Three-Pillar Enterprise AI Strategy
Sequence AI adoption from general tools to partners and proprietary agents
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 99%
The Three-Pillar Enterprise AI Strategy divides adoption into increasingly specific layers. First, the organization enables general efficiency tools for questions, research, and early writing drafts. Second, it works with external partners that are already investing in AI and can help extract value from organizational data. This layer depends on cleaning the underlying data because incorrect inputs produce false findings regardless of model capability. Third, the company develops proprietary agents for narrowly defined workflows and client questions that create distinctive value. Across all three pillars, business and technology specialists must translate the problem, constraints, data, and expected result for one another. The sequence captures immediate productivity while building the foundations needed for trustworthy, differentiated automation.
Origin
Kate Burke described this three-part strategy as Allspring Global Investments' approach to adopting AI.
Core principles
- 01Start with broadly useful efficiency tools
- 02Use partners where external capabilities accelerate progress
- 03Clean data is a prerequisite for trustworthy answers
- 04Build proprietary agents around specific workflows and client needs
- 05Translate continuously between business problems and technology
How to run it
- 1
Enable general efficiency
Provide governed access to broadly useful AI tools for questions, research, and first drafts.
Pro tip Begin with reversible, low-risk tasks where humans review every output.
Watch out Do not paste restricted client or company information into unapproved services.
- 2
Prepare the data foundation
Identify data needed for higher-value use cases, then clean, govern, and validate it.
Pro tip Assign owners to critical datasets and define acceptable quality thresholds.
Watch out Querying incorrect data can produce confident false findings.
- 3
Leverage capable partners
Work with vendors and platforms whose AI investment can accelerate data and workflow solutions.
Pro tip Evaluate partners against a specific business outcome rather than feature lists.
Watch out A partnership cannot compensate for unclear ownership or poor internal data.
- 4
Build specific agents
Develop internal agents for clearly bounded questions, workflows, and client needs unique to the organization.
Pro tip Prioritize cases with repeatable inputs, reviewable outputs, and measurable value.
Watch out Avoid building a general agent when a narrow workflow would be safer and easier to evaluate.
- 5
Bridge business and technology
Pair domain experts and technologists so the business issue is translated into accurate requirements and controls.
Pro tip Keep the business owner accountable for whether the output is useful, not merely whether the system runs.
Watch out Technical success without business relevance creates unused tools.
In the wild
An asset manager first enables approved drafting tools for internal communications. It then cleans product and reporting data with a technology partner. Finally, it builds a governed agent that answers a narrow class of client-service questions using validated internal sources and human review.
→ The firm gains early productivity while progressing toward a differentiated, trustworthy workflow.
Common mistakes
Building on dirty data
Incorrect source data creates unreliable answers and false findings even when the model performs as designed.
Starting with an oversized agent
A broad proprietary system is harder to validate than a specific business case with bounded inputs and outputs.
Separating builders from users
Without translation between business and technology, the system may solve the wrong problem.
Is it for you?
Best for
Established organizations introducing AI across general productivity, data analysis, and proprietary workflows.
Not ideal for
Teams without a defined business problem, usable data, or the capacity to validate AI outputs.
From the transcript
“So I think of. AI or strategy around AI in really three ways.”
“It's all about data at the bottom. You need really clean data.”
“you need to be able to put the business issue and the technology. You need to be able to be able to translate between the…”
From the episode
Building an Asset Allocation Strategy: Masters in Business with Kate Burke
Kate Burke