CASE STUDY · AI-NATIVE REWRITES OF LEGACY SAAS (CRM/ERP/ITSM)
Rebuilding a legacy CRM as an agent-first platform.
- 50%Manual data entry down about 50%
- 60%Duplicate records down 60%
- 5 hoursSales reps save about 5 hours a week on admin
01 · PROJECT OVERVIEW
The client and the brief.
An anonymised mid-market B2B services company. Name withheld under confidentiality.
Agents for record updates, follow-up drafts and deal-risk flags, introduced in stages alongside the old CRM. The company kept its CRM as the system of record during the work. The agents read emails, calendar entries and activity, proposed updates, drafted follow-ups for the rep to edit and flagged deals that showed signs of stalling.
- 1A 10-year-old CRM with manual reconciliation and duplicate entry.
- 2Sales reps typed the same details into the CRM and into other tools.
- 3Records drifted apart, duplicates piled up, and managers did not trust the pipeline reports.
- 4Follow-ups were forgotten when reps were busy.
- 5A full replacement had been discussed for years and always postponed because of the risk.
02 · WHAT WENT WRONG
What went wrong, and how we fixed it.
Every engagement hits problems. These were the ones that mattered here, and what we changed so they did not come back.
- PROBLEM 01Early merge suggestions combined records for different contacts at the same company, because the matching leaned too heavily on the company domain
Two reps found wrong merges in the first week. We stopped automatic merging, changed the matching to require stronger signals, and moved all merges to approval until the error rate stayed low in review. We also added the undo window.
- PROBLEM 02The reps accepted the change because they could see the reasons for each suggestion
- PROBLEM 03A second hurdle was adoption
In the first month, several reps kept using their own notes and ignored the agent's suggestions, because the first drafts did not sound like them. We sat with the most sceptical reps, collected examples of their own follow-ups and changed the drafting instructions to match their tone and length. We also let each rep switch off drafts for a given account type.
- PROBLEM 04Use rose once reps saw that they could shape the tool and that it saved them time on the dull parts of the job
03 · THE PROCESS
How we ran it.
- 01Weeks 1 to 2
Watched how reps and managers used the CRM. Measured time on admin, counted duplicates and agreed baselines.
- 02Weeks 3 to 5
Built the record-update agent in suggestion mode only. Reps saw proposed changes and accepted or rejected them.
- 03Weeks 6 to 9
Added follow-up drafts and the duplicate matching step. Switched on automatic updates for low-risk fields.
- 04Weeks 10 to 12
Added deal-risk flags, reviewed them with sales managers and trained the team.
- 05After launch
Weekly review of rejected suggestions to improve the agents.
04 · RESULTS
What changed.
Measured against the way the work ran before the engagement began.
- NOW 01
Reps review and approve agent suggestions instead of typing records.
- NOW 02
Managers rely on pipeline reports because the data is more consistent.
- NOW 03
The team has a standing review of rejected suggestions, which acts as a feedback loop.
- NOW 04
The company has also stopped treating a CRM replacement as a single risky event.
- 50%
Manual data entry down about 50%
- 60%
Duplicate records down 60%
- 5 hours
Sales reps save about 5 hours a week on admin
05 · WHAT WE BUILT
How it works.
- An integration layer read from the CRM's interface and from email and calendar sources.
- A matching service compared records using names, domains and contact details, and proposed merges with a score.
- Agents produced structured proposals, each with the evidence it used.
- A policy layer decided whether a change could apply automatically, needed approval, or was not allowed.
- Every proposal and decision was stored with the source data, the model version and the person who approved it.
- Permissions followed the CRM's existing roles.
- Controls after launch
- A weekly report shows how many suggestions were accepted, edited or rejected, by agent and by rep. A rising rejection rate prompts a review. Automatic updates are limited to a named list of low-risk fields, and any change to that list needs sign-off from the sales operations lead. The audit log records the source of every change, so a manager can see why a field has its current value.
- Handover
- Sales operations learned to adjust matching rules, edit the draft instructions and review the audit log. IT received the architecture notes and a list of the credentials and interfaces the agents use. The next stage, retiring parts of the old CRM, is now planned with the company's own team.
06 · QUICK ANSWERS
The short version.
- What problem did the client have?
- A 10-year-old CRM with manual reconciliation and duplicate entry. Sales reps typed the same details into the CRM and into other tools. Records drifted apart, duplicates piled up, and managers did not trust the pipeline reports. Follow-ups were forgotten when reps were busy. A full replacement had been discussed for years and always postponed because of the risk.
- What did Brandsmashers build?
- Agents for record updates, follow-up drafts and deal-risk flags, introduced in stages alongside the old CRM. The company kept its CRM as the system of record during the work. The agents read emails, calendar entries and activity, proposed updates, drafted follow-ups for the rep to edit and flagged deals that showed signs of stalling.
- What were the results?
- Manual data entry down about 50%; duplicate records down 60%; sales reps save about 5 hours a week on admin.
YOUR TURN
Have a similar requirement?
Tell us what is stuck. We will come back with a plan and matched, pre-vetted engineers.
