Rolling out Claude or Copilot on top of Salesforce or HubSpot makes individuals a little faster. It doesn’t make your business better than your competitors’, because they have the same tools and the same defaults. And behind the box marked “we’ve done AI”, most firms are carrying three problems they can’t see:
1. You’re leaving revenue on the table. The AI only sees the CRM, so your people spend their time on the wrong advisers and clients.
2. You’d struggle if audited. Staff are pasting client data into AI tools nobody approved or logged. One in five organisations in IBM’s 2025 breach study had a breach involving exactly this.
3. You’re paying more without investing better. A 20-user US vendor stack costs roughly $486k in its first year. Datafabric delivers comparable scope for roughly $160k, and builds a data foundation that compounds, instead of a stack of licences.
Same tools as everyone else means the same results as everyone else.
- Two kinds of firms
- Problem 1: you’re leaving revenue on the table
- Problem 2: you’d struggle if audited
- Problem 3: you’re paying more without investing better
- Side by side
- What it adds up to, for your products and your P&L
- When we’re not the right fit
- Five questions to ask before you call it done
- Sources
Two kinds of firms
Every asset manager has had this meeting in the last year. IT announces the rollout: Copilot or Claude for everyone, connected to Salesforce or HubSpot. The marketing manager who owns the CRM gets a new connector to switch on. A few people try it, draft some emails, summarise a few accounts. A box gets ticked: we’ve done AI.
Here’s what we’ve learned building data and AI platforms for Australian asset managers: that meeting splits the market in two.
Illustrative. Same tools, same start. The gap opens after “we’ve done AI”.
The first kind sees software as a cost. It takes whatever its vendors hand over, accepts the defaults, and stops there.
The second kind sees software as an edge. It uses the same tools as a starting point, then builds on them to win more business, stay audit-ready and spend smarter than the firms it competes with.
Let’s give credit where it’s due. The vendors have done real work in 2026. Claudeforce puts Salesforce data and actions inside Claude, with 37 ready-made sales skills. HubSpot’s free connector lets Claude read and update HubSpot records. Copilot for Sales works with Salesforce, and Microsoft 365 Copilot can query HubSpot through a connector still in preview. A BDM can now summarise an account or draft a follow-up in seconds.
That’s useful. It’s also exactly what every one of your competitors can switch on tomorrow.
Be honest. Does any of this sound like your firm?
- BDMs decide who to call from memory and gut feel.
- Nobody can say what client data has been pasted into an AI tool this year.
- Board pack numbers get rebuilt by hand every month, and two people still disagree on them.
- Your AI rollout is measured in licences switched on, not flows won.
- One person built the reports, and everyone quietly hopes they don’t leave.
If you ticked two or more, you’re the first kind. That’s not a criticism. It’s the default, and it’s where most firms sit. But it hides three problems, and each one has a cost.
Problem 1: you’re leaving revenue on the table
What it looks like. Your AI assistant can only reason over what it can see, and out of the box it only sees the CRM. Web and LinkedIn engagement, media mentions and history over time live somewhere else. So it can’t tell a BDM which advisers have been reading your fund updates, and it won’t volunteer anything unless someone thinks to ask.
What it costs you. Time spent on the wrong relationships. Every week your people call whoever they called last, while the advisers and clients most ready to act go quiet. That’s flows you never see, because nobody knew to ask.
How the second kind fixes it.
All your data in one place
- The Foundry pulls in what the CRM doesn’t hold, cleans it and joins it to your records on a schedule
- Outcome: the AI, and your team, see the whole story
Who to contact next
- Compass recommends which advisers or clients each BDM or relationship manager should contact next
- Outcome: time goes to the relationships most worth the call
And AI that speaks fund distribution. Sherpa is set up for each firm’s own CRM fields and checks its answers against the actual records, instead of guessing like an off-the-shelf assistant.
Problem 2: you’d struggle if audited
What it looks like. Right now, people across your business are pasting client names, holdings and flows into whichever AI tool is open: a personal account, a browser extension, a free app someone found last week. Nobody approved it. Nobody logged it. Nobody can say where that data went. Meanwhile, the AI that was approved writes a fresh answer each time, so two people can take two different numbers to the same meeting.
What it costs you. The evidence says this is already costing firms money:
For any firm holding client data, that’s not just an IT issue. It’s a client trust issue, a privacy issue and a reputation issue: the kind that costs mandates.
The audit test. If an auditor asked tomorrow which client data went into which AI tool, when, and who approved it, could you answer? The first kind of firm usually can’t.
How the second kind fixes it. AI runs inside a platform built to be checked:
- One number, every time. Compass dashboards show AUM, flows and pipeline from the same source, and each person sees only the clients, funds or territory they’re allowed to see.
- Independently audited. SOC 2 Type 1 (Trust Center), with role-based access, SSO, MFA and AES-256 encryption as standard, and audit logs of who accessed what.
- AI that stays where your data lives. Models are called through AWS in the environment we operate, not through staff’s personal accounts. Compare that with Claude Enterprise, for example, which routes requests to servers worldwide by default, and whose only residency option is US-only.
- Nothing hidden. A public Trust Center and a live status page.
Problem 3: you’re paying more without investing better
What it looks like. Licences for the CRM, licences for the assistant, add-on data platforms, a consultant to set it up, and someone in-house to keep it running. Every line is a cost. None of it is an asset that makes you better than the next firm.
What it costs you. For the same 20-user scope in year one:
| First year, 20 users | US vendor stack | Build it yourself | Datafabric |
|---|---|---|---|
| Roughly | $486,000 | More than either A capable in-house team alone runs $700k+ a year. |
$160,000 |
These figures assume 20 users and an AI assistant working across the firm’s data in its first year. Your numbers will differ with your size, data sources and usage. The full method is in our first-year cost comparison and build vs buy, and we’re happy to run it for your firm.
That gap has almost nothing to do with AI. The AI itself is cheap. The expensive part is the people and plumbing around it.
A licence is where the work starts. An AI assistant bought off the shelf is a licence, nothing more. The day it’s switched on, the real work begins: connecting it to the data the CRM doesn’t hold, cleaning that data, setting who can see what, teaching it your fund names and fields, checking its answers, training your people, and keeping it all running as models and vendors change. Done badly, you’ve paid for a tool nobody trusts. Done well, it’s a project with people and budget behind it, and most firms never planned for one. And skipping it isn’t free either: Gartner expects 60% of AI projects without AI-ready data to be abandoned through 2026.
How the second kind fixes it. It invests in the layer that compounds, and stops paying for the rest:
- Data loads someone else runs. Our pipelines retry when a source fails, check quality and record what changed. No data engineering team to hire, and no reports quietly going stale.
- No extra CRM seats. Atlas maps advisers, OpsFlow runs workflows, and Treaddesk lets clients pull their own data. Executives, operations and clients get answers with one sign-in.
- No lock-in. Sherpa can use Claude, OpenAI, Gemini or Grok, so you can switch to a better or cheaper model as the market moves.
- Costs that scale the right way. You pay for what runs and what you use, not for another hire each time you add a data source, a report or a team. The cost per outcome falls as you grow.
Side by side
| What you need | First kind: what vendors give you | Second kind: Datafabric |
|---|---|---|
| Revenue | ||
| Data from outside the CRM | Only with paid add-ons and setup | Built in |
| Tells your team who to contact next | Only if asked | Yes, a ranked list |
| Built around fund distribution | Generic sales | Yes |
| Audit | ||
| The same number every time | No, answers are written fresh | Yes, fixed dashboards |
| Each person sees only their data | Within the CRM | Across every report and tool |
| Show an auditor how AI uses client data | Rarely, if staff use their own tools | Yes: SOC 2 audited, access controls, audit logs |
| Spend | ||
| Data loads run and monitored | Not included | Yes, run by us |
| Access without a CRM seat | No | Yes, one sign-in |
| Choice of AI model | Whatever the vendor offers | Claude, OpenAI, Gemini or Grok |
| First-year cost, 20 users | ~$486,000 | ~$160,000 |
What it adds up to, for your products and your P&L
At business-outcome level
For the head of distribution
- More revenue: time goes to the relationships most worth the call
- One picture of every client and adviser, across sales and marketing
- AI answers in your firm’s language, with your firm’s numbers
For the COO
- Audit-ready: AI use you can show, and one trusted number from desk to board pack
- Smarter spend: around a third of a US vendor stack, no data team to hire, no extra CRM seats
- A data foundation that compounds, with costs that grow with use, not headcount
At product level
| Product | Solves | What it adds to your CRM and AI assistant |
|---|---|---|
| The Foundry | Revenue, spend | Brings in non-CRM data, cleans it and keeps it up to date |
| Compass | Revenue, audit | Trusted dashboards, secure access, and who-to-contact-next recommendations |
| Sherpa | Revenue, audit | An AI assistant set up for your CRM fields and for fund distribution |
| Atlas | Revenue, spend | A map of advisers for BDMs and marketing |
| OpsFlow | Spend | Workflows and tasks for operations teams |
| Treaddesk | Spend | Self-serve data for clients |
Under the hood. All of it runs on one data platform in AWS that we operate for you:
| Layer | What we use | In plain English |
|---|---|---|
| Data loads | Dagster, dbt, Python | Brings data in, cleans it and checks it on a schedule |
| Data store | Aurora PostgreSQL, S3 | One place every report and tool reads from |
| Reporting | Amazon QuickSight | Dashboards that show each person only their data |
| AI | Claude, OpenAI, Gemini, Grok via AWS Bedrock and OpenRouter | The right model for each job |
| CRM link | Our own Salesforce MCP server and HubSpot integration | Lets the AI read your CRM safely and accurately |
| Sign-in and hosting | Cognito, IAM, EKS | One login, tight access control, reliable hosting |
When we’re not the right fit
Every vendor article is written by the vendor, so here’s where our argument stops.
Strong internal team, simple data. If you already have data and engineering people with spare capacity, and your data sits cleanly inside your CRM, the default stack plus some care may be enough. We set out where that line falls in our cost comparison.
Large, multi-country transformations. Our case is strongest for small and mid-sized firms. A global group re-platforming across several jurisdictions may need a different scale of partner.
A fair question back at us. Choosing a smaller, specialist supplier carries its own concentration risk. Price it, and make it a term of the contract. Our answer is that the knowledge sits in a platform built to serve every client, not in a few people’s heads. We make that case in full in Software as a Service is broken.
Five questions to ask before you call it done
They’re short. The answers tell you which kind of firm you are.
If even one answer gives you pause, the default stack isn’t enough. Not in 2026, and not at the speed adviser flows, regulation and AI are moving.
Sources
- First-year cost of a US vendor stack vs Datafabric for a 20-user asset manager, with full method. Datafabric first-year cost comparison
- Cost of an in-house team and build risk. Datafabric, build vs buy
- Client meeting and data consolidation outcomes. Datafabric deployments, self-reported
- 60% of AI projects without AI-ready data expected to be abandoned through 2026. Gartner, February 2025
- Shadow AI breaches, added cost and AI access controls. IBM Cost of a Data Breach 2025
- Organisations unsure whether they were compromised through AI. Help Net Security, August 2025
- Claudeforce and the Salesforce in Claude plugin with 37 sales skills. Salesforce, August 2026; CX Today, August 2026
- HubSpot connector for Claude features. HubSpot
- Microsoft 365 Copilot and HubSpot connector status. Carly
- Claude Enterprise features and US-only inference. Anthropic
Figures are illustrative and rounded. Cost figures follow our published first-year cost comparison for a 20-user deployment; real costs vary with scope, data sources and usage. Client outcomes are self-reported. Vendor features are as described by each vendor or reported publicly as of November 2026.