Clean Consulting Data Before Agents Use It
Agents can be wrong in a very convincing way
A consulting firm’s shared drive can look like a valuable knowledge base. It may hold years of proposals, project plans, research, client presentations, meeting notes, and delivery guidance. But the files often disagree, and nobody can tell which version is safe to use.
That’s a problem before an agent ever touches the material. Once an agent searches the firm’s records and turns them into an answer, inconsistency can become harder to spot. It may cite an old case study as current, reuse a proposal’s outdated pricing, or apply one client’s delivery guidance to another engagement. The result can sound polished and still be wrong.
The VB Pulse Agentic Context Layers Report from September 2026 puts attention on the context agents rely on. For a consulting firm, that context isn’t abstract. It’s the project record, proposal library, and internal guidance that the firm uses to make decisions and serve clients.
Before you connect an agent to those materials, check whether they’re current, consistent, and appropriate for the work. Cleaning up the context is not a separate admin project. It’s part of making sure an agent’s answers have a chance of being useful.
The costly work starts before the agent
In many consulting firms, senior people often carry a lot of the firm’s knowledge in their heads. They know which proposal is the closest match, which past project is safe to reference, and which delivery method changed after the latest engagement.
That can work while the firm is small. It gets expensive when the same people repeatedly reconstruct the firm’s knowledge for new work.
A major proposal can take substantial senior and specialist time. Some of that effort is thoughtful tailoring. Some is hunting for the last relevant deck, checking whether its case study is still accurate, asking who owns the latest pricing, and rebuilding material that already exists somewhere.
Research has the same pattern. Teams may spend weeks on secondary research at the start of an engagement, then repeat similar work for another client because findings were never stored with clear sources, dates, and context. Each project creates IP, but the firm pays for the same insight twice when nobody can reliably find or reuse it.
Repeated work can create significant costs for consulting firms. Treat that as a working range, not a guaranteed saving or a universal figure. The amount that applies to your firm depends on proposal volume, team mix, utilization, project economics, and how much work truly repeats. The point is that rework has a dollar value, even when it hides in partner time and write-offs rather than appearing as a line item.
Why clean context comes before agent access
An agent can retrieve information quickly. It can’t reliably infer which of two conflicting documents represents the firm’s current policy unless you give it a way to tell. A newer file name doesn’t prove the content is approved. A complete-looking project folder doesn’t prove the client permitted its work to be reused.
Start by checking three types of information.
Project records. Confirm each project has an owner, client or industry context, dates, status, and a clear distinction between approved deliverables and working drafts. Mark confidential content and any restrictions on reuse. If project names, sectors, or outcomes differ across the CRM, project management system, and final deck, resolve the mismatch or keep those records out of agent search.
Proposals and pitch material. Identify the current proposal template, approved case studies, pricing rules, and descriptions of services. Archive drafts and expired pricing rather than leaving them beside approved documents with equally plausible names. Check that claims about past results match the source project and have permission to be used in new pitches.
Internal guidance. Find the current delivery playbooks, review steps, and quality standards. Give them an owner and a review date. If two guides disagree, ask the person accountable to decide which one applies. Don’t ask an agent to settle an internal policy dispute by guessing from whichever document it finds first.
This is less about making every file perfect and more about making the important differences visible. A file marked “draft,” “approved,” or “retired” gives an agent and its human reviewer a better basis for deciding what to use. So do clear dates, owners, source links, and access rules.
Firms considering a pilot can also read why agent pilots stall before production in consulting firms. A common lesson is that a working demo doesn’t prove the underlying records are ready for client work.
What the workflow looks like end to end
Take a new proposal request. A partner wants a first draft for a prospective client in a particular industry. Today, a senior consultant may search old folders, message colleagues for the right example, and copy pieces from several decks. They then spend time checking whether those examples are accurate and whether the pricing fits the opportunity.
A Proposal Generation Agent should do more than search and paste. First, it should receive the opportunity details and identify which approved proposal materials, case studies, and pricing guidance are relevant. It should respect permissions and exclude records that are stale, restricted, or not approved for reuse. Then it can produce a draft with references to the source material it used and clearly mark anything that needs human confirmation.
The partner still owns the proposal. They check the fit, validate claims and pricing, and decide what to send. The agent takes on the search and first-draft work, while source references make its choices easier to review.
An engagement kickoff can follow a similar sequence. A Research Agent gathers structured company and industry research, records sources and dates, summarizes the findings, and prepares a one-page brief. The team can review the source trail, correct weak or outdated information, and use the brief as a starting point rather than rebuilding the same research from scratch.
After delivery, a Knowledge Agent can help staff ask questions across decks, documents, and meeting transcripts. But it should answer from material the firm has made searchable and appropriate for the question. The answer should point back to its sources. If a question crosses client boundaries, access controls should prevent the agent from exposing restricted project material.
These agents depend on the same basic foundation. If records have no owner, if project outcomes conflict, or if internal guidance is out of date, faster retrieval only makes it easier to reuse the wrong thing. For a closer look at the build and buy decision, see how consulting firms can approach AI agents.
Set rules for what the agent may trust
A practical readiness check doesn’t require a full data-cleaning program before a pilot. It does require a defined boundary around the information the pilot can use.
Choose one workflow with meaningful repetition, such as proposal drafting or engagement research. Identify the systems and folders it depends on. For each important source, answer a few basic questions:
- Who owns this record or document?
- Is it current, approved, and complete enough for this task?
- Does it contain client information that can’t be reused?
- Is there a more authoritative source that supersedes it?
- Can a reviewer trace an agent’s answer back to the source?
Then decide what the agent should do when the answer is unclear. It might flag conflicting materials, leave a field blank, or ask for review. That’s preferable to presenting an unverified guess as a firm-approved answer.
Access and governance matter alongside accuracy. An agent that can search every client folder may find useful precedent, but broad access can also expose information to the wrong team or workflow. The firm should set boundaries before connecting systems, not after an uncomfortable output. This is the concern behind the insider-risk questions consulting firms should address.
The initial scope can stay small. Pick a limited, approved set of documents, define who can use the agent, and test it against questions where your team already knows the right answer. Review the sources it selects, the answers it produces, and the cases where it should have declined or asked for help. Expand only after the team can explain what the agent is drawing on and where it still needs human judgment.
Find the first useful workflow in an Omni Audit
In an Omni Audit, we map the manual work, identify the systems and records behind it, and define a practical first step. No deck.
That conversation should make the data problem concrete. For example, we may find that proposal work repeats because case studies aren’t tagged by service and industry, while the pricing document has two active versions. Or the research process may be sound, but teams have no consistent way to save sources and dates for reuse. Those are different problems and call for different first steps.
If you want to map that work with us, Book a 60-min Omni Audit. You can also see Omni for consulting firms and the kinds of workflows we assess.
Make the economics visible
A useful business case compares the cost of today’s repeated work with the cost and review effort of a focused agent workflow. Count the time spent searching, reconciling, and rebuilding, not just the time spent typing. For proposals, separate the hours required to tailor a credible pitch from the hours spent locating and checking old material. For research, distinguish new analysis from research that’s repeated because previous work wasn’t findable or reusable.
Then account for the checks that remain. Partners still need to review proposals. Consultants still need to judge whether research fits the client’s question. Somebody still owns the records and decides when guidance is current. If a proposed workflow saves time but leaves the same review burden, its value may be smaller than expected. If it makes sources visible and cuts repeated search, it can help the firm spend more of its people’s time on judgment and client work.
This is why readiness should be measured in the workflow, not as a vague percentage of “clean data.” Can the agent find approved examples? Can it tell current guidance from retired guidance? Can it show where an answer came from? Can the right person review the output without searching from scratch? Those are useful pilot questions. They also help a firm decide what to fix first.
For a practical worksheet, use Deploy Your First Business Agent. It helps you scope an initial workflow, identify the records it needs, and set out the checks people should make before the agent’s work reaches a client. You can download the worksheet directly.
Start with the records behind one recurring task
Don’t begin by connecting an agent to every shared drive and asking it to make the firm’s knowledge reusable. Start with one repeated task, then find the project records, proposals, and guidance that task depends on. Reconcile what matters, label what’s approved, set access boundaries, and decide how the agent should handle gaps or conflicts.
That’s a sounder route to a useful first deployment than hoping the agent will sort out the firm’s filing habits. It also gives partners a way to judge progress by work that matters: less time searching for proposal material, fewer repeated research steps, and clearer review of the sources behind a client-facing answer.
If you’re ready to choose that first workflow, Book my Omni Audit. We’ll use the 60 minutes to find the work worth improving, identify the data conditions that could trip it up, and leave you with a pilot plan. For the consulting-specific details, visit the Omni Audit for consulting firms.
Your guide is ready
Check your downloads folder. If it did not open automatically, use the button below.
Download the GuideYour guide is ready
Check your downloads folder. If it did not open automatically, use the button below.
Download the GuideTalk it through
Talk it through with Sam
30 minutes on what a Command Centre would look like for your business.
Book a call