The selling day is mostly not selling
Ask a business development team where their week goes and the answers are consistent: preparing a quotation, updating the customer record, chasing an internal approval, writing notes that nobody reads, rebuilding the history of an account before a call, and handing an order to operations in a form somebody else has to interpret.
None of this is selling and all of it is a precondition for selling. The commercial consequence is easy to see and difficult to attribute: fewer calls made, calls made later than the customer expected, and responses to a tender that arrive after the deadline. The revenue organisation's own reporting usually shows a pipeline number rather than a measure of this loss, which is why it survives.
CRM data quality is a pipeline visibility problem, not a tidiness problem
The usual response to poor CRM data is a data cleansing exercise. Cleansing improves the record and then the record decays, because the decay is caused by the work of selling rather than by the data itself.
Ask what makes a record dirty. Most entries are made after the call, when the detail has to be reconstructed, under pressure, by somebody for whom the CRM is not the priority. Duplicate accounts are created because nobody could find the right one. Opportunities sit at the same stage for months because moving them requires a decision nobody makes. The remedy is to remove the decisions, not to police the records.
Where the administrative time actually goes
- Quotation preparation. Reconstructing price, terms and scope from previous deals, then having it reviewed, then reformatting it.
- Data entry after the fact. Notes, stages, contacts and next steps recorded once the interaction has already ended.
- Internal routing. Chasing an approval or a legal review with no visibility of where it currently sits.
- Account preparation. Rebuilding the history of an account before a meeting, because the history is not usable.
- Order handover. Passing a commitment across a boundary in prose and having it re-entered on the other side.
Two characteristics run through all five. Each requires information that already exists somewhere, and each is done by a person whose time is expensive and whose attention is finite. That combination is what makes them good candidates for support and bad candidates for headcount argument.
Quote preparation is a retrieval problem disguised as a writing problem
When a representative asks how long a quote takes, the honest answer is usually that most of it is spent working out what to charge. What the customer will accept, what was agreed last time, which discount was approved by whom, and whether this one is an exception.
That knowledge exists. It sits in past quotes, in approval threads, in a pricing table nobody has opened in a year, and in the memory of two people. The useful intervention is retrieval with provenance — the comparable precedent, the approved discount band, the clause the customer negotiated last time, each attached to where it came from — rather than a generated price. A representative who is handed a number with no source will not send it, and will be right not to.
Follow-up depends on a memory the organisation does not hold
Most revenue is lost to a specific, ordinary event: a stated intention to follow up that lives only in the representative's head, is not recorded as a dated commitment, and is overtaken by the next call. Pipeline reviews then report a stage rather than an expected action, and a forecast built on stages cannot distinguish a deal waiting on the customer from a deal nobody has contacted for three weeks.
This is a capture problem with a narrow specification. A commitment has an owner, a date, an action and a trigger, and if it is written down at the moment it is made, follow-up becomes a queue rather than a personality. The conversation is unstructured and the business meaning is buried in it, which is why this is one of the clearer cases for reading what was said and turning it into a dated record for a person to confirm.
The order handoff is where a commitment quietly changes shape
A customer agrees to something in a call. It is written up as an order by somebody else, who reads the note, applies reasonable assumptions and sends it. Operations discovers the assumption on the first day, the customer discovers it later, and the deal that was won becomes a credit note conversation.
The handoff is where pipeline data stops describing reality, because this is the point at which an interpreted document replaces an agreed commitment. Making that exchange explicit — what was agreed, on what terms, with what excluded, acknowledged by the customer — costs a paragraph and removes an entire category of dispute. The structured record of the commitment is the deliverable, and it is closer to a contracting problem than to a reporting problem.
What should not be delegated
Assembling the pricing evidence, drafting a first version, filling the fields and flagging the missing ones are delegable. What is not: what to concede, whether this customer gets an exception, whether a deal should be forecast, and when a commitment should be walked back. Those are judgements about the account, made with knowledge the system does not have, and a system that appears to make them removes the accountability without improving the decision. The comparison of approaches is in when to use AI, rules or conventional automation.
What to start with on Monday
- Have three representatives log their time for one week against predefined activity categories. The distribution is the business case.
- Select one administrative activity with high volume and low consequence — quote assembly, record updating, account preparation — and count how long it actually takes today.
- Identify the information each instance needs and where that information already exists.
- Specify the missing business rules: discount bands, approval thresholds, required fields, and what constitutes a valid commitment.
- Build the support for that one activity, with the source of every extracted figure visible so the representative can check it.
- Record the baseline before launch and review at ninety days against selling time, not against pipeline alone.
The measures that matter are selling hours recovered, the interval between customer request and first substantive response, the age of the oldest open commitment, and the share of records that would pass a forecast review without someone asking a question. The capacity argument deserves the same discipline as any other, and the method for establishing it is in calculating the value of enterprise AI.