbot: Retrieve: Looks up the return policy in the knowledge base. Generate: Responds with the relevant policy details.

Sounds huge , sounds expensive : No Worries ?? GenAI can do it in less than 5 months, in all areas of S2P at near to ZERO Margin Cost.

Want to know HOW? then read this article. Get Bonus feature about a quick early win in S2P that will give you atleast 18% benefit right away along with code !

Read on ...

Here's how ( reference Kotter's 8 step process and business canvas model framework )

1. Get a sense of urgency

Plug your long pending gaps and do your POC' wireframes in less than 4 weeks (2 sprints) to see what good looks like. assess your solutions buy make , LLMESH or GPT plug and play , Symbolic AI or hybrid AI or plain o'l RPA. Bottom line : "MAKE LIFE EASIER , PROCESS simpler , Cheaper and FASTER" (add a buffer cost for AI governance , Data accuracy , dependability and efficacy , security guardrails as you build your model not at the wireframe level)

2. Define your use cases and know what's available on the S2P GenAI use case Menu : Lessons learnt

>>>>Dont build when you can Plug and Play :

>>>>Opt for small nudges big impact as a priority.

Know your use cases and where you can get Plug and play features

• Category management : A genAI agent can define /sort/recommend and sort out category strategies that take months of stakeholder convincing .requirements gathering , spend planning - in Real time and personalized for every procurement requirement department wise.
• Strategic Sourcing : SS is still bidding but digitized . now imagine the RFx development , Response Review Stage 1 , Summarization is done automated , Insights from categorization management can be used to maximise value sustainable year on year . This feature is available in most S2P ISV's but only 13% of the people use it , leverage it more.
• Contract life cycle management (CLM): Painpoint -central visibility in a command center - single plane of glass is key.

• Dashboards are in one system , data is in data lakes , procurement uses other tools that needs a heavy itnegration layer. This makes it cumbersome , manual intensive and preparedness to response becomes limited. What if you are able to embed best customer strategies , contract payment term matching capabilities in your CLM governance STP and get a dashboard on demand real time at Zero marginal cost. It would move you to "somewhat satisfied" to STP digitally AI intuitive procurement in a few months.
• Supplier risk management (Auto Detect Auto Alert) and Relationship Management : The greatest gaps in SRM is 87% 3rd party risk , vendor screening and due diligence 75% and performance management capabilities is 75%. A generative AI agent could assess these factors in comparision to enterprise internal factors , changing demand and auto generate predictive risk and recommend "Action based strategies" for resolution for a ZERO marginal cost. Give the Supplier a persona and a score card (apple has done this years ago and metadata it. Simulate "what if scenarios on existing "power BI dashboards and thereafter analyze historical trends , demand profiles , and supplier performance using GENAI
• Requistion to Receive : Most procurement leaders feel they are already automated. But their turnaround time does not reduce .Why ? Req2Recieve do not Always optimize the automated path , deviations include non compliant spend potential , internal procurement cost , frustrated users who are told No!. All these requests amount to nearly 41% of total periodic request. An agent can provide ARN based intuitive contextual responses with high touch point support to help users get their answers apriori , find buy what they need from best of deal suppliers , negotiate quotes and what could get you a best deals (PS: When I tried this ,I felt as if I was booking my holiday on Agoda or Kayak)
• Invoice to pay : Payment checks are embedded in the CLM cycle .OCR , payment terms auto checks , exception management detection and remediation can be done by an agent that works continuous on a monitoring mode .
• Information Synthesis: using Data Retrieval and Insights: GENAI can retrieve, summarize, and extract insights from unstructured data sources along with real time -auto optimization opportunities.
• Augmenting Engagement: Walmart uses a auto negotiation tool called Pactum for better supplier deals . Optimize bid structures and bid outcomes to prevent errors.
• Automate Procurement Ops

2. Build momentum and remember to get a seat at the decision making table : you can do that if you can authenticate the "benefit $ numbers" if you do step 1 well. You need to build Trust with fact and conviction rather than HYPE.

3. Respect the S2P process for your decision making ( Tech leaders need to respect it ) Procurement is unique

Think specifically about S2P architecture not just Tech . ISV's will not tell you this . A bit of Human intervention in AI will be necessary but a lot of the current manual governance can be done by AI or RPA automation or AI with intuitive process orchestration , that solves nearly 48% of the problem and releases human manual grunt work - ( contract term matching , contract evaluation with Master service level agreement , vendor payment terms validation with PO/Invoice) takes nearly 65% of an AGENT's time.

A S2P framework specifically for GENAI has to be built in addition to the existing S2P framework , that is true to its risk tolerance , culture , complexity , investment appetite and value realization targets that is wants to achieve through GENAI investment.

This is how the workflow could look like - think beyond automation - think intuitive , sentiment Human AI decision making capability in choosing your AI/GENAI model to realize max value .

4.LEGO based progress in three steps in adopting GENAI.

(Spoiler alert: Too many silos in step 2 , without genAI governance will lead to 5.7X higher tech debt in just three years and 3.5X higher cost to serve in Tech). Remember to put a RAG-Ops/LLMESHOPS ops or GENAops in place to plug in the use cases in phase 3.

Phase 1: Pilot out prompt based use cases in silos- for workforce automation opportunities

Phase 2: embed AI plug and play solutions with end to end provisioning into S2P data domains of "thematic cure pain point areas discovered in point 1. verify the quality of insights gained before and after

Phase 3: 50%-80% S2P Procurement effort reduction but innovation , prompt , responsible AI checks have to continue as a support function. A S2P innovation leader role emerges herein to steer this journey forward.

Conclusion

After nearly 20 years in dealing with procurement systems , and measuring the value realized from digitalizing procurement, I would say think 30:70 paradigm. If Digitalization gave you 30% benefit , what is the operating and procurement innovation model you need to gain 70% benefit at zero additional marginal cost. That my friends is what GENAI would do to change the dynamics of Procurement in the next 2-3 years.

Nupour Mukherjee is a AI Board Advisor and provides GENAI that are designed to provide maximum value to your use case ,at the least cost with less of the hype. Follow her newsletter for more #GENAIValueFirst insights

Bonus feature : Early Win :

Matching Payment terms in a Contract life cycle management process using Generative AI

Model use :

• Traditional view : NLP TF-IDF, Neural networks , or Logistic regression . Avoid using Random forest =errors will be high as 68%.
• What I did :-): Used BERT/GPT : 79% accuracy first run !!!

Feature Extraction focus area : Word embeddings to represent contract text and extract relevant features to payment text.

What you need: Labelled dataset with contract 'payment terms' and corresponding text

Train the model : LR, NN and GPT helps reinforced learning much faster in lesser iterations saving on nearly 70% compute costs.

Code logic (simple python)- please include any of your own custom use case hyperparameterization for best results)

import pandas as pd

from sklearn.feature_extraction.text import TfidfVectorizer

from sklearn.model_selection import train_test_split

from sklearn.linear_model import LogisticRegression

from sklearn.metrics import accuracy_score

# Sample dataset (contract text and payment terms)

contracts = [

"Supplier agrees to net 30 payment terms.",

"Payment due within 45 days of invoice date.",

# ... more contract examples ...

]

payment_terms = ["net 30", "within 45 days", ...]

# Create a DataFrame

df = pd.DataFrame({"contract_text": contracts, "payment_terms": payment_terms})

# Feature extraction using TF-IDF

tfidf_vectorizer = TfidfVectorizer()

X = tfidf_vectorizer.fit_transform(df["contract_text"])

# Split data into train and validation sets

X_train, X_val, y_train, y_val = train_test_split(X, df["payment_terms"], test_size=0.2, random_state=42)

# Train a logistic regression model

model = LogisticRegression(max_iter=1000)

model.fit(X_train, y_train)

# Make predictions

y_pred = model.predict(X_val)

# Evaluate model accuracy

accuracy = accuracy_score(y_val, y_pred)

print(f"Model Accuracy: {accuracy:.2f}")

# Example prediction for a new contract

new_contract = "Payment terms: net 30 days."

new_contract_vec = tfidf_vectorizer.transform([new_contract])

predicted_term = model.predict(new_contract_vec)[0]

print(f"Predicted Payment Term: {predicted_term}")