Home · Business · Sales · Pipeline & forecasting

PARTLY

As of 13 August 2026, AI can only partly predict which deals will close.

This still needs a person who signs their name to it.

Can you do it?

15 minutesto a draft.

1 hourto something you’d act on.

Cost, all in£0

Skill neededpower-user

Who has to check ita colleague

What the alternative costsA sales operations review of the same CRM export is the alternative; no price is stated here.

If this goes wrong: you give attention to a weak deal, neglect a late-stage opportunity and build a forecast on signals that have already changed.

What to actually do

  1. Hand it to a person

    The route this page recommends

    A person who owns the outcome does this end to end, worth it when the failure is dear.

  2. Use a tool built for this

    Second choice

  3. Do it yourself

    The distant third

    A chat interface, power-user skill, and roughly 1 hour until you can act on the result.

    How to actually do it

    1. Export the open pipeline from your CRM with deal ID, value, stage, owner, creation date, last activity, next step, expected close date and forecast category.
    2. Gather the latest anonymised meeting notes, call summaries, emails and buyer commitments for each open deal, and remove personal data that the model does not need.
    3. Open a chatbot and paste the prompt followed by the export and supporting notes, keeping the same deal ID across every source.
    4. Ask the model to rank the deals by likelihood of closing by your chosen target date and to separate facts, inferences, missing information and stale signals.
    5. Compare each ranking with the current CRM stage, forecast category, close date and the salesperson's own view, then ask the model to explain every disagreement without changing the CRM.
    6. Use a historical export of previously closed and lost deals to test whether the signals it cites were associated with outcomes in your own business, and record where the prediction was wrong.
    7. Have the sales manager decide which deals receive attention and update the CRM only after checking the latest buyer communication.

    Prompt

    You are helping a UK sales manager assess an existing pipeline. Using only the anonymised CRM export, activity history, current deal notes and call summaries below, rank the open deals by likelihood of closing by [target date]. Do not invent missing facts or treat an empty field as a negative signal. For each deal, provide: deal name or ID, current stage, likelihood band of high, medium or low rather than a false-precision percentage, evidence supporting the band, evidence against it, missing information that could change the assessment, and the next question a salesperson should ask. Separate observed facts from inferences. Flag stale activity, contradictory notes, unrealistic close dates and deals whose data is too thin to assess. Do not change the CRM stage, forecast category or close date. Finish with a short explanation of the assumptions and a list of the five data fields that would most improve this assessment. Data: [paste anonymised CRM export, activity history, deal notes and call summaries here].

    Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

What it gets wrong

  • It cannot know whether a buyer is privately losing interest when that change is absent from your CRM and call records.
  • It turns recorded signals into a ranking, but it cannot replace a salesperson's judgement about politics, trust and competing priorities inside the account.
  • It cannot prove that a pattern from past deals will hold when your market, pricing, product or sales process changes.
  • It can present a confident ranking from incomplete or stale pipeline data, so the forecast still needs a human owner and a historical accuracy check.

What caps this at PARTLY: real time truth, judgement under ambiguity and verification cost.

How we scored this

Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.

AxisScore (0–2)
Output2
Inputs1
Verification1
Liability1
Effort delta2
Total7 / 10

The methodology and its thresholds are published in full.

FAQ

Can AI accurately predict which sales deals will close?
It can rank deals using recorded pipeline activity, stage history and call evidence, but it cannot see unrecorded changes in buyer intent. Treat the result as a prioritisation aid, not as a reliable replacement for a sales forecast.
What data does AI need to predict deal closure?
Give it a consistent CRM export with stages, activity dates, next steps, close dates and forecast categories, plus anonymised call summaries and buyer commitments. Missing or stale fields make the ranking weaker and can make inactivity look like a meaningful signal.
Can ChatGPT predict which of my leads will convert?
It can analyse a pasted, anonymised export and produce a ranked list with reasons and missing information. It cannot access your live CRM unless you connect an appropriate tool, and you still need to check the ranking against current conversations and past outcomes.
Should I trust an AI sales forecast?
Use it as one input alongside your CRM, salesperson judgement and historical results. Do not let it automatically downgrade or abandon a deal without checking the latest buyer evidence and the consequences for your forecast.

Nearby answers

All Pipeline & forecasting rulings

Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.

The newsletter

AI news, new answers and product picks, straight to your inbox.