Why your team keeps re-explaining the same context to AI tools (and how to fix it)

A client meeting, with a smiling woman shaking hands across the desk

The short answer

Your team keeps re-explaining context because people assume every new piece of information means rewriting the old rules, which is a structural problem. Organise context in layers, from the whole business down to each client, capture new insight as dated notes and review it regularly. Start with one shared pack for the job your team re-explains most.

Key takeaways

76% of workers say their preferred AI tools lack access to company data or work context (Salesforce and YouGov, 2026).
Most generative AI tools forget context between conversations and don't learn from feedback (unless specifically instructed), which MIT researchers link to most organisations seeing no return from AI (MIT NANDA, 2025).
One fix is to organise context in layers, from the whole business down to each client, so new insight is added as a note and reviewed, and nobody has to rewrite the brief from scratch.

Someone on your team opens a new chat and pastes in the same background they pasted in yesterday: who you sell to, what you offer, how you like things written and what's going on with the account. Someone else does the same with a different tool and a slightly different version of the story. Across a sales, marketing and BD team, a surprising amount of time goes on taking AI back to what is really an illusion of a starting point.

It's a familiar problem in a new place. Atlassian's research with 12,000 knowledge workers found that leaders and teams lose 25% of their time just searching for answers (Atlassian, State of Teams 2025). AI tools were meant to help with that, but most of them only know what someone tells them in the moment.

I wrote about the individual version of this in How to set up your first Claude project in 15 minutes. For a team it's harder, because the context isn't one person's alone to write down. Here are the four places it usually goes awry, how to organise context so it stays current, and where to start.

The context lives in people, not in a shared place

Most of what makes your team good at selling sits in people's heads or is played out in the moment during a sales call: which customers buy and why, which objections come up, which proof points land and what's changed this quarter. Some of it is stored in inboxes, call notes and personal documents. More and more of it sits in individual AI chats and private projects that nobody else has access to.

Each person ends up building their own version of the story, so five people get five slightly different answers from the same tools.

What helps: Start writing it down in one place the whole team can use, beginning with the background people paste in most often.

Every tool starts from zero

Most AI tools don't carry what they learned in one conversation into the next, and they don't share it with each other. MIT's research into why most organisations are seeing no return from generative AI describes the core problem as tools that forget context, don't learn and can't evolve. One interviewee said their tool "doesn't retain knowledge of client preferences or learn from previous edits" (MIT NANDA, 2025).

It's made harder by the number of tools in play. Salesforce found that workers toggle between an average of four different AI options, so the same background gets typed into several places (Salesforce and YouGov, 2026).

What helps: Use the features built for this, such as projects, custom instructions and shared knowledge files, and agree which tools the team uses for which jobs so the context lives in fewer places.

What workers expect if AI tools could access company data

Share of US workers who say secure access to company data would bring each improvement

Less time spent searching for information

62%

Better quality work

60%

Faster task completion

59%

Source: (Salesforce and YouGov, 2026)

Nobody keeps it current

Even when a team writes its context down, it dates quickly. Pricing changes, the offer gets repositioned, a new competitor appears or an account moves on, and the brief the tools are working from was built three months ago.

People notice the output is slightly off, stop trusting it and go back to pasting in their own version. It's the same pattern I see with CRM data: when nobody oversees and owns the update, nobody trusts the report.

Part of the problem is a belief that every new insight means rewriting the whole brief. A client mentions a new priority, or someone's view on a market shifts, and it feels easier to start again in a fresh chat than to work out what needs to change. Most of the context hasn't changed at all. What's changed is usually one detail about one client, one person or one idea.

What helps: Give each context pack a named owner and a review date, and make updating it part of the change itself. Maintenance is as critical as creation. When the pricing changes, the pack should be updated at the same time.

Handovers drop the context

Marketing generates a lead and it gets passed over to sales. What that person engaged with, what drew them in and the question they were trying to answer often doesn't transfer with them, so the first call starts cold with a customer who expects you to know the history.

The same thing happens between AI tools. The research doesn't transfer over to the call prep, and the call notes don't help to shape the proposal. The customer experiences one relationship with your business and feels it every time the conversation starts again.

What helps: Map where context needs to move across from one stage to the next, and decide what needs to be transferred with the customer at each handover. I set out how to do that in how to map your sales, marketing and BD workflow.

From experience

I see this often. Individuals get a good result from an AI tool once, then start from zero the next time because the most useful insight wasn't logged for the future. I hit the same problem on a bigger scale when I built an agentic system across my own commercial function. Insights from discovery weren't reaching the follow-up and the proposal, until I rebuilt it around the customer journey with shared objectives and context passed forward at each stage.

Zoom out: how context could be built

Before fixing any one tool, it helps to step back and look at how context works across the whole business. Some of it is true for everyone, some only for one team, and some only for one client or one person within that client. When it's all mixed together in one long brief, every small change feels like a rewrite. In the same way you might wireframe a website and decide how to organise its information hierarchically, organise your context in layers, so that each layer can change at the appropriate pace.

It's worth separating context from data before going further. Data is the facts and records: CRM fields, call transcripts, deal history, market news. Connecting AI tools to it helps, and it's why so many workers say their tools lack access to company information. Context is what tells the tool which of those facts matter and what to do with them: who you're for, what good looks like, what this stakeholder cares about and what has changed. Access to data can be solved with integrations. Context needs people to decide, write it down and keep it current, and that's where most teams get stuck.

The right structure will look different in every organisation, depending on its size, the markets it works in and how its teams are set up. What follows is a working hypothesis, a starting point to test against your own business and adapt.

A working hypothesis: how context could be layered

Horizontal layers run from the whole business down to each conversation. Optional verticals cut across them, and the external track brings in what's happening outside the business.

Internal layersExternal track

EnterpriseWhat's true for the whole business: positioning, offer and pricing logic, brand voice, approved proof, policies.

Market and macroEconomic, regulatory and technology trends

Verticals or offeringsOptional

Vertical A

Vertical B

Vertical C

Sector dataMarket data and news for each vertical

Role and functionSales, BD, marketing

CompetitorsPositioning, pricing, wins and losses

ICP and accountSegments, clients, stakeholders

Company signalsFunding, hiring, leadership changes

Live insightDated notes from calls and meetings

External track: market and macro trends beside the enterprise layer, sector data beside each vertical, competitors beside each function and company signals beside each account.

Each client inherits the enterprise layer, its vertical, the function serving it and its own account context. New insight is added at the bottom and only moves up once it's been reviewed. Adapt the layers to your organisation.

The enterprise layer. What's true for the whole business: positioning, your offer and how pricing works, brand voice, approved proof and policies. Owned by leadership, and it changes a few times a year, typically alongside your strategic planning reviews.
The role and function layer. How sales, BD and marketing each do their work: playbooks, qualification, messaging by role, templates and what good looks like. Owned by each function lead, and reviewed monthly.
The ICP and account layer. What's true for each target segment, then for each client: the stakeholders, what each person cares about, how they like to work, what's been agreed and the last thing they said that matters. Owned by the account lead, and updated after each meaningful conversation.
Live insight. Dated notes from calls, meetings and emails, captured as they happen.

If your business works across several vertical markets, there's a useful optional layer that sits just below the enterprise core: a stream for each vertical, or for each product or service you offer. When I visited Ping An, I saw a similar architecture, with separate data streams for each of the vertical markets they operate in, feeding the context their teams and tools work from. Each vertical carries what's specific to that market, such as its regulation, terminology, buying patterns and the proof that lands there, while drawing on the same enterprise context. Your functions then work across the verticals, and each client inherits the enterprise layer, its vertical, the function serving it and its own account context.

Alongside these, you might also choose an external track that draws in what's happening outside the business, matched to each layer: market and macro trends beside the enterprise layer, sector data beside each vertical, competitors beside each function, and company signals such as funding, hiring or leadership changes beside each account. Keep these as dated facts with a source, separate from what the business has decided to do about them, because they go out of date much faster.

When someone picks up a piece of work, their AI tools read the layers that apply to it: the enterprise layer, the client's vertical if you work across several, their function, the client's segment and the account itself. The context is assembled for them, so nobody has to type it in again. A lower layer can add detail for one client, but it shouldn't contradict the layers above it. If it does, that's a conversation for the team, and a useful one.

How it stays current: capture, route, promote, review

The structure and how this runs will also vary from one organisation to the next, but much of it can now be automated, with people reviewing the changes that matter most. A simple rule keeps it safe: the more of the business an update affects, the more human review it needs.

Capture. After each call, meeting or important email thread, AI drafts a short insight note: what's new, who said it, which client and stakeholder it relates to, and the date.
Route. The note is matched to the layer it affects. Account updates are added straight away, with the account lead able to check and correct them. Changes to a segment or a function go into a weekly review for the function lead. Anything that would change the enterprise layer is a proposal for leadership to approve.
Promote. When the same insight turns up across several accounts, such as one objection coming from three clients in the same sector, it's flagged to move up a layer. That's how the business learns from individual conversations without anyone rewriting a brief. When I visited Nvidia last year, I heard about a similar approach being applied incredibly effectively: every employee shares the equivalent of the top five things they're working on every fortnight, and AI analyses them for insight that can be extracted and acted on.
Review. Every entry has an owner and a review date, and anything past its date is flagged as possibly out of date.

The same approach works for people's thinking, which keeps developing, and that's healthy. When someone's view on a client, a market or a competitor shifts, add a note saying what changed and why, and keep the earlier note in the log. It shows how the thinking has developed, and it stops the next person repeating a conversation that has already moved on.

You don't need specialist software to start. Most teams can do this with what they already have: a call recorder that produces summaries, the CRM for accounts and stakeholders, shared projects or knowledge files for the enterprise and function layers, and a simple review list.

Zoom back in: your first shared context pack

You don't need the whole structure in place to start. A first context pack covers the enterprise and function layers for one job: a short set of documents your AI tools read before they respond, the same way you'd brief a capable new starter on their first day. Account notes can follow once the pack is working. One pack for each commercial job, such as outreach, proposals or content, works better than one pack for everything. Keep it focused: three well-chosen documents usually beat fifteen random ones.

The six parts of a shared context pack

Part 1Who we sell to

Your ideal customer profile, the roles you sell to, their priorities and the problems you solve for them.

Part 2What we offer

Your products or services, how pricing works, and what's included and what isn't.

Part 3How we sound

Tone of voice, words you use and avoid, and two or three pieces of writing you're proud of.

Part 4Proof that lands

Results, case studies and testimonials you're allowed to use, plus the objections you hear most.

Part 5What's live right now

This quarter's focus, priority accounts, current campaigns and anything that has recently changed.

Part 6What good looks like

Strong examples of outreach, proposals or posts, with a line on what made each one work.

How to set it up in a week

Choose the job your team re-explains most often. Outreach, proposals and LinkedIn content are common starting points.
Ask the two or three people who do it best what they always tell the tool. Write their answers into the six parts above.
Put the pack where the tools can read it. That might be a shared project, custom instructions or a shared knowledge file, and the team agrees to work from that version.
Name an owner and set a monthly review date. Capture new insight as dated notes, and use the monthly review to move anything that has become widely true into the pack.
Test it with real work for two weeks. If people are still pasting in background, find out what's missing and add it.