Agentic AI for Contract Automation
Key Points
- Traditional contract and ECM systems store agreements in centralized databases but still require experts to manually locate, read, and extract key terms, making the process slow and inefficient.
- A common use case involves lease agreements where stakeholders must repeatedly reference specific clauses to determine next actions, highlighting the burden of manual document handling.
- The emerging paradigm of automated contract processing leverages AI to instantly analyze complex documents—whether multi‑party contracts or simple terms—and surface critical information as soon as the document arrives.
- By integrating “agentic” conversational interfaces, AI can reason about contract content and interact with users, turning static document repositories into interactive assistants that streamline decision‑making.
- This AI‑driven approach aims to reduce expert workload, accelerate business workflows, and provide real‑time insights from contracts that were previously buried in manual reviews.
Sections
- Understanding Modern Contract Automation - The speaker outlines current contract management tools—dedicated contract repositories and generic ECM platforms—and explains their limitations while illustrating a lease‑contract use case where automating agreement data informs next‑best‑action decisions.
- Agentic AI for Document Automation - The speaker describes how an agentic conversational AI can coordinate multiple specialized bots to automatically extract, reason about, and summarize complex legal documents, dramatically improving subject‑matter expert efficiency.
- AI‑Powered Orchestration for Contract Automation - The speaker outlines how a central orchestration hub leverages generative and traditional AI, along with a vector store for metadata, to automatically ingest, index, and process incoming contracts for greater efficiency.
- AI-Powered Contract Metadata Extraction - The speaker describes using an AI model and a crawler to ingest contracts, generate searchable metadata (e.g., force majeure, cancellation terms), and store the processed content in a vector store for easy retrieval via the hub.
- Conversational AI for Contract Retrieval - The speaker explains that simple keyword searches miss contractual nuances, so metadata‑enriched, contract‑trained AI models must index documents in a vector store, enabling users to query contracts through a conversational hub.
- AI‑Powered Contract Clause Comparison - The speaker explains a workflow where relevant contract excerpts are fetched from a vector store and supplied to a large language model with specific prompts to compare and clarify differences between clauses and related obligations.
- Integrating Business Rules via Hub - The speaker explains how a central hub uses AI to combine contract data, business rules, and other systems, automating decision‑making and communication generation.
- AI‑Orchestrated Contract Retrieval Workflow - An AI hub coordinates email, document, CRM, and ERP services in response to user conversational requests for contracts, delivering context instantly and reducing onboarding friction.
- Scaling Contract Automation with AI - The speaker explains how an AI‑driven orchestration hub integrates existing systems (ERP, CRM, business rules) to replace manual scaling, enabling faster, higher‑value processing of complex contractual documents.
Full Transcript
# Agentic AI for Contract Automation **Source:** [https://www.youtube.com/watch?v=E0Pd8mUpr-M](https://www.youtube.com/watch?v=E0Pd8mUpr-M) **Duration:** 00:26:30 ## Summary - Traditional contract and ECM systems store agreements in centralized databases but still require experts to manually locate, read, and extract key terms, making the process slow and inefficient. - A common use case involves lease agreements where stakeholders must repeatedly reference specific clauses to determine next actions, highlighting the burden of manual document handling. - The emerging paradigm of automated contract processing leverages AI to instantly analyze complex documents—whether multi‑party contracts or simple terms—and surface critical information as soon as the document arrives. - By integrating “agentic” conversational interfaces, AI can reason about contract content and interact with users, turning static document repositories into interactive assistants that streamline decision‑making. - This AI‑driven approach aims to reduce expert workload, accelerate business workflows, and provide real‑time insights from contracts that were previously buried in manual reviews. ## Sections - [00:00:00](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=0s) **Understanding Modern Contract Automation** - The speaker outlines current contract management tools—dedicated contract repositories and generic ECM platforms—and explains their limitations while illustrating a lease‑contract use case where automating agreement data informs next‑best‑action decisions. - [00:03:02](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=182s) **Agentic AI for Document Automation** - The speaker describes how an agentic conversational AI can coordinate multiple specialized bots to automatically extract, reason about, and summarize complex legal documents, dramatically improving subject‑matter expert efficiency. - [00:06:07](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=367s) **AI‑Powered Orchestration for Contract Automation** - The speaker outlines how a central orchestration hub leverages generative and traditional AI, along with a vector store for metadata, to automatically ingest, index, and process incoming contracts for greater efficiency. - [00:09:12](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=552s) **AI-Powered Contract Metadata Extraction** - The speaker describes using an AI model and a crawler to ingest contracts, generate searchable metadata (e.g., force majeure, cancellation terms), and store the processed content in a vector store for easy retrieval via the hub. - [00:12:20](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=740s) **Conversational AI for Contract Retrieval** - The speaker explains that simple keyword searches miss contractual nuances, so metadata‑enriched, contract‑trained AI models must index documents in a vector store, enabling users to query contracts through a conversational hub. - [00:15:20](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=920s) **AI‑Powered Contract Clause Comparison** - The speaker explains a workflow where relevant contract excerpts are fetched from a vector store and supplied to a large language model with specific prompts to compare and clarify differences between clauses and related obligations. - [00:18:46](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=1126s) **Integrating Business Rules via Hub** - The speaker explains how a central hub uses AI to combine contract data, business rules, and other systems, automating decision‑making and communication generation. - [00:21:51](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=1311s) **AI‑Orchestrated Contract Retrieval Workflow** - An AI hub coordinates email, document, CRM, and ERP services in response to user conversational requests for contracts, delivering context instantly and reducing onboarding friction. - [00:25:01](https://www.youtube.com/watch?v=E0Pd8mUpr-M&t=1501s) **Scaling Contract Automation with AI** - The speaker explains how an AI‑driven orchestration hub integrates existing systems (ERP, CRM, business rules) to replace manual scaling, enabling faster, higher‑value processing of complex contractual documents. ## Full Transcript
Hi. Today we're going to talk
about contract automation.
So let's set the stage real quick.
Contract automation
and the processing of,
like, agreements and complex
documents, like contracts,
today exist in various forms of technology.
Most of the time, you will have,
for example,
maybe a contract system.
And, that could be,
where all the agreements go.
They're stored in a database.
So that contract system
would be where the users have to go
to access the content
and deal with any sort of terms
and conditions that would relate to
any particular entity
mentioned in the contract.
Not the best,
but, at least, it's all in one place.
But that's today.
Other systems and other organizations
might have
a system called
Enterprise Content Management.
We'll call that ECM.
And, in an ECM system...
This is far more a generic
workflow and document management system.
And what you find in an ECM system,
really, are just,
kind of, electronic file cabinet
with workflow, records management, and
a lot of the base
functionality
to deal with electronic records.
However, in these two systems,
when it comes to dealing
with contracts, in particular,
there's a lot of instances, really.
Where this video was inspired
was a use case
where contracts and lease
agreements were frequently referenced
in order to determine and, like, next
best action, because there were entities
that were involved,
and properties that were involved.
And, so it was important for the
the contents, or the specific Ts and Cs
of that document,
to be at the hands of the expert,
which, by the way, would have to go in
and access from this system,
or in some cases
this system, to pull the document back.
But here's the thing.
They had to read this with their own eyes,
and take the time to digest
where in the contract
they needed to focus their reading
to extract
the key information
they needed to make a decision.
Well,
in our video today,
we're going to talk about a new paradigm.
And that is really automated
contract processing.
And this is a paradigm that fits really
any sort of complex document.
So, whether it's a lease agreement,
or some sort of terms and conditions between,
either multiple parties
or just a simple two party system,
you really need to have this information,
when it hits the door,
you need to understand what's in it.
It's beneficial
to have these types of insights
because it speeds the process along.
But, with the tools of today, they really
require the subject matter expert
to constantly dig into the document
in order to get that information.
In this new paradigm,
we're going to talk about
how we can automate that
and use AI in the process
to help
the subject
matter expert be more efficient.
Now, there's a term
today called agentic.
You may
read this
term and correlate it to
a type of AI bot or virtual assistant.
But, essentially,
the term agentic is really referring
to some sort of
conversational interface
that allows the
understanding and reasoning
of complex problems in order to carry out
a task, or multiple tasks. And the ability
for AI to reason
through complex type
tasks, really,
kind of,
has been something that has
come of late because we have systems
now that have multiple agents.
So you have bots for certain
types of tasks.
That's all they do.
And, now, you have this new kind
of virtual assistant
that sits on top of those other bots
to sort of coordinate or orchestrate
those activities.
Because many of the things
that we want to automate in business
are more complex than just a simple
single transaction, right?
Doing a lookup, in a ticketing system
to find the status of a ticket.
While that's something
a bot can do today,
we certainly want to do more than that
when we apply AI
to something
that is, it's tied to as much value
as these complex
documents and contracts in particular.
So what you will see
with the enterprise
content management system,
or the contracting system, is...
We have some new roles
or new players in this picture.
So at the foundation,
of virtually any sort of
enhancement or automation today,
is our good friend generative AI.
So...
The AI,
or the large language model,
or the traditional AI, has power today
to really decipher and
go multimodal from image
recognition, image
generation, audio
generation, audio understanding,
and, definitely, free flowing
text understanding and text generation.
And that's the piece that we will use
from AI in this particular scenario.
The other thing that's at play here
is an orchestrator,
or an orchestration hub.
So we're just going to call this "the hub".
So,
when you have...
and as I mentioned before,
the agentic concept,
you have lots of different bots
and lots of different,
what we'll call, skills that sit inside
or are controlled by, the hub.
And one of the primary skills
is the ability for the hub
to leverage AI
when it makes sense.
So, in this scenario,
what we're going to see
is a flow that allows you to leverage
generative AI and traditional AI
throughout the process
to make it more efficient.
So.
We'll
grab our contracts
as they hit the door.
They'll either be still stored
into our contract system
or into the ECM system.
And what we'll find
is that, as these
documents arrive,
we'll have...
a triggering event
that will put
our documents...
into our vector store.
Now, the vector store has a purpose.
And, in this scenario, it will contain
all the metadata about the documents,
but also all the contents
of those documents.
And the metadata that we will create
will be helpful in determining elements
of the contract,
because we'll have a model
that when the documents are ingested...
So we're just going to put
this model here.
Okay.
And then we'll put this model here.
And,
with these models,
we're going to break up the documents,
and we're going to create metadata,
so that they can be stored
neatly in our vector store.
So, today, you can see in the blue,
we have
places for these documents to reside.
That doesn't change.
What does change is now
these documents are being siphoned off
and the contents of those documents
are being processed,
so that they can be searchable and usable
from our hub.
So, once the documents are in the hub,
by the way, our hub also has...
a skill
to retrieve
from the vector store.
So, in this process,
we've now been able to wire up
all the contents
of where our
essential agreements reside today.
So minimal impact.
We have a crawler.
It's a little
crawler that will go out and grab
all of those agreements.
And, sometimes,
the agreements are in contract systems,
they're actually stored
as blobs in the database,
so the crawler actually
has to open the database,
pull that blob,
column out,
that field out, and process it that way.
So, once we get that content,
we run it through our model
to create the necessary metadata
relating to things, like: What's a force
majeure clause?
What are the terms of cancellation?
Those types of "contract"
sections that are fairly common,
the model is going to identify
that,
and it's going to place some metadata
around that text,
so that when we, from our hub,
want to pull back specific sections
of that contract,
that metadata helps us pull it back
much more efficiently.
Because, remember, when it comes to
contracts,
you really can't do
word find or a word search.
That doesn't work.
Because it's the context of the phrase,
or the context of the clause,
in that particular agreement,
that makes all the difference.
And those clauses and phrases
have lots of different
disparate words
in them to describe
or attain the same meaning.
So having even a semantic search,
oftentimes, will leave you
without being able
to locate certain things.
So what you need in
these models are contract
trained traditional AI type models.
Or, you could also, use generative AI
to identify those things as well.
But, either way, we need a model
to create some metadata
before that content goes into the
vector store.
Now, once it's in the vector store,
we have...
this user,
who is over here,
looking at a screen.
And, previously,
the user accessed
either the content, the contract store
system together
with ECM, or they just went into ECM.
Depending on
how an organization has deployed,
they, the management and the access,
we still have a user sitting out here
who's remotely connecting
to accessing these documents.
And, sometimes,
those user interfaces are native,
in the case of the contracting system,
those user interfaces
can be native
to the actual contract system.
However,
when we look at our new workflow,
we're going to connect this user
to our hub.
And we're going to do it
conversationally.
That's where, if you consider...
the conversational aspect
of what previously was probably
a serial workflow of moving documents
through a process.
Now, in this new agentic
type architecture,
we use a chat or,
what we would call, an agent of agents
to leverage
all the skills we have in the hub
to accomplish our contract use cases.
Now, there are multiple contract
use cases.
Many of them, kind of circulate around,
confirming terms and conditions,
looking at dates, as they
become effective and expire.
But a lot of that work can be done
interrogatively through chat.
For example, you can type in:
What's the difference
between the 2023 force majeure clause
and what we have
in the 2024 force majeure clause?
And, what will happen in that scenario...
So, when we have a question,
what happens first
is, step one,
we retrieve
the relevant content
from the vector store.
Now, keep in mind,
the original contracts are still
in the record system.
We haven't changed that.
What we're doing
is we're accessing the content
that we've indexed into the vector store.
Remember,
our model is helping us
get to that content much faster,
because we're identifying things
that are in a contract,
such as parties involved,
terms and conditions.
So we're searching,
including, those types of elements
in our metadata. Okay.
Now. Once we retrieve
that relevant content for the question,
the second thing we do...
is we provide the AI,
in this case, we could say,
you know... LLM.
Because we're doing text.
We would provide that content
to our large language model.
And, then, a set of instructions,
giving the AI
a purpose for the inquiry.
Compare these two clauses.
Provide guidance or information
on what is missing
when comparing one or the other.
A lot of times in contracts,
we really just need to understand
what's changing,
in addition
to what we're on the hook for.
So we can ask multiple questions...
And what the AI will do is will take,
this is what you might see
as a retrieval augmented generation
scenario
that is quite popular,
otherwise known as RAG.
But what the instruction is doing
is we're leveraging
our hub,
and we're taking
the information that we get from the LLM,
and we can now,
with our hub,
evaluate for
business rules.
We'll call this BR.
Now, our business rules...
That could be something that
helps the subject matter expert determine
the best course of action.
Because, inside our business rules,
we have thresholds and conditions
that really are strategic
and can inform the user,
rather than having the user
get information from the contract.
Because today
this is what they have to do.
They have to go to the contract system,
they have to open up the contract,
then they have to read it and digest it.
Then, they either have to reference
the business rules
separately in a different system,
or they have to
pull some sort of expert in to supplement
their expertise around the business rules
in order to have the full context.
But, look at what we have here.
We have a hub
that's bringing all of this together.
So, in step three,
we can evaluate our business rules.
What effects do we have?
What important thresholds are,
you know, present in the context of our query.
And, then, once
that's done,
there could be yet another step
for generating communications.
And what we'll do
is we'll call this "other".
Because "other"
would encapsulate
any other type of function that may
or may not be applicable to this process.
And, once were
exhaustively complete
on touching all the necessary systems,
keep in mind,
the hub is now doing all the integrating.
It's leveraging the AI.
It's leveraging our content,
which is our truth system of record.
It's leveraging applicable business rules
and other systems,
and keeping it
conversational and interactive
with the user.
So what does this mean?
This means that the user has less
complexity,
has less friction,
less cost of time,
because the AI, and the LLM in
this case, is providing summaries,
fact extraction,
as well as comparison,
and all the things that
they would have to do,
this user would have to do, themselves
with their own cognitive ability.
Now, they're getting assisted by the
AI orchestrated by our hub,
and the hub brings in
all the other services.
Could be that we need to send an email,
could be that we need to
pull in other documents as it relates
to our particular agreement.
Maybe there is a record in Salesforce,
or our ERP or CRM system,
that is necessary
to establish the full context.
That's where "other" comes in.
So all of this is triggered
by this one user conversationally.
Saying, I need to process,
or I need to investigate,
this year's agreement,
or last year's agreement.
Or I need to access the agreement,
the lease agreement, for this property.
All of these types of things
are just conversational and can trigger
a multitude of actions.
Simultaneously, by the way,
for our user to bring
the full context to their screen.
Now, what does this do?
As I mentioned before,
this essentially saves...
time.
Okay.
Saving time
means that we have less friction,
we have greater efficiency, and,
I mean, in this particular scenario,
it's more natural.
Because the user is not having to learn
or adapt.
I mean, keep in mind,
this user doesn't stay the same.
There's obviously career churn that
this population of users,
the folks that deal with these contracts
and documents, they can be,
they can move on
to other areas of the business.
And then we have a new person.
And what this new person has to do today,
without the benefit of this orchestration,
is have to really learn
and apprentice under
somebody who knows
and has been doing this.
But this new architecture,
and new approach, not only saves time
but accelerates
value.
Because
this person can be a new person,
or they can be a seasoned person,
and still have access
to the same orchestration of events.
And with the AI model,
tuned for this particular
organization's content,
they can get trusted results.
And, last but certainly not least,
we can scale.
Many organizations have trouble
in complex documents
when they approach scale,
because the only way they can scale
is to actually add people,
because it's really through the subject
matter expert
that this content
is evaluated and adjudicated.
But, if we're able to let the AI help us
along with our orchestration hub, then...
all of a sudden we're moving faster,
we can have stronger value sooner,
and that will allow us to do more
in a shorter amount of time.
So,
in summary,
contract automation
is...
the addition
of an orchestration hub
that leverages key systems,
such as business rules,
your,
you know,
your existing
investments, like your ERP and CRM system,
as well as AI,
both traditional and generative.
All for the complex,
but yet simplistic to the user, purpose
of processing these complex documents.